Machine Learning

Self-Taught Lookahead improves LLM search via self-supervision, matching costly methods at a fraction of the compute!

KODCODE: A new synthetic coding dataset with verified solutions and tests, enabling state-of-the-art performance for coding LLMs.

MultiAgentBench: A benchmark for evaluating collaboration and competition in LLM agents across diverse, interactive scenarios with novel metrics and protocols.

PQI: Accurately identify sensitive weights in post-quantization to enhance LLM compression & performance!

SoS1: O1 and R1-Like Reasoning LLMs are Sum-of-Square Solvers.

LLM can now reason and correct itself using self-generated data, achieving performance on par with external reward models!

OneRec: A unified generative model that replaces the traditional retrieve-and-rank strategy, significantly improving recommendation quality in real-world scenarios.

Stable-SPAM stabilizes 4-bit LLM training, outperforming Adam.

DVPO: A lean RLHF framework that decouples value & policy optimization with global value guidance, cutting GPU use by 40% and training time by 35%.

TAG: A decentralized framework for scalable multi-agent hierarchical reinforcement learning.

f-distill: One-step diffusion models through f-divergence minimization, outperforming reverse-KL with better mode coverage and lower variance.

MONSTER: Large datasets for time series classification!

UPCORE reduces unintended unlearning effects via coreset selection, balancing knowledge removal and utility preservation.

S*: Hybrid test-time scaling for code generation, boosting both coverage and selection accuracy.

ReQFlow: Efficiently generate high-quality protein backbones with rectified quaternion flow, outperforming existing methods in speed and designability.

MLGYM: A new framework & benchmark to advance AI Research Agents

Logic-RL unlocks LLM reasoning via rule-based reinforcement learning, generalizing to math problems after training on logic puzzles.

PURE: LLM-driven user profile management boosts recommendation by harnessing user reviews for personalized insights while tackling token limits. PURE enhances LLMs for better recommendations.

Unveiling the surprising potential of noise: transferable knowledge in semi-supervised heterogeneous domain adaptation (SHDA).

AdaptiveStep: Divides reasoning steps automatically through model confidence, enhancing PRM training & performance.