Foundation models: accepted in review but not delivered (212)

简体中文

These entries passed an earlier review and lack a current catalog entry or an explicit cost closure. This is not user authorization to implement every item, nor does it mean each item passed the current cost gate. Do not use this as an automatic implementation queue. Sorted by paper ID descending.

Confirmed decisions not to reproduce full experiments are marked separately. Other items first need cost and scope reduction, not automatic execution.

ID / paper	Current disposition	Priority organization verification	Prior review rationale
2610.19139 Online Tokenization with Optimal Delay in Linear Time	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Online tokenization formalizes stable output prefixes and delay, combining special tokens, pretokenization, and BPE. It gives a lower-delay joint algorithm and impossibility boundaries. The public Morgan Stanley implementation allows CPU-level exact token-ID and streaming equivalence checks. Accepted as a low-cost candidate; not implemented here and no full throughput study.
2610.08791 World Models' Last Exam in Physics	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Forty physics-relation measurement tasks across nine phenomena; official images, prompts, evaluators and external video batch scoring support consideration as public evolve infrastructure, not a new world-model algorithm or completed integration.
2610.08779 ALIVE: Interaction-Aligned Object Insertion for First-Frame-Guided Video Editing	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ALIVE concatenates source and noisy target channels, uses a clean first frame and normalizes cross-attention by temporal overlap and description length. Qwen VLM predicts P4 from the same inputs rather than reading targets. Training uses 35,800 pairs and two benchmarks; some baselines get extra P2 prompts, so conditions are not fully matched.
2610.08718 When Forgetting is not Catastrophic: On the Mechanics of Spurious Forgetting	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Fine-tuning new facts separates shared representation drift from per-fact erosion. Removing the first singular component of the update has a same-norm random rank-1 control on OLMo2-1B; suitable for mechanism diagnosis, not a universal anti-forgetting algorithm.
2610.08713 SpaTime: Streaming Vision-Language Models for Spatio-temporal Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Freeze causal geometry and vision encoders, add projected 2×2 merges, train LoRA/state/head, and model first-Response timing. Strict counts earliest answers; charitable last-frame forced answers stay separate. Train on StreamVSTI and evaluate StreamVSI OOD tasks.
2610.08678 Secure Speculative Decoding for Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Mix strict/relaxed acceptance and rejection recovery distributions by global generation position, validating strictly early and quickly later. Defined sufficiently for genuine speculative decoding.
2610.08674 EC-RAG: Event Chain Retrieval-Augmented Generation for Long Video Understanding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Decompose query modalities without training; sample frames by CLIP variance; build ASR/OCR/DET event descriptions and time-adjacent chains; use Contriever for related events and adjacent context. Three long-video benchmarks, no learned causal event graph. Count tools and tokens.
2610.08659 Selective Transfer of RL Updates for Visual Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Separate R−B from B−V, use rank-1 SVD per matrix then restore original Frobenius norms, and reconstruct donor interpolation or directly add the RL delta. Three 8B VLM backbones; select donor on source validation and λ on H1 validation, freeze both, and preserve nonvisual module support alignment.
2610.08649 Stable Scores, Unstable Answers: Frame Phase and Option Order in Video Multiple-Choice Evaluation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Map 3×8 frame-phase posteriors to content space and arithmetically average, independent of option-order control. With n=442, the 0.45pp gap versus 32 frames is not significant. Origin flips fall but order flips persist. Serial 3.18s exceeds 24-frame 2.86s; attention units are not measured speedups.
2610.08639 Forensic Reserve: Eliciting Latent Knowledge for Image Forgery Detection	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Locate a reserve via site-wise ICA of frozen DINOv3 and real/fake discovery. Fix output basis U and train only zero-initialized A in residual UA h; freeze classifier/backbone and exclude benchmarks from discovery. Reusable visual adapter, with task-specific forensic evidence.
2610.08560 Have I Seen Enough? Frozen Video-Language Models Encode Evidence Readiness	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Nested-CV logistic probes on frozen Qwen residuals predict evidence readiness. Stop on joint readiness/confidence at training quantiles, or choose highest readiness by deadline. Video holdout and crossed quads remove independent image/question shortcuts. Fixed-window policy does not read gold evidence times; annotated growing-prefix sampling is diagnostic, not deployment latency. §4.4 gives matched-video-seconds improvements.
2610.08463 UNREAL: Unifying Retrieval and Long-Context with a Single Model	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Use frozen decoder middle-layer chunk features, learn 64 retrieval tokens and cross-layer mixing for MaxSim, then re-encode selected text to generate. Includes multi-positive InfoNCE, public QA training split, and independent long-context validation.
2610.08414 Image Bitstream Fine-grained Understanding for Privacy-Friendly AIoT	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. BFG encodes JPEG bytes to 192 dimensions using ByteFormer hierarchical convolution/shifted windows, projects to 512, fuses global context with local cross-attention for autoregressive captions, and adds a 120-class head. CFU-D isolates source IDs/splits and tests substitution/deletion/repetition corruption. Dogs 58.51% uses ImageNet initialization; skipping decoding is not a privacy guarantee or broad foundation result.
2610.08403 SSR: Sparse Segment Reduction for Ternary GEMM Acceleration	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SSR separates ternary matrices into positive and negative binary matrices, sorts column blocks by pattern, drops all-zero patterns, and replaces multiplication with segmented sums and tree reduction. Public C++/AVX2 code is verifiable.
2610.08401 GeoPID: Decomposing and Steering Visual Information in Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GeoPID estimates shared and modality-unique subspaces from unlabeled calibration features and amplifies vision-unique directions by I+αUUᵀ. Across 22×14 combinations, 100 separate labeled calibration questions per group select layer and α before freezing; final evaluation sees no gold. Rényi task decomposition is offline diagnosis, and the mean relative 7.63% must retain heterogeneity and regressions across 308 combinations.
2610.08341 DIPrune: Task-Aware Token Pruning with Dual Importance for Efficient Multimodal Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DIPrune combines inverted text-to-vision attention rank normalization with truncated normalization of cross-layer rank gains, preserving tokens initially low-ranked but semantically important later and removing corresponding KV. Multiple VLM/video backbones and aligned-layer ablations. At 64 LLaVA tokens, retains 97.7% on average; POPE TTFT 73.23→32.50 ms, with 0.24 ms extra. Use measured TTFT, not FLOPs, for speed.
2610.08183 Compact Robot Policies Need Fine-Grained Visual Representations	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Train DINOv2 with a 48-query resampler/FiLM text and small flow DiT for actions. λ=.01 weights cross-layer task-classification CE, not an information bottleneck. Remove classifier after training and output action chunks with 10 Euler steps. LIBERO/RoboTwin are within task distributions; real hardware uses an additional private 10,000-hour/5,887-task pretraining corpus, not public simulation alone.
2610.08164 Align, Then Correct: Training-Free Two-Stage Low-Rank Compensation for Extremely Quantized Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Two-stage low-rank compensation: asymmetric activation targets use full-output Fisher-weighted SVD, then rank-constrained SVD of the existing adapter plus K-FAC natural gradient. Stage two replaces rather than adds a rank-r adapter.
2610.08162 The Failure Is in the Readout: Fine-Grained Emotion Recognition Benchmarks Measure Elicitation, Not Perception	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. EmoNet audits first-token yes/no probabilities for 40 classes against free JSON generation, with answer mass and single-answer validity gates. Candidate multimodal evaluation diagnostic.
2610.08159 Making COMET Comparable Across Scripts: Diagnosis and Correction of Tokeniser-Induced Script Bias in Indic MT Evaluation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Public COMET script-bias diagnostics and rank-safe quantile calibration/LOLO correction are candidate evaluation-audit infrastructure, not a new LLM or integrated dynamic scorer.
2610.08133 VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Freeze the VLA and train only a five-layer patch scorer. Laplace soft Top-K has total mass K, with a square-root black-image embedding gate, L1 full-context teacher and ground-truth action consistency. Hard Top-K preserves RoPE positions at inference. LIBERO and real hardware/A100: K64 success 97.0 versus 96.8, 63.49→46.82 ms; retain K32 Long regression.
2610.07990 A Broader Look at Model Merging: Rethinking Implicit Regularization Induced by Task Arithmetic	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. S-RandOpt adds one auxiliary-gradient dimension to the task-vector subspace, then samples models; compares actual weight GD with coefficient GD. Candidate merging search mechanism.
2610.07987 VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. VisionWeave learns a gated 2×2 pooler and intermediate ViT granularity router with fine/coarse MRoPE centers aligned. Three teacher top-512 forward-KL stages train pooler, soft-mix router, then hard-route pooler plus LLM; not OPD. A training subset calibrates 50% savings; eight benchmarks and SGLang dual-A100 long video show 2.30× throughput. Label the modified post-encoder FastV baseline.
