An 8B Model Beats a 235B One at Science Reasoning

Also Worth Noting

• Mat-Pref: Verifiable-Reward Training Improves Compositional Reasoning in Inorganic Materials score 4
  关键词(3): fine-tuning, GRPO, reasoning; 顶会接收: ICML
• Beyond Value Benchmarks: Measuring Value-Structure Alignment in Large Language Models via Symmetric Q-Sorts score 4
  关键词(1): reasoning; 顶会接收: ACL
• Denoising-Enhanced Coarse-to-Fine Infrared Small Target Detection with Attention Prior-Guided Knowledge Distillation score 4
  关键词(3): lightweight, distillation, real-time; 顶会接收: ECCV
• Provably Efficient Policy-Reward Co-Pretraining for Adversarial Imitation Learning score 4
  关键词(1): pretraining; 顶会接收: ICML
• Drowning in Routine: Signal Dilution in Multi-Turn Agent Training score 4
  机构: Mila; 关键词(2): scaling, GRPO
• Multi4D: High-Fidelity Dynamic Gaussian Splatting via Multi-Level Competitive Allocation score 4
  关键词(1): real-time; 顶会接收: ECCV
• Beyond Flat Labels: Level-Restricted Contrastive Learning for Hierarchical Fine-Grained Vision Classification score 3
  顶会接收: CVPR
• Residue-Level Attributions in Protein Language Models Do Not Recover Allergen Epitopes score 3
  机构: ETH Zurich