Daily Papers of 2026-09-28

  1. FuseReg: Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders 159 upvotes, #1 of 2026-09-28
  2. Disaggregated Quantization: Specializing LLM Prefill and Decode 91 upvotes, #2 of 2026-09-28
  3. RayOrch: Programming and Executing Lineage-Controlled Multi-Grain Dataflows for Foundation-Model Data Preparation 56 upvotes, #3 of 2026-09-28
  4. TimeEvo: Failure-Driven Self-Evolution of a Time Series Agent 32 upvotes, #4 of 2026-09-28
  5. Block Sparse Attention with Log-Linear Complexity 30 upvotes, #5 of 2026-09-28
  6. InternW0-Δ: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data 29 upvotes, #6 of 2026-09-28
  7. LastOPD: Taming Collapse in Latent On-Policy Distillation 28 upvotes, #7 of 2026-09-28
  8. Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors 24 upvotes, #8 of 2026-09-28
  9. FoMo: Forking Moment in Generative Trajectory as a Perceptual Distance 23 upvotes, #9 of 2026-09-28
  10. IndicBankBench: Evaluating Safety and Reliability of Language Model Assistants in Indian Retail Banking 21 upvotes, #10 of 2026-09-28
  11. D-JEPA: A Decision-Aligned Latent World Model 17 upvotes, #11 of 2026-09-28
  12. Morphometric Imitation: From Morphology and Contact Aware Hand Retargeting to Sim-to-Real Visuomotor Policy 16 upvotes, #12 of 2026-09-28
  13. Jev in the Wild: A Data-Driven Analysis of the Jev Model's Functionality, Applications and Ecosystem 15 upvotes, #13 of 2026-09-28
  14. Enhancing Photogrammetric Digital Surface Models with Pretrained Diffusion Models and Multimodal Conditioning 14 upvotes, #14 of 2026-09-28
  15. TrackEverything: Long Horizon Dense Tracking via De-Duplicating 3D Scene Representations 13 upvotes, #15 of 2026-09-28
  16. AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs 13 upvotes, #15 of 2026-09-28
  17. Depth-adaptive Inference of Looped Language Models via Continuous Depth Batching 12 upvotes, #17 of 2026-09-28
  18. SLCA-GRPO: Resolving Cross-Segment Credit Misattribution in Tool-Calling RL 12 upvotes, #17 of 2026-09-28
  19. Game Arena: Strategic LLM Evaluation in Competitive Environments 12 upvotes, #17 of 2026-09-28
  20. LightMIS: Ultra-Lightweight Medical Image Segmentation Without a Stage-Wise Decoder 11 upvotes, #20 of 2026-09-28
  21. TRACE: Temporal Audit and Condition-aware Evaluation of Streaming Video Understanding 11 upvotes, #20 of 2026-09-28
  22. Do Implicit Personalization and Explicit Styles Conflict? PsPLUG: A Lightweight Plug-in for Balancing Personalization and Style in Customized LLMs 10 upvotes, #22 of 2026-09-28
  23. VLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models 9 upvotes, #23 of 2026-09-28
  24. SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance 8 upvotes, #24 of 2026-09-28
  25. MOPD-Router: Rethinking Teacher Routing in Multi-Teacher On-Policy Distillation 8 upvotes, #24 of 2026-09-28
  26. CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation 7 upvotes, #26 of 2026-09-28
  27. Not All Ranks Are Equal: Budget-Aware LoRA Merging Across Tasks 7 upvotes, #26 of 2026-09-28
  28. BoundInk: Boundary-Aware Online Handwriting Generation 6 upvotes, #28 of 2026-09-28
  29. Paragraph Boundaries Are Not White Space:Compression Depth as the Signature of Hierarchical Structure 6 upvotes, #28 of 2026-09-28
  30. Evidence-Grounded Auditing of Identification Assumptions in Climate-Policy Causal Evaluations 6 upvotes, #28 of 2026-09-28
  31. Softmax Reparameterization for Output-Head Quantization 6 upvotes, #28 of 2026-09-28
  32. CodeGraph: Open-Taxonomy Knowledge Graph for Source Code with Wikidata Grounding 4 upvotes, #32 of 2026-09-28
  33. ZooWork-ShopRanker: An Open, Preference-Aligned E-Commerce Reranker 3 upvotes, #33 of 2026-09-28

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.