2610.07940 Hybrid Latent Attention for Looped Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. HLA shares rotary latent KV across loops while keeping an exact recent window. A frozen backbone with learned writer/reader and real same-engine vLLM/A100 speed/quality comparisons supports systems consideration.
2610.07925 TF-PRVR: Training-Free Partially Relevant Video Retrieval	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TF-PRVR freezes the VLM, uses db4 DWT on frame-feature velocity direction changes for hierarchy, builds within-level semantic/temporal and adjacent-level IoU edges, then propagates by (1−β)(I−βW)^−1 and takes max after per-time cross-scale averaging. Three datasets/backbones, fixed β=.7. Count offline matrix inverse and features; no new VLM training.
2610.07903 Unsupervised Long-Tailed Adaptation of Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MARS learns prompts/text adapters on frozen CLIP: zero-shot visual prototypes detect unsupported probability shifts and fix weak targets with strong-view CE. Teacher soft frequencies and margin EMA control KL/pseudolabel CE. Nine datasets, γ=10/20/50, three seeds, unknown priors but known label set; not open-category discovery.
2610.07887 Visual Abstention in Unified Multimodal Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. VisTA on BAGEL trains feasible/infeasible pairs: reason then generate image or ABSTAIN; text CE for both, flow matching only for feasible inputs. Separate 38,224 pairs and 1,050 DoD image/maze pairs. Without reminder: 93.0% refusal, 0.8% false refusal, 74.3 edit. Paired minimal-change criteria do not establish real-world impossibility or safety alignment.
2610.07868 CueRator: Agentic Search for Symbolic Rules to Adapt Frozen Multimodal Encoders	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CueRator uses agent discovery of AST-constrained closed-form gates on frozen ImageBind. Per-video Sobol training oracles are structure upper-bound diagnostics; a Gaussian contextual policy predicts ten parameters with K4 leave-one-out REINFORCE. Select among 3×20 candidates on validation then test. No deployment access to oracle GT; encoder is frozen, training uses seen-only data, validation includes unseen.
2610.07848 Dynamic Positional Attention Modulation for Parameter-Efficient Fine-Tuning of Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DyPAM uses low-rank hidden features for RoPE-paired Q/K scaling with head/layer biases. Core PEFT formulas, validation checkpoint selection, and multiseed public tasks are present.
2610.07819 \(α\)Transfer: Coefficient Transfer for Efficient Model Merging	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. alphaTransfer moves global/task/layer merge coefficients searched on a small proxy to a larger same-family model, copying/interpolating by relative depth. Actual task-vector edits run on the native larger model.
2610.07809 MASKerade: Token-Routed Mask Experts for Dense-to-MoE Upcycling	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MASKerade trains expert binary masks and a token router on frozen FFNs, with true hard masks, STE, and top-1 retaining softmax probability for router gradients. Matched nominal active budget and mask-type comparisons support multimodal MoE adaptation.
2610.07774 Reading, Not Manipulating: Leveraging Router Logits for Multimodal Safety in MoE Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Concatenate last-prompt-token router logits across MoE layers; train-standardize and fit class-balanced L2 logistic detection. After prefill reject at validation-selected τ=.4 or generate normally. Two MoE models, HoliSafe/OOD and MMMU. Readable safety signals do not imply steering control or router uniqueness; residual features also work.
2610.07758 Later Is Better: Token Reduction for ViTs Under Distribution Shift	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Late-concentrated power-law token allocation is an executable ViT/VLM compression operator with public official code. Multiple shifts/backbones and five reducers benefit under conservative iso-GFLOPs.
2610.07729 Foveated Compression: Selective High-Resolution Preservation for Token-Efficient VLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Foveated Compression defines mixed-resolution representations and tie-aware region selection, suitable for token-compression diagnostics; do not turn test-oracle or worse-than-resize results into gains.
2610.07706 WASD: Wasserstein-based Knowledge Distillation for Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. WASD uses debiased Sinkhorn divergence with teacher-embedding costs and detached potential differences to optimize student probabilities. Official train/eval scripts and equal-wall-time comparisons exist.
2610.07689 Unlocking Fine-Grained Perception in CLIP via Structurally-Aware Latent Masked Modeling	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SALM adapts CLIP with patch relations and norm-difference matrices plus masked reconstruction of teacher latents; SALM-Self removes the external teacher. Core losses and curriculum are specified.
2610.07643 Monte Carlo Estimation for KV Cache Eviction	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Sample short continuations from a frozen model to estimate future queries, weight them by centered log-probability, entropy, and disagreement, then select KV by counterfactual deletion cost after output projection. Discard sampled continuations.
2610.07585 REViT-v2: Hierarchical Windowed Roto-reflection Equivariant ViT for Equivariant Feature Extraction	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. REViT-v2 uses group-equivariant QKV convolution, full representation-field multihead local windows, and group-convolution downsampling, with O(NM²C) rather than global N². ImageNet and RotMNIST include a same-stem global control. Preserve finite p4m group and window geometry; no arbitrary-angle exact equivariance claim.
2610.07572 Two Vectors Replace In-Context Demos: Structured Task Adaptation via Embeddings	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. STAVE adds two vectors to structured frozen LM/LMM inputs with paired demo/no-demo supervision, genuine 2d adaptation rather than gold inference or per-layer simulation.
2610.07533 SkillFormer: Skill-Decomposed Adaptation for Audio Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SkillFormer clusters question embeddings into six skills, uses parallel rank-16 adapters and top-2 question routing, trains 500 steps per cluster then 2,000 balanced joint steps. 50k QA, three audio models/benchmarks. §2.4 Phase B explicitly unfreezes the LLM despite Phase A 'throughout' wording; reproduce phases and report trainable parameters, not LoRA-only throughout.
2610.07532 Safeguarding LLMs via Model-Agnostic Latent Safety Signals from Dark Knowledge	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. LADE selects safety tokens by harmful/benign first-token probability differences, maps first valid subtokens across tokenizers with duplicate weights, and gates using kNN distances from L1-normalized features to harmful references. Implementable defense mechanism.
2610.07522 Activation Denoising: A Robustness View on Parallel vs Sequential LLM Quantization	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Estimate H/C/N second moments with clean/noisy forwards, linearly pretransform weights, then apply GPTQ/QuIP# rounding under the new metric so layers can quantize in parallel.
2610.07457 AlignQuant: Tile-Aligned Mixed-Precision Quantization for Efficient LLM Generation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. AlignQuant aligns assignment, packing, and GPU execution to 64×64 tiles. Normalize prefill/decode Fisher sensitivities separately, take their max, use 4 bits for low-sensitivity tiles and 8 otherwise. Real integer execution is essential.
2610.07389 Inference and learning in sparse autoencoders as natural gradient flow	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. BeFOND casts Bernoulli sparse inference and dictionary learning as natural-gradient flow, with closed-form updates, executable official core, and public SAEBench tasks. Candidate representation diagnostic/intervention.
2610.07385 Semantic Capability Acquisition and Specialization During Vision-Language Model Fine-Tuning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SSR isolates NAME and ATTRIBUTE routes during causal training, fuses summaries, and optionally drops NAME with probability .5; evaluation restores ordinary EOT. P3 without names and CAPP counterfactuals test description readability. Six CLIP-like models across datasets; P3 trajectory peaks are retrospective, not selection, which should use P1. Missing fork revision and subset generator limit exact reproduction.
2610.07348 Stepped MoE: Segment-Level Routing with Configurable Inference Complexity	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Use the first 12 dense layers to predict expert sets for the remaining 44, refresh on each segment's first token, and train budget embeddings with loss scaling across sparsity levels. Real 12B, 1T-token training supports variable-compute consideration.
2610.07332 Structuring MoE Expert Selection for Agentic Reinforcement Learning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Agentic RL combines operation-label/router mutual information, adjacent-token expert-set consistency, and an entropy gate. Actual AppWorld/AutomationBench training; cannot replace model routing with a proxy.
2610.07324 Scale-Invariant Training for Time Series Foundation Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ScaleIn evaluates loss in context-normalized target space as a candidate time-series foundation training configuration. It audits an existing practice theoretically and empirically, not a new network.
2610.07132 CroissantMiner: Automated Extraction and Validation of Croissant Metadata for ML Datasets	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CroissantMiner provides 102 human-gold and 500 silver papers with 30 fields, single-pass and four agentic extractors, rule/LLM field scoring, and NULL penalties. Prefill Sonnet4.5 is not ranked; use final κ=.708/hard set .663 instead of early .890. Separate external enrichment from paper-only gold. Official MIT data/evaluation/CLI support evolve extraction evaluation.
