Spectral Decay Recovers 7% Accuracy in W4A4 Quantization

Featured

• High Precision Audience Expansion via Extreme Classification in a Two-Sided Marketplace score 5
  机构: Google; 关键词(3): efficient, search, recommendation
• S2D: Selective Spectral Decay for Quantization-Friendly Conditioning of Neural Activations score 5
  机构: Amazon; 关键词(8): scaling, efficiency, quantization, deployment, fine-tuning
• Unbiased Approximate Vector-Jacobian Products for Efficient Backpropagation score 5
  机构: INRIA; 关键词(2): efficient, cost
• Depth Completion as Parameter-Efficient Test-Time Adaptation score 5
  机构: NVIDIA; 关键词(3): efficient, inference, fine-tuning
• Refine Now, Query Fast: A Decoupled Refinement Paradigm for Implicit Neural Fields score 5
  关键词(4): efficient, fast, inference, cost; 顶会接收: ICLR

Also Worth Noting

• A Critical Look at Targeted Instruction Selection: Disentangling What Matters (and What Doesn't) score 4
  机构: Microsoft Research; 关键词(1): fine-tuning
• A Generative AI Approach for Reducing Skin Tone Bias in Skin Cancer Classification score 2
  关键词(2): diffusion, synthetic data
• A Trajectory-Based Safety Audit of Clawdbot (OpenClaw) score 2
  关键词(4): agent, evaluation, safety, jailbreak
• InnoEval: On Research Idea Evaluation as a Knowledge-Grounded, Multi-Perspective Reasoning Problem score 2
  关键词(5): production, reasoning, search, benchmark, evaluation
• Competition for attention predicts good-to-bad tipping in AI score 2
  关键词(3): edge, attention, safety
• Differentially Private Retrieval-Augmented Generation score 2
  关键词(2): retrieval-augmented, RAG
• Event-based Visual Deformation Measurement score 2
  关键词(3): search, benchmark, cost
• Adapting VACE for Real-Time Autoregressive Video Diffusion score 2
  关键词(5): latency, real-time, attention, diffusion, cost
• Beyond Token-Level Policy Gradients for Complex Reasoning with Large Language Models score 2
  关键词(2): coding, reasoning
• LRD-MPC: Efficient MPC Inference through Low-rank Decomposition score 2
  关键词(6): efficient, efficiency, inference, latency, attention
• Multi-Turn Adaptive Prompting Attack on Large Vision-Language Models score 2
  关键词(3): vision-language, safety, jailbreak
• pFedNavi: Structure-Aware Personalized Federated Vision-Language Navigation for Embodied AI score 2
  关键词(2): vision-language, embodied
• Boule or Baguette? A Study on Task Topology, Length Generalization, and the Benefit of Reasoning Traces score 2
  关键词(2): reasoning, benchmark