Adaptive DiT Patches Hit 3x Speedup, Mamba Improves by Subtraction

Featured

• Unified Latents (UL): How to train your latents score 8
  入选 HF Daily Papers; HF 热度: 44 upvotes (+4); 关键词(1): diffusion
• Arcee Trinity Large Technical Report score 8
  入选 HF Daily Papers; HF 热度: 16 upvotes (+3); 关键词(2): attention, MoE
• DDiT: Dynamic Patch Scheduling for Efficient Diffusion Transformers score 8
  入选 HF Daily Papers; HF 热度: 11 upvotes (+3); 关键词(4): efficient, inference, diffusion, cost
• ArXiv-to-Model: A Practical Study of Scientific LM Training score 7
  入选 HF Daily Papers; HF 热度: 7 upvotes (+2); 关键词(4): scaling, pretraining, transformer, reasoning
• 2Mamba2Furious: Linear in Complexity, Competitive in Accuracy score 7
  入选 HF Daily Papers; HF 热度: 7 upvotes (+2); 关键词(4): efficient, efficiency, attention, mamba
• Modeling Distinct Human Interaction in Web Agents score 6
  入选 HF Daily Papers; HF 热度: 2 upvotes (+1); 关键词(3): agent, agents, agentic
• Safe Continuous-time Multi-Agent Reinforcement Learning via Epigraph Form score 5
  关键词(3): efficient, agent, safety; 顶会接收: ICLR
• SpectralGCD: Spectral Concept Selection and Cross-modal Representation Learning for Generalized Category Discovery score 5
  关键词(4): efficient, distillation, multimodal, cost; 顶会接收: ICLR
• Retrospective In-Context Learning for Temporal Credit Assignment with Large Language Models score 5
  关键词(4): efficient, efficiency, agents, evaluation; 顶会接收: NeurIPS
• Position: Evaluation of ECG Representations Must Be Fixed score 5
  机构: MIT; 关键词(3): inference, pre-training, evaluation

Also Worth Noting

• Retaining Suboptimal Actions to Follow Shifting Optima in Multi-Agent Reinforcement Learning score 4
  关键词(1): agent; 顶会接收: ICLR
• Dataless Weight Disentanglement in Task Arithmetic via Kronecker-Factored Approximate Curvature score 3
  顶会接收: ICLR
• Fail-Closed Alignment for Large Language Models score 2
  关键词(3): alignment, safety, jailbreak
• Characterizing the Predictive Impact of Modalities with Supervised Latent-Variable Modeling score 2
  关键词(3): inference, multimodal, audio
• Fundamental Limits of Black-Box Safety Evaluation: Information-Theoretic and Computational Barriers from Latent Context Conditioning score 2
  关键词(3): deployment, evaluation, safety
• Bending the Scaling Law Curve in Large-Scale Recommendation Systems score 2
  关键词(7): scaling, efficiency, production, serving, inference
• WSDM Cup 2026 Multilingual Retrieval: A Low-Cost Multi-Stage Retrieval Pipeline score 2
  关键词(2): evaluation, cost
• Conv-FinRe: A Conversational and Longitudinal Benchmark for Utility-Grounded Financial Recommendation score 2
  关键词(3): alignment, recommendation, benchmark
• Dynamic Delayed Tree Expansion For Improved Multi-Path Speculative Decoding score 2
  关键词(3): efficiency, throughput, evaluation
• Sonar-TS: Search-Then-Verify Natural Language Querying for Time Series Databases score 2
  关键词(3): search, benchmark, evaluation
• Persona2Web: Benchmarking Personalized Web Agents for Contextual Reasoning with User History score 2
  关键词(5): agent, agents, reasoning, benchmark, evaluation
• Action-Graph Policies: Learning Action Co-dependencies in Multi-Agent Reinforcement Learning score 2
  关键词(2): agent, agents
• M2F: Automated Formalization of Mathematical Literature at Scale score 2
  关键词(2): scaling, agentic
• ReIn: Conversational Error Recovery with Reasoning Inception score 5
  入选 HF Daily Papers; 关键词(6): efficient, fine-tuning, agent, agents, reasoning
• WS-GRPO: Weakly-Supervised Group-Relative Policy Optimization for Rollout-Efficient Reasoning score 2
  关键词(5): efficient, efficiency, GRPO, preference, reasoning