2610.07121 SoloQ: Calibration-Free Quantization for Diffusion Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SoloQ K-RPBH combines permuted block Hadamard and cross-block Haar, E8P weights, and Lloyd/NVFP4 activations without calibration. The scale \|\|z\|\|²/<q(z),z> restores a projection, not exact values. Block diffusion quantizes only committed KV and keeps active KV precise. Five dLLM experiments; separate H200 quality/low-bit performance from A100 compatibility.
2610.06963 WavePrune: One period is often enough for RoPE	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. WavePrune limits each 2D RoPE channel's QK contribution to its first rotation period. FlashWavePrune uses the maximum window per 16-dimensional group for TensorCore sparsity, usable in inference or retraining.
2610.06950 Learning to Decide, Not to Reason: Parameter-Efficient Decision Operators via Low-Rank Activation Steering	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DecSteer trains shared low-rank activation residuals and token gates, initially identity with output-branch gradients. Focuses on decisions rather than full model training. Real SearchQA/LiveMath BC and multiple benchmarks; preserve RMS, SiLU, and gating.
2610.06932 RADC: Risk-Aware Dual Caching for Vision-Language Test-Time Adaptation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. RADC's foreground/global dual cache and Gaussian-risk admission define a VLM TTA candidate.
2610.06844 Learning to Read the Contextual Tokens in Diffusion Transformers	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Contextual Reader adds layer/timestep embeddings to MM-DiT context tokens, uses a two-layer Q-Former, and probes frozen LLM QA. CoAl uses a one-query early aligner with negative cosine to frozen clean-image SigLIP during flow training, upweighting null conditions to limit prompt shortcuts. Discard aligner at inference. Separate COCO from-scratch and SD3 Fine-T2I tuning; probing is not generation capability.
2610.06817 Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Paradee distills Kokoro's single voice into 8.07M TTS, training text and decoder in stages with multiresolution spectral loss, GAN refinement, and phase locking. WikiText103 teacher audio and 200 evaluation sentences can be reconstructed.
2610.06751 MatrixFormer: A Foundation Model for Matrix Completion	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. A 78M general matrix completion model with 2D cells, row/column axial attention, and synthetic-prior training. A tabular foundation candidate, not industrial A/B approval.
2610.06750 Balancing Memory Pathways: Analyzing and Improving Memory Utilization in Hybrid LMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Keep two hybrid-LM memory paths and add recurrent auxiliary SFT with cross-segment attention masked. Public implementation supports executable training integration.
2610.06686 OVAL: Output-Aware Local Page Bases for KV Cache Retrieval	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. OVAL normalizes and mixes key energy and key/value output sensitivity matrices, builds page representations from principal subspaces, and selects decode pages via logsumexp of low-rank reconstructed logits without rereading value statistics.
2610.06666 What Matters for Latent Reasoning with Flow Matching	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FLaRe combines symbolic-CoT VAE, noise-shifted flow matching, and backpropagation through verified rollout paths. An executable latent reasoning training mechanism, not only measurement.
2610.06636 Differentially Private Mixing of Public Datasets Improves Private Learning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DP-MixMin convex proxy optimization of public pretraining mixture weights with privacy accounting is a foundation LM data mixture candidate, not only medicine. Full GPT2-Medium pretraining and DP tuning on two text domains.
2610.06617 RealtimeWAM: One-Step Asynchronous World Action Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. RealtimeWAM anchors consistency distillation with interval velocity to the teacher's ten-step endpoint; λ=.2 equals 1/t² endpoint weighting. Frozen video shares KV and only action LoRA trains. CUDA two-stream wavefront records events after layer KV projection and waits at action attention. LIBERO/RoboTwin/Plus and matched distillation; latency includes VAE but not once-per-episode text.
2610.06616 Video Encoders Built on Image Representations	Full experiment not reproduced; review DONE; affordable components considered separately	Google/Meta confirmed; see priority list	Accepted in review; core implementation and public verification pending. Image-first video preserves source positions, selects DPP/cross-frame associations under fixed token budget, and optionally uses an 18M temporal refiner. Appendix C.1 separates training-free controls.
2610.06611 Lens3D: Target-Conditioned Visual Foveation for Fine-Grained 3D Understanding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Lens3D grounds object boxes, selects depth-visible voxel coverage across views with wireframes, then uses a 2D VLM. LensDistill uses privileged teacher captions and target center to supervise a native 3D LLM, without extra views/VLM at test. LensBench 2,068 objects/141 val scenes separate from 561 training scenes, but references are three-model silver with 100 human-evaluated objects. Keep GT-box variants separate from end-to-end grounding.
2610.06598 SimForcing: Distilling Simulation Motion Priors into Real-Domain Robot World Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SimForcing trains action-conditioned video dynamics with simulation replays of real trajectories and a disturbed teacher. Distill adjacent clean-latent differences under shared noise/time using L2; zero-init convolution conditions each block with per-example shared dropout and noise. Bridge/InternData and downstream LIBERO. World pretraining predicts no actions; simulation deliberately excludes object interaction, not general physics.
2610.06479 Behavior-Preserving KV Cache Compression	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Match the next-token distribution under full cache. Approximate removal perturbations at logits through the final norm Jacobian and LM head, then greedily optimize mask-level marginal KL over singleton proposals.
2610.06413 SpatialChain: A Benchmark for Auditing Spatial Reasoning Faithfulness in VLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SpatialChain generates and answer-filters 28,350 train and 899 test CoTs from GQA scene graphs. Evaluated models receive image/question; judges can read GT graphs for AC/RF/RC. Surface overlap is not faithfulness and RF is graph agreement, not model-causal faithfulness. Official data/eval/judge CLI supports public evaluation with scorer gold isolated.
2610.06404 Quantifying the Stability of Multi-Step Reasoning via Error Amplification	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Random ordered CoT step subsampling with STE quantization-aware SFT aims to reduce stepwise error amplification. CLRS-Text/LEGO experiments and Jacobian diagnostics.
2610.06324 Readout Blindness: VLM Scores Miss the Spatial Direction Their Frozen Encoders Retain	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ADR uses frozen patch and independent text features for Sinkhorn excess coupling or dense centroids, then signed displacement along parsed subject-object relation axes. No training; context-free/unary blindness theorem has assumptions, not a claim about all VLM encoders. Nine models and five direction-balanced datasets; derangement null adjusts grounded gain for language priors. Candidate readout and diagnostic.
2610.06286 DeferKV: Rethinking Eviction Timing for One-Shot KV Cache Compression	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DeferKV waits until the first real generation token yields a decode query before compressing prompt KV. Combine that query with trailing prompt queries using temporal discount and reuse base smoothing, budgets, and selection.
2610.06226 LeAVJEPA: A Minimalist Architecture for Audio-Visual Self-Supervised Learning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. LeAVJEPA trains early-fused shared ViT, modality-specific local views, symmetric invariance, and SIGReg. Appendix C.3 Table A3 controls encoder, 30 epochs, and batch 512 for local views. Research candidate only, not SOTA or reproduced.
2610.06161 Introducing Code-Switched Contexts to Cognitively-Inspired Bilingual Model Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Dictionary/POS-controlled code-switching and four-stage density curriculum are executable data augmentation diagnostics for small bilingual pretraining. Preserve English–Chinese regressions.
2610.06147 Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TTC stability correction for candidate memory subspaces; public code/features support gradient-free VLM adaptation.
2610.06026 Differentiable Bit-Widths: Co-optimizing Pruning and Quantization via SVD for Ultra-Efficient LLM Compression	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DBW learns per-SVD-component bit widths including zero, with STE, Hamilton leftover allocation, matrix budgets, and singular-value entropy regularization for joint pruning/quantization. Public train/eval code supports compression consideration.
2610.06021 Scalable Minimal-Change Learning for Controllable Image Editing	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Arro audits missing and unintended edits with a VLM, deduplicates shared rubrics with an LLM then validates Booleans with the VLM, and separately normalizes negative counts before weighting and batch renormalization for flow-GDPO. MinEval 10k/600 disjoint and MagicBrush/AnyBench/EmuEdit. Model judges are not A/B; account for group auditor and external judge budgets.
2610.05994 How (and How Not) to Use Data Augmentation in VLA Post-Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Critic-only image augmentation in VLA PPO affects clean actor/rollouts through advantages. A reproducible robot/post-training intervention, not industrial recommendation evidence.
2610.05993 From Transformation to Target State: Rethinking Query Representation for Zero-Shot Composed Image Retrieval	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Frozen MLLM reconstructs several holistic target descriptions and weighted atomic facts from reference plus edit. CLIP global/patch-softmax matches are query-wise z-scored and fused. FashionIQ val/CIRR test1/CIRCO test; no task training but dataset-specific prompts and extra MLLM cost remain.
2610.05978 Byte Language Models: Scaling, Emergent Abstractions, and Information Allocation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Byte model uses 256 vocabulary, hashed n-grams/DeltaNet convolution 16, four-byte TST first 30%, and NTP last 70%. IsoFLOPs with subwords is not equal text exposure. Late-layer pooled-span interventions are diagnostic; keep the 200M segmented model. Speculation is an eval-text oracle simulation, not real latency; 3.4× accepted tokens is not 3.4× speed.
2610.05940 ReMem: Streaming Video Understanding With Long Context Retention	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ReMem's training-free SCM uses fixed general-query attention for recursive compression, retaining original RoPE positions plus temporal offset. RVM retrieves top clips by cosine of pooled keys and query, with finite FIFO eviction. Streaming/OVO/long-video and NIAH. r=.7 selected on StreamingBench SQA; create validation rather than reusing tuning results as test.
2610.05897 Fitting Vision Adapters at Frontier Scales	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. An approximately 50M visual projector connects frozen encoder/reader, initializes with shared-vocabulary affine alignment, and uses empty-think SFT followed by adapter-only CISPO image-math RL to restore reasoning. Positive reward requires nonempty reason/answer. Qwen4–32B matched recipe supports controls; 550/743B MoEs differ in training and rewards. Blind/corrupted-image controls are essential.
2610.05896 TasteRoute: Personalized Routing for Video Generation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TasteRoute uses frozen request embeddings, generator/user embeddings and cross features, BT pairwise plus ordinal quality, five prompt-disjoint general folds and per-annotator train/val/test with user dropout .4. Ranks within budget without generating video. Nine annotators and 3,128 videos; 55.2% conflicting-preference accuracy is limited. No population generalization or deployment claim from oracle verifier.
2610.05885 The Optimization Landscape of Learning Compacted Context Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Fast Diamond simplifies a Perceiver KV writer with explicit single/recursive compression and matched budgets. Candidate learned KV method. The canonical appendix resolves inconsistent main-text readout wording as frozen, with discrepancy noted.
2610.05879 Learning to Learn a Language	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. PFLM samples a fresh recurrent SCM prior per sequence for byte-level pretraining, then freezes weights and predicts using long prefixes. Core generator definition and 300M model/weights are public; candidate training paradigm.
2610.05872 Off-Policy Merging Beats On-Policy Self-Distillation for Continual Learning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Merge off-policy donor fine-tune deltas after Fisher/KFAC retention-sensitive projection and scaling. Qwen/OLMo continual learning, distinct from 2610.00767 belief transplant and earlier GRAFT 2609.37868 trajectory swapping.
2610.05861 Imagine to Act: High-Fidelity Data Synthesis via Image Editing World Model for Scalable GUI Agent Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Infinite-Dreamer learns deltas from before/after/action/next action. AppDesigner text drives delta-conditioned FLUX2Klein screenshot edits, VLM filters steps, and SFT trains GUI actor. AndroidControl train/val for world model; actor only synthetic, online AndroidWorld/MobileWorld tests. Future design is allowed in synthesis, not gold next state/plan at evaluation. Not a real app state machine.
2610.05842 HLA: Expressive Hybrid Linear Attention via Chunk-Wise Dynamic Mixing	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. HLA interpolates complete chunk affine GDN transitions with identity via query gates and composes them temporally, unlike static state weighting. Candidate long-context operator.
2610.05774 AdaSpark: Adaptive DSpark with Online Learning for Tree Verification and N-gram Fill	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. AdaSpark merges DSpark proposals with request-local n-grams, ranks legal tree prefixes by acceptance probability, and chooses verification width from online lower-quantile latency and acceptance models. Candidate speculative controller.
2610.05744 CIPHER-MoE: Balancing Efficiency and Routing Fidelity in Trillion-Scale MoE Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CIPHER-MoE keeps token Top-K, adds expert cosine affinity, capacity admission, and locality loss, and separates strict drop from reroute. Actual large-scale SFT efficiency and quality evidence.
2610.05715 Rotated, but How Far? Diagnosing and Improving Object-Rotation Reasoning in VLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. RotationCue uses three learned relation queries for object identity, rotation presence, and angle range, injecting conditional angle cues into a frozen VLM. ALOI/ModelNet40 can construct training; OR-Bench is object-disjoint full test.
2610.05685 Spend Bytes on Breadth: Precision-Count Trade-offs for Decode-Time KV Compression in Long Chain-of-Thought Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. BreadthKV combines KIVI quantization and SnapKV-D eviction under equal decode-KV bytes and calibrates bits independently on AIME22–23. Studies broader low-precision versus narrower high-precision cache.
2610.05664 StageVLN: Spatial and Trajectory Auxiliary Guidance for Efficient Vision-Language Navigation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. StageVLN aligns multilayer current-image features to a frozen geometry teacher by cosine, and trains auxiliary relative heading sin/cos and expert forward-step progress. Pre-action state avoids teacher-forced GT current action. Joint R2R/RxR training and unseen validation; remove teachers/projectors/heads at inference. Progress is not success or a STOP rule.
2610.05637 Visual Grounding Safety in Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Grounding safety matches VQA and localization by intent. SFT mixes harmful refusal, benign self-distillation with neutral evidence, and 1,548 grounding capability examples; plain and placeholder formats train separately. Score refusal from final answer/coordinates, not thinking. Failed localization can mimic safety, so report PointArena and over-refusal separately. Placeholders are not actionable coordinates.
2610.05630 Atomic Visual Entailment: Enhancing Zero-Shot Vision-Language Reasoning through Atomic Fact Decomposition and Learned Selection	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. AVE freezes the decomposer, splits original text into atoms, and pools direct/atomic candidates from two VLMs. XGBoost predicts labels from score/entropy/agreement without requiring pool membership, then GroundingDINO locates representative justifications. Select on 20% validation of 47,203 training items before fixed SNLI-VE dev/test. 94% oracle is not deployment; Flickr30kEntities approximates localization, not full faithfulness.
2610.05538 LiFT: Loop Flow Transformers	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. LiFT uses random continuous-depth coordinates to supervise iterative velocity correction with a shared DiT core. Detached initial velocity anchor and prelude auxiliary loss are essential. Matched tokens, blocks, depth, and near-equal FLOPs support comparison to dense.
2610.05484 Universal Test-Time Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. uTTT shares fast-weight pools across layers with separate read/write routers. Matched state, active compute, and tokens support the core architecture.
2610.05418 EvoMem-VLA: State-Evolution Memory for Long-Horizon Robot Manipulation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. EvoMem-VLA freezes paired DINO delta tokenizer and interleaves initial/recent/event state and delta memories. Keyframe head predicts future record offsets, committing only when reached. Routing is configured, not learned. GT subtasks in training vs generated test subtasks cause exposure bias; closed loop sees observations only. RMBench/RoboMME and 80 real episodes across two embodiments.
2610.05402 Task Vector Descent: Learning from Non-IID Batches	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TVD partially combines task vectors from consecutive same-distribution minibatch updates, interpolating Adam first moments/steps and second moments in square-root space. Fixed sample budgets and isolated train/evaluation for SFT and GRPO.
2610.05318 Robust Parameter-Efficient LLM Adaptation on Analog Hardware	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Use functionally equivalent channel scaling on frozen base, calibration-outlier splitting for LoRA, and digital residual accumulation for subthreshold updates. Candidate analog-error training diagnostic.
2610.05274 Answer with Evidence: Consistency-Aware Grounded Visual Question Answering for Roadside Traffic Scenes	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. EtA uses a box-then-answer sequence. ECPO clusters boxes by cross-sample IoU over G rollouts with support ≥max(2,ceil(.25G)) and applies Hungarian coverage plus format/reference answer reward before GRPO. No GT boxes in RL does not mean label-free. RoadSceneVQA/gRefCOCO/CARPK span tasks; structural consistency existence does not prove full grounding.
2610.05273 When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLA	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SAPrune calibrates three stable action-attention depths without reward, then budgets top-attention and functional-diversity paths using QR explained mass times attention concentration. Prune only visual tokens. Three VLAs on LIBERO/SIMPLER and four real runs; RTX5090 SDPA latency differs from FA2 and cannot be replaced by FLOPs.
2610.05190 FORGE: Verification-Gated Behavioral Repair for Generative Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FORGE uses prefix/LOO toxicity localization, classifies tokens, then edits lm_head by low-rank QP or MLP null space, regenerates, and commits/rolls back. Candidate auditable weight repair.
2610.05185 Recurrent Latent Visual Search for GUI Grounding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ReLaViS feeds latent slots back as spatial distributions and evidence vectors; CWC combines candidate-quality lower bounds and truncated final Gaussian KL. Train warmup, GT-latent teacher forcing, and predicted self-conditioning. Test screenshot/query only with 3×3 peak-neighborhood coordinates. 100k train/5k validation and five isolated GUI benchmarks; grounding is not agent task completion.
2610.05126 Selecting Repetition Counts Across Model Scales in Data-Constrained Pretraining	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Use seed bootstrap/log-N regression on small-scale repetition-loss curves to select Top-3 and a two-term curve argmin. Executable pretraining budget selection with frozen choices before target prediction and independent test.
2610.05097 ReMAP: Restoring the Perceptual Cycle with Reasoning-Time Latent Visual Memory	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ReMAP uses global/local latent visual memory, region localization, staged formation/access training, and RL for when to retrieve. Ten benchmarks; retain real visual memory, not a text oracle.
2610.05066 Salvation Lies Within: Eliciting Inherent Style Transfer in Step-Distilled Diffusion Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. StyleForge's training-free agent extracts render, color/light, composition, and reference-content states, then compiles constraints from color/brightness statistics in fixed order to preserve style, weaken composition, and exclude content before few-step FLUX2Klein. 31 styles × 16 prompts and three Qwen judge controls. Structured prompt operator, not diffusion training or distillation; model scores are not human preference or live A/B.
2610.05026 GeoBridge-VLA: Geometry-Aware Residual Adaptation for Vision-Language-Action Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GeoBridge-VLA decodes K4 neighborhood geometry from native visual features with simulated depth or real DA3 pseudodepth, then freezes geometry/VLM and adds zero-init 1×1 residuals gated near .002. Flow plus RMS residual penalty. SmolVLA LIBERO/OMY only; unchanged token count does not remove decoder compute. No cross-backbone verification.
2610.05025 Triggering Generalist Reasoning via Predictive Uncertainty for Dual-System VLA	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TUD separates π0's heavy PaliGemma cache and current SigLIP/Q-Former/action expert. Across observations, re-predict fixed slots under one anchor and use SD to proxy plan change. Online p70 of running means and maxskip trigger refresh; execution uses committed chunks, not every probe. VLA-Arena/CALVIN/real hardware; variance is not calibrated epistemic uncertainty and probes cost compute.
2610.05024 LightVLN: Efficient Aerial Vision-and-Language Navigation with Compact Memory and History-Guided Local Aggregation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. LightVLN keeps one learned-query token per historical frame in FIFO and uses history selection to aggregate fixed 32 local regions in the current 16×16 grid. Separate OpenFly/AerialVLN training and unseen campuses. Full model 1.276B, not .5B; separate 4090 profiling from closed-loop routes. No extra memory encoder does not remove vision backbone.
2610.05023 Look Where You Say You're Looking: Self-Grounded Attention for Visual Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Self-Grounded Attention localizes each CoT observation with GroundingDINO, combines selected-head regional saliency, exact-match/judge, and format reward for GRPO. Head selection uses VisualCoT relevance and attention is not causal faithfulness. SaliencyR1-8k training and 25 VQA lmms-eval tests. Preserve real locator/judge and eight-H100 budget, not answer-aware fixtures.
2610.04805 ExStereo: Lifting 2D Vision-Language-Action Models to 3D with Explicit Stereo Representations	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ExStereo builds four-direction orthographic depth/XYZ/valid features from frozen FastFoundationStereo, calibrated fxB/d, and tabletop RANSAC. Action-query cross-attention residuals join the VLA without replacing prefix/VLM. Seven 3,500-episode midtraining tasks and ten evaluation tasks/objects are disjoint; post-training uses 100 episodes/task. RoboTwin and three real tasks with 20 runs each. Count depth/calibration cost and preserve percentage-point success differences.
2610.04792 Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GU edits Grassmann subspaces for missing images; candidate multimodal robustness operator.
2610.04753 More Value per Key: Asymmetric Sparse Attention for Faster LLM Decoding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SAGA decouples key and value head counts and refreshes approximate top-N thresholds periodically. Convert pretrained GQA with data-weighted key merging, minimum-weight matching, and KL+CE distillation, not head deletion alone.
2610.04721 Knossos and Ariadne: Benchmarking and Learning Complete Diagram Topology Extraction with Vision-Language Models	Full experiment not reproduced; review DONE; affordable components considered separately	Google/Meta confirmed; see priority list	Accepted in review; core implementation and public verification pending. Meta-coauthored Ariadne predicts node inventory, emits typed edges from groups of three active sources to all predicted nodes, merges deterministically, and traces separately. Inference does not read GT inventory. Knossos has 18k/1200 across six domains with different seeds, plus 162 TopoBench diagrams. Within-template seed isolation is not unseen grammar generalization. Public dataset/repo links are given, with independent value in method and graph evaluation.
2610.04651 GS-Codec: A Gaussian-Splatting Bottleneck for Neural Audio Coding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GS-Codec fits latents with an inner loop of continuous 1D Gaussians, trains the codec with stop-gradient and commitment, trains a latent reconstruction predictor instead of 300 Adam steps, and calibrates k-means scalar quantizer parameters on a holdout after training. Fixed centers need no transmitted positions; bitrate remains discrete. Distinguish 2,339-hour and 54k-hour configurations across LibriTTS/LJSpeech/LibriSpeech. Some baseline values are citations rather than reruns. The 7.9 ms L40S number is encoder-only for noncausal three-second clips, not end-to-end streaming latency.
2610.04616 PerturBot: Breaking Shortcut Priors in Vision-Language-Action Models with Perturbative Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. PerturBot's V/C/R respectively use task-preserving viewpoint perturbations, state/skill language relabeling, and relabeling of executed off-target motions, mixed with the original flow loss rather than invented recovery to failed targets. Paired GroundFscore uses shared noise to test null invariance and causal sensitivity separately; calibrate Δaction with two expert records from the same state and preserve denominator flooring and filtering. This is a diagnostic, not a capability certificate. Fifty real PnP layouts and 50 RoboTwin tasks include equal-budget/full-factorial ablations; 84% success is not merely an offline score.
2610.04607 ForeAct3D: Policy-Grounded Future World Modeling for VLA Policies	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ForeAct3D decodes current/future depth, three-class masks, and poses from six policy-feature queries; future decoding accepts only policy-generated actions. Training uses background staticity, Kabsch rigidity by GT instance, and kinematics; inference needs no future or GT inputs. Same-pixel correspondence is a short-term approximation, and strong supervision and rigid-body assumptions are material. LIBERO 98.3, CALVIN 3.73, and 37.8% real-robot performance have ablations. Geometry baselines use different horizons, and predicted-mask localization still uses GT instances; do not claim GT-free instance detection.
2610.04580 Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DCAT is a cross-modal alignment and transfer candidate with a clear closed-form solution and public project code entry point.
2610.04518 Length Generalization Needs Proper Regularization	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. VPAD computes affine dropout preserving mean and variance for each token's uncentered activation, with a shared mask after RMSNorm. Public training, evaluation, and weights make it a candidate length-generalization training operator.
2610.04512 Correctness Is a Direction: Geometric Answer Selection in Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Geometric Answer Selection uses 50 labeled calibration questions to form a direction from correct-minus-incorrect mean normalized hidden states. One forward per candidate projects to select an answer, making it a frozen-model readout candidate.
2610.04506 EgoExo-Next:Benchmarking Vision-Language Models on Visual-Option Next-State and Cross-View Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. EgoExo-Next provides 2,503 visual four-choice questions and four ego/exo present-matching or future-state tasks. Later distractors discourage choosing the farthest future. Candidate-only, shuffled-sequence, mismatch, and CLIP/DINO controls support diagnostic value. Official HF provides input images, input_meta, separate answers, and loading rules; give the model only images and input metadata, not gold or time. Paper §4.1 averages both directions, but the HF example describes requiring both directions correct: freeze and report the scoring version. Images retain six source licenses, and annotations are noncommercial research only, not uniformly open for commercial use.
2610.04381 Beyond Plausibility: Verifiable Fine-Grained Image Editing on Structured Assets	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. VeriEdit-Bench renders exact targets and protected areas from editable SVG, chart, and HTML sources. Lab distance decomposes fidelity, preservation, localization, and magnitude with per-example harmonic scores, adding executable image-editing evaluation.
2610.04318 Rethinking Long-Video Efficiency: A Joint Allocation Perspective on Frames, Pixels, and Front-End Latency	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. LoHi splits one decoded frame pool into low-resolution video and a few high-resolution images with common token and front-end decoding budgets. SemDiv combines query relevance and DPP to choose high-resolution frames.
2610.04303 What to Preserve in Recursive Computation: A Local Predictive Sufficiency Principle	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Local predictive sufficiency estimates compression deficits from finite ridge tests with full-covariance balancing and projects real optimizer proposals onto half-spaces defined by grouped predictive constraints. Language memory, TGN, VLA, and recurrent policies provide trainable mechanisms and controls.
2610.04225 FlashGaze: Training-Free Multi-Scale Patch Pruning For Efficient Video Understanding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FlashGaze chooses Drop/Merge/Keep before vision encoding by motion-compensated error, using three-level quadtree dynamic programming under a pixel budget. Qwen3-VL/NVILA and public video benchmarks support verification.
2610.04204 Clean: Second-order LLM Training at Linear Memory Cost via Nyström Sketching	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Clean/Q-Clean reduces optimizer state through two-sided thin Nyström factors and four complementary block moments. Clear definitions and measured LLM pretraining make it an optimizer candidate.
2610.04152 Kepler4D: Controllable Future Video Generation via 4D Scene State Evolution	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Kepler4D uses DA3, SAM2, and a VLM to obtain object proxies. Chain-of-Motion produces structured movement and renders 3D control for VerseCrafter. DL3DV/DAVIS support conditioned continuation experiments.
2610.04140 ExpertMuon-Compass: Alignment-Guided Step Sizes for Mixture-of-Experts Training	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ExpertMuon-Compass preserves Muon matrix directions and changes expert step radii with family- and row-alignment scalars. Separate tuning seeds and paired evaluations define a training-operator candidate.
2610.04139 From Sight to Foresight: Predictive Spatial Reasoning in Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SpatialMind aligns vision tokens with bounded log-depth correction plus metric and motion encodings, training state chains and answers in GROUND/CURRENT/DYNAMICS/FUTURE stages. External VSI/OSI/VLM4D allow transfer checks.
2610.04118 LongSocialBench: Do Long-Context LLMs Understand Online Discussion Threads?	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. LongSocialBench releases 1,462 questions over 94 full discussions with an exact MCQ protocol, covering structured long-context social reasoning. Admit only gold-free inference evaluation; report oracle conditions separately as diagnostics.
2610.04047 Periscope: Extending Frozen Language Models Beyond Their Context Window	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Periscope defines local/cross-span probing and evidence graphs for frozen models, making it a long-context reasoning and retrieval operator candidate.
2610.04009 SUAVE: Unified Video-Action Models via Masked Diffusion	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SUAVE unifies discrete video and actions under masked diffusion as a new trainable model candidate.
2610.04002 Do Language Models Need a Trainable Input Embedding Table? Fixed Minimal Token Codes at 1.7B-Class Scale	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Fixed minimal binary token identities plus deterministic lifting can form an executable architecture/mechanism diagnostic, with public checkpoints. Do not present it as a quality or efficiency winner.
2610.03741 Logit-Aware MIMO AirComp for Distributed Mixture-of-Experts LLM Inference over Wireless Edge Networks	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Finite differences estimate MoE aggregation sensitivity to next-token logits, guiding power-constrained MIMO AirComp transceiver weighting. This enables mechanism diagnostics under wireless noise.
2610.03695 Language Models that Play Chess and Explain Their Moves	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Queen connects a frozen Leela position encoder to SmolLM3, runs a four-stage spatial/temporal curriculum, then seven rounds of self-evolving SFT on Stockfish-verified recursive subposition explanations. Lichess, public engines, and held-out puzzles support checks.
2610.03632 World Embedding Benchmark	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. World Embedding Benchmark provides public physical video/text retrieval, regression, pair discrimination, and new-model evaluation interfaces, suitable as physical-understanding infrastructure.
2610.03516 XGenAct: Geometry-Enhanced World Action Models through Cross-Task Generation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. XGenAct combines a deterministic geometry codec with shared video-flow task mixtures as a world-and-action joint-training candidate.
2610.03389 From Patching to Pruning Visual Computation in Vision Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. P2P tests replacing cached vision projection with means, scans prefix/suffix sensitivity for safe layers, and actually skips vision attention/MLP, optionally with patched-graph LoRA. Seven public benchmarks and separate proxy, validation, and test data support reproduction.
2610.03199 Predicting and Repairing Merge Collapse in Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Prism norm-scales and averages task vectors, soft-thresholds by cross-expert variance, and applies cancellation/dispersion gates to avoid removing unique knowledge. The deterministic, data-free repair is fully specified.
2610.03198 KV\(^2\): A Self-Refining KV Cache	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. KV² uses KeyDiff averaged across layers and heads to choose sparse reconstruction queries per chunk, appending them after the full cache for actual causal forwards. Local maximum attention creates global retention; iterative proxy replacement is possible. It was already queued, so improved evidence is not a new paper count.
2610.03123 Foresight: planning future perception in streaming VLMs without retraining	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Foresight separates one frozen model into a video-writing Ingest and read-only-snapshot Think. It plans sampling rate, retained classes, compression, and response timing. StreamingBench and OVO support real causal streaming verification.
2610.03084 NegT2IBench: When Negation Changes the Picture. A Polarity Benchmark for Text-to-Image Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. NegT2IBench uses 4,800 positive and negative attribute/relation prompts in six types, checking object existence and grading with detector, SAM, depth, and SigLIP2. It adds executable negation evaluation for image generation.
2610.03039 HyperThink: Text-to-Parameter Hypernetworks for Efficient Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. HyperThink generates frozen-LLM layer biases from queries under VQ constraints, a distinct algorithm candidate.
2610.03034 Adaptive Second-Order Solvers for Fast Stochastic Diffusion Sampling	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. The PI diffusion solver uses stochastic Heun first-/second-order error plus previous error for step size and aggregates adaptive traces into static modality-specific schedules. FFHQ/ImageNet64, LM1B, and public EDM checkpoints support verification.
2610.03002 Recursive Self-Improvement in Unified Multimodal Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Unified multimodal RSI uses execution-verified drawing programs as labels. Diagnostics only choose data distribution; separately update generation and comprehension experts and then merge, with equal-step and equal-verified-count controls.
2610.02951 Dynamic Expert Pruning for Multi-Agent Systems	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DEP is already queued. A request text encoder and MLP predict layerwise expert top-K, trained with impact-label BCE and soft-mask self-distillation KL. A CPU expert pool delta-loads GPU slots. Retain the candidate without double-counting.
2610.02887 Revealing Epistemic Uncertainty in MLLMs via Causal-Invariant Masking	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CIM compares semantic distributions after masking while preserving question-relevant ROI and estimates uncertainty with EED expected embedding drift. VQAv2, OKVQA, AdVQA, POPE, and MME supply public data and equal-sampling controls.
2610.02880 Found but Not Read: When Extracted Text Closes the Retrieval-Reading Gap in Document Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FoveDoc-Bench freezes ColQwen2 top-16 pages and compares image-only with image plus CPU OCR/PDF text for the same pages. Lexical top-16 chunks have 300 characters each. This separates retrieval and reading bottlenecks.
2610.02876 Seeing, Saying, but Not Using: From Reportable Spatial Facts to Usable States in Multimodal Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. OSS normalizes shared spatial facts across contexts and uses ordinary autoregressive training on answer, state alignment, and state trajectory pools. Derivatives inherit observation-world splits; test three SpaceConflict types and paired scores.
2610.02841 How Robust Is Multimodal Claim Verification to LLM Rewriting?	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SciClaimEval pairs claims with supporting and refuting images and measures verdict changes under natural paraphrases and single-word injections, for multimodal classification and reward-robustness diagnostics.
2610.02840 PointWAM: 3D World Action Modeling for Dexterous Robotic Manipulation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. PointWAM pretrains unified 3D scene and hand trajectories, then retargets actions, qualifying as a robotics foundation-model candidate.
2610.02832 FastOPD: On-Policy Distillation for Lightweight VLA Deployment	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FastOPD samples on-policy states via one student flow-map jump, matches diagonal teacher velocity, and propagates finite-step maps using midpoint self-consistency. LIBERO, RoboTwin, and real robots were evaluated; single-step action-expert training needs a genuine checkpoint.
2610.02718 Revisiting Visual Representation Enhancement of VLMs via Kernel Canonical Correlation Analysis	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. 3vKCCA trains three-view projections from CLIP vision, frozen DINO, and text, a vision alignment candidate with theoretical limits.
2610.02713 WakeKV: Reactive, Reversible KV Residency for Heads That Change Their Minds	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. WakeKV keeps a full CPU reservoir and per-head GPU LRU residency with reversible retrieval when demand changes, suitable for cache-migration experiments.
2610.02697 GeoScaffold: Learning Compact Geometric Latents via Reconstruction for Efficient Vision-Language Navigation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GeoScaffold first trains a frozen depth VQ-VAE, then eight geometry queries reconstruct depth codes, depth maps, connectivity, and traversability under imitation learning. Remove reconstruction heads and tokenizer at deployment. R2R-CE/RxR-CE support checks.
2610.02683 RAOA: Alternating-Operator Neural Computation with Programmable Radio Propagation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. RAOA alternates normalized energy-gradient problem steps and Hadamard mixing, sharing two control parameters across iterations. A zero-initialized residual adapter supports recurrence studies in frozen LLMs.
2610.02666 CHASE-VLA: Post-Training Quantization Framework for Vision-Language-Action Models with Chunk-Aware Scale Estimation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CHASE-VLA predicts layer/denoising-step group activation-scale distributions from earlier generated action chunks. Quantile loss and variance jointly train a scale estimator, preserving a W4A4 action expert.
2610.02657 Context-Tower Conversion Preserves Generation While Freezing Retains Knowledge: Low-Budget AR-to-Diffusion Conversion of MoE LLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Low-budget AR-to-diffusion conversion uses a frozen context tower and layerwise KV concatenation with matched token/parameter protocols and public-task evidence. Admit as a research conversion operator, not serving acceleration.
2610.02597 GRAFT: Growing Agglomerative Foundation Models via Continual Teacher Distillation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GRAFT retains old teacher functionality with a historical student anchor and admits new vision teachers using TSRT and image–text relation KL. It is a defined continual-distillation operator.
2610.02593 Fisher-Guided Submodular Data Selection for Continual Pre-Training of Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Dual Fisher geometry and streaming submodular selection define a CPT data algorithm candidate, not industrial deployment evidence.
2610.02462 Capability Scaling-Down Laws for LLM Compression	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Predict pruning, quantization, and distillation loss separately by capability: shared pruning-density curves, bit/group-size grid interpolation, and distinct data-volume and exposure terms for distillation. This can inform Evolve budget allocation.
2610.02388 Octrees as an Explicit 3D Language	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. OctLLM uses a sparse six-level octree, separate 3D branch, and 3D RoPE with frozen Qwen2.5-VL. It also trains occupancy completion and uses TRELLIS decoding. Objaverse-XL, HSSD, ABO, ShapeNet, Toys4K, and PointLLM200 support checks.
2610.02344 Drive vs. Decay: On the Training Dynamics of Joint-Embedding Predictive Architectures	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ResidualPred uses residual predictors and diagonal attention initialization to suppress JEPA collapse. Table 5 has five seeds and matched ImageNet1K 30-epoch controls, supporting the mechanism.
2610.02323 World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ProAct publishes the training core. Proposal, world calibration, and anisotropic flow are candidates, starting with executable LIBERO.
2610.02298 EditHero: A Benchmark for Long-Horizon Part-Level 3D Editing and Vibe Modeling	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. EditHero reconstructs exact 3D targets for 457 chains and 2,755 turns from part libraries and operation logs. Self-rollout separates edited and preserved areas and compares no-op and regeneration controls for multi-turn generation evaluation.
2610.02196 InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. InterEvolve evolves staged reward programs over a frozen forward/backward motion prior. Inner CMA-ES tunes parameters and an outer LLM edits structure, using fixed scenarios and reusing verified skills. OMOMO/GRAB and eight simulated task types support evaluation.
2610.02091 GeoLatent: Geometry-Guided Latent Structuring with Routed Optimization for 3D Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. GeoLatent separates shared geometry and residuals and uses three-stage routing optimization to block visual shortcuts into answers. DA3/VGGT targets supervise only latents. SPAR/SpatialLadder and SPBench/ViewSpatial enable actual evaluation.
2610.01962 SIEVE: Selective attention-value Suppression for Vision-Language Models Unlearning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SIEVE alternates negative CE and zero attention values on forget data with positive CE and frozen-reference value matching on retain data. Synthetic Profile Visual-QA/Textual-QA supports selective forgetting tests.
2610.01921 Cross-Lingual Alignment for Decoder-Only Models using MoE Routers	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. XLR aligns languages in middle layers by KL over sequence-mean sparse routing probabilities on parallel text. Its CPT auxiliary loss and packing are defined, with public training and evaluation, making it a foundation-model training candidate.
2610.01917 MoLE: Mixture of Latent Experts for Complementary Visual Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MoLE introduces routed vision evidence, expert value transforms, and isolated latent experts plus summary topology. Public data and a detailed appendix support a faithful vision-reasoning implementation.
2610.01815 Debias Anything: Fairness with Diversity without Supervision in Diffusion Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Debias Anything trains a semantic adapter on frozen diffusion models using attribute directions, target proportion guidance, and disagreement diversity for batch-level debiasing. Public faces and diffusion checkpoints support quality/fairness comparisons.
2610.01741 ATI-VLA: Action-Centric Predictive Vision-Language-Action Models via Actionable Alignment Then Adaptive Injection	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ATI-VLA shares a quantized action/observation space with hierarchical side injection, a VLA algorithm candidate.
2610.01710 CoEvolve: Construct-to-Edit Visual Grounding with Bidirectional State Refinement	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CoEvolve trains ordered multistep bounding-box progress with RER-GRPO, then denoises coordinates with BDR while retaining reasoning text. RefCOCOg train and held-out RefCOCO/+ support tests.
2610.01687 Architectural Sampling: Test-Time Scaling via Computational Diversity in Frozen Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Architectural Sampling repeats frozen VLM layer windows with different recursion counts to generate architectural rollouts, then trains with label-free self-reward TTRV-GRPO. Public models and 12 benchmarks support verification.
2610.01649 CrossGMN: Graph Metanetworks for Cross-Architecture Weight-Space Transformations	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CrossGMN is a small-model compression or Evolve meta-operator candidate, not established LLM compression support.
2610.01640 Not All Error Yields to Scale: Where Scaling Stops in Vision-Language Inference	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Separable Inference Scaling Law fits error floors and power laws separately by backbone parameters and actual vision token count, using a cost surface to choose held-out settings. It can inform multimodal budget planning.
2610.01637 Fusing Visual and Textual Representations via Multi-layer Fusing Transformers for Vietnamese Visual Question Answering	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MFT fuses PhoBERT/ViT through multilayer bidirectional cross-modal attention for Vietnamese VQA classification. Public ViVQA supports training and EM/token F1, a narrow multimodal candidate.
2610.01625 Beyond Domain-Level Adaptation: Margin-Oriented Semantic-Appearance Interaction Correction for Personalized Federated Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MOSAIC combines shared/private prompts, domain–class residuals, low-rank class corrections, and instance gating over frozen CLIP. Public Office31/Home/DomainNet federated classification supports tests.
2610.01605 Hob-VL: A Benchmark for Visually Grounded Boolean Reasoning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Hob-VL provides public symbolic and natural-language visual Boolean composition and De Morgan consistency evaluation, supporting new-provider inference and strict offline grading.
2610.01595 Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TAI compares forward/reversed video activations by layer and injects input-specific peak vectors into later layers with decay. Three frozen VideoLLMs and four public temporal benchmarks support evaluation.
2610.01537 FedFit: Federated Fine-Tuning of LLMs via Vector-Bank Parameterization and Quantization	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FedFit's disjoint dual banks, alternating federated aggregation, and low-bit communication are reproducible candidates, with formula ambiguities requiring review.
2610.01513 Decision Titan: Test-Time Training for Long-Term Memory in Offline Reinforcement Learning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Decision Titan combines RTG/state/action Decision Transformer with Titans fast-weight memory, updated by trajectory chunk. Repeated-memory X-Maze tests length extrapolation.
2610.01499 VTR-Bench: A Systematic Benchmark for Evaluating Visual Text Rendering in Video Generation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. VTR-Bench publishes 300 prompts and CoQ/carrier-transcription WER entry points for newly generated videos. Agentic keyframe generation is also researchable, but unmatched budget gains do not establish efficiency.
2610.01496 SALD: Self-Referenced Advantage Learning for Diffusion Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SALD compares a same-model easy-noise teacher without gradients and a hard-noise student, with batch-normalized differentiable advantage weights, sample EMA noise curriculum, and frequency discrepancy gates. Public diffusion data/checkpoints support actual training.
2610.01492 Q-SPT: Learnable Query-Based Compression for Low-Frame-Rate Speech Tokenization	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Q-SPT downsamples SenseVoice/DAC features using sliding-window query cross-attention, semantic FSQ plus conditioned acoustic RVQ, and joint spectral, adversarial, and text reconstruction training. Public LibriSpeech reconstruction, ASR, and TTS support verification.
2610.01286 Dyna3: VLM-Guided Training-Free 4D Reconstruction via Depth Foundation Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Dyna3 detects motion with best-match cross-frame DA3 feature similarity and uses Qwen2-VL/SAM3 semantic masks to separate static aggregation and per-frame dynamic point clouds. DAVIS/TUM/Sintel/DyCheck support checks.
2610.01243 When the Judge Acts: Auditing VLM-Guided Image Selection on Culturally Situated Prompts	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. JudgeActs freezes the CulturalFrames candidate pool and uses three cyclic order checks and a preset abstention gate, comparing against exact random expectation and a post hoc human oracle. It audits VLM selection reliability.
2610.01192 FlashBack: Knowing When to Remember in Streaming Vision-Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FlashBack separates native streaming state and a temporary Recall branch, routing by temporal evidence to historical retrieval/SideKV without writing Recall into native history. Public video benchmarks can test causal long-term memory.
2610.01172 Learning Rate Transfer for Hybrid Transformer-SSM Architectures	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Naive μP with AdamW transfers learning rates across widths in hybrid Transformer–SSM models, an executable parameterization/tuning recipe rather than a new guarantee.
2610.01127 Counting and Min-Cost Encoding for Tokenization in Large Language Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. CNF–MCE builds a vocabulary from direct substring statistics and actual-use filtering, then globally segments with boundary-aware dynamic programming. This is a defined foundation tokenizer candidate.
2610.01054 Capturing In-Context Learning Dynamics with Task Operators	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Task Operator extracts multiplicative softmax mass and additive context values by head and position from ICL forwards for demonstration-free replay. Analytic definitions, public code, answer-excluded validation, and isolated test make it a candidate.
2610.01037 SLIM: Simplex-Lattice Interpolation Merging	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. SLIM fits merging performance with a minimal quadratic simplex lattice of expert vertices and pairwise midpoints, then optimizes predicted AVG under constraints. It compares to random design under 36 evaluations with held-out OOD.
2610.00983 The Devil Is in the Reconstruction Loss Scale: Rethinking Optimization in LLM Quantization	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Replace PTQ reconstruction MSE with grouped RMSE, making output-gradient norm independent of nonzero error scale. Sample, channel, token, and element variants belong to one family and fit existing PTQ training comparisons.
2610.00981 NarrativeFlow: Flow-Based Vision-Language-Action Model Using Robot Velocity Fields	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. NarrativeFlow uses two task tokens and narrative differential supervision to strengthen language-conditioned robot flow, a candidate algorithm.
2610.00952 A Matched-Budget Audit Framework for Recaptioned Image-Text Supervision Distributions	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Matched-Budget Recaption Audit compares caption coverage, prefix concentration, CBU density, and image-support risk on the same images and text budgets, as a pretraining-data diagnostic.
2610.00899 TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. TOAST's stochastic action-token supervision is a defined VLA training operator, suitable for matched public LIBERO verification.
2610.00878 UniTrackPLA: Unified Panorama-Language-Action Model for Instruction-Guided Navigation and Dynamic Person Tracking	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. UniTrackPLA maps DINOv2/SigLIP panoramic multiview input with time/heading encodings to waypoints. WAC predicts post-action latents and uses observed error to replan. OmniTrack/OmniVLN support route-isolated verification.
2610.00864 Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. K-MF splits the MeanFlow time derivative into two interval JVPs, defining a trainable one-step action generator.
2610.00717 Sequential Functional Structured Tucker Compression for Large Language Model Attentions	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FTC measures actual sequential student activations, jointly budgets shared QKV Tucker and sparse core bytes, and separately calibrates WO against a teacher map. Same-metric SeqSVD and equal-storage core ablations make it a compression candidate.
2610.00666 VisionQ: VLM-as-a-Judge Taxonomy, Dataset and Benchmark for Qualitative Analysis in Computer Vision	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. VisionQ provides new VLM-judge benchmark execution and DPO code for auditing agreement with author claims, not objective visual quality or automated peer review.
2610.00663 Backdoor Containment via Expert Quarantine and Shutdown in LLMs	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. QES uses attention-based soft trigger localization, sample peakiness modulation, and four routing constraints to train expert LoRA and disable isolated experts. It needs no gold trigger and was run on multiple models, two tasks, and three attacks.
2610.00604 MIKASA-Robo-VLA: Benchmarking Memory in VLA Models for Long-Horizon Manipulation	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MIKASA-Robo-VLA offers 90 ManiSkill3 tasks, two RGB views and unified 7D state/action, success_once, and 22,500 demos, enabling executable VLA memory evaluation.
2610.00573 FORTE: Adaptive Scoring and Exact Keyframe Selection for Long-Video Question Answering	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. FORTE uses banded RBF-GP to estimate frame relevance and temporal-coverage marginal gains within a scoring budget. Exact O(KN) upper-envelope dynamic programming selects 64 frames. Public long-video QA supports checks.
2610.00559 PhysVista: Benchmarking Physical Intelligence in VLMs via a Perception-Reasoning-Assessment Loop	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. PhysVista uses public video frames and questions for perception, causal, and physics tasks, adding public physical-VLM infrastructure.
2610.00541 Random Recursive Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. RRM uniformly samples reusable layers with replacement at each step per example in actual stochastic recurrent training. TRM+RRM holds TRM fixed except layer ordering, providing a core control for small recurrent architecture research rather than established large-model ability.
2610.00499 Denoising Surface: Modeling and Predicting Inference Cost for Diffusion LLM Serving	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DWS separates diffusion decoding load into block survival and conditional denoise-step survival. Two-stage multiscale/Brier MiniLM supervision and CPU INT8 predictions drive SJF with hardware cost profiles.
2610.00497 Gumbel Straight Flow: Distilling Autoregressive Models into One-step Flow Maps	Full experiment not reproduced; review DONE; affordable components considered separately	Google/Meta confirmed; see priority list	Accepted in review; core implementation and public verification pending. Google DeepMind Amsterdam contributes to GSF: straight paths from AR Gumbel coupling, monotone Gaussianization, and diagonal-teacher/semigroup self-consistent flow-map loss. The paper includes architecture- and iteration-matched FMLM-re and a teacher-forced posterior-Gumbel training algorithm.
2610.00487 ScaffoldM3C: A Multimodal Sequential Monte Carlo Framework for Generative Stable Construction Planning	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. ScaffoldM3C conditions Gemma to generate bricks using position, category, and candidate-likelihood heads. SMC penalizes collision-repair displacement, reweights samples, and selects the longest stable sequence. StableText2Brick provides reconstructable supervision.
2610.00432 XOR-Trellis: Ultra-Low-Complexity Dequantization and Curvature-Aware Hadamard-Free LLM Quantization	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. XOR-Trellis uses 16-bit state and two branch bits per step to create a four-value FP4 palette. LDLQ objective adjustment and D-weighted Viterbi preserve curvature without Hadamard rotation.
2610.00399 Metacognitive Reasoning in Energy Based Models using Instance Based Learning Theory	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. MERITED uses a 191M EBT energy distribution and question-embedding experience similarity to choose MCMC depth or stepwise stopping. Twenty training-memory items form state/action returns, supporting a dynamic-compute diagnostic.
2610.00251 The Null Is the Hard Part: Exact Tests for Memorization in Generative Models	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. Model-level randomized label permutation and image-level matched conformal/BH controls can extend generative memorization evaluation. Public code includes actual test entry points and outputs. Candidate evaluation only, not an executed test or proof of no memorization.
2610.00204 Query Independent Variable Rate Visual Token Coding	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. RDTok uses query-independent PCA to construct token-level quantization rate–distortion curves and integer dynamic programming to allocate a total bit budget. Public ChartQA/COCO images and frozen VLMs can test compression.
2610.00040 DSSR-3D: Decoupled Reasoning for View-Dependent Referring in 3D Gaussians	Needs cost/scope reduction or delivery; not auto-queued	Not independently confirmed this round; not excluded	Accepted in review; core implementation and public verification pending. DSSR-3D uses FLAN-T5 to parse target, direction, and anchor, locates anchors with semantic Gaussian softmax, and fuses target scores with local directional decay in the camera XZ plane. Public Ref-LERF scenes support tests.
2602.16490 From Growing to Looping: A Unified View of Iterative Computation in LLMs	Full experiment not reproduced; review DONE; affordable components considered separately	Google/Meta confirmed; see priority list	Accepted in review; core implementation and public verification pending. A historical Google collaboration was missed: grow-first/loop-later plus middle-block recurrence has actual SmolLM training, 22 public tasks, and parameter, training, and inference budget controls. It is a recurrent-model candidate, not a new October paper.

See reconciliation.json for per-entry ledger references, titles, and current status.