A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Guanghao Ye, Khiem Pham, Xinzhi Zhang, et al.
Proposes RLSP, a post-training framework that decouples exploration and correctness signals during PPO to enable emergent reasoning behaviors in LLMs.
Qi Sun, Edoardo Cetin, Yujin Tang
Transformer-Squared is a self-adaptive LLM framework that adjusts only the singular components of weight matrices in real-time, using a two-pass dispatch and reinforcement-learned expert vectors to outperform LoRA with fewer parameters.
Rujikorn Charakorn, Edoardo Cetin, Shinnosuke Uesaka, et al.
Doc-to-LoRA (D2L) is a lightweight hypernetwork that meta-learns to perform approximate context distillation in a single forward pass, generating LoRA adapters for a target LLM to reduce inference latency and KV-cache memory.
Roeland Scheepens, Niels Willems, Huub van de Wetering, et al.
This paper introduces a flexible architecture for multivariate trajectory density maps using composable blocks and expressions, enabling analysts to model domain knowledge for visual exploration.
Bingxin Xu, Yuzhang Shang, Emilio Ferrara
BATON improves long-horizon robot manipulation by making subtasks the unit of exploration and adding transition-aware memory, boosting task success by 11.6% over SoTA.
Bogdan Georgiev, J. Gómez-Serrano, Terence Tao, et al.
This paper demonstrates AlphaEvolve, an LLM-guided evolutionary coding agent, as a tool for autonomously discovering novel mathematical constructions, rediscovering best-known solutions and improving several across 67 problems.
Ning Li, Qiqiang Lin, Zheng Wu, et al.
ColorAgent is an OS agent that uses step-wise reinforcement learning and a multi-agent framework to achieve state-of-the-art success rates on AndroidWorld and AndroidLab benchmarks.
Vatsal Venkatkrishna, Nico Daheim, Iryna Gurevych
Introduces parameter-space exploration for LLM RL via Perturbed Parameter Policy Optimization (3PO), improving downstream performance over GRPO at similar FLOPs.
Unknown
This paper proposes a method to transfer exploratory policies from simulation to real-world reinforcement learning, overcoming the sim-to-real gap even when direct transfer fails.
Unknown
This paper introduces a safe model-based reinforcement learning approach that guarantees stability during exploration and learning.
Unknown
This survey systematically reviews Low-Rank Adaptation (LoRA) methods and their variants for Large Language Models (LLMs), emphasizing their efficiency in adapting models without full fine-tuning.
Unknown
Introduces Correlated Low-Rank Adaptation (CoLoRA), a novel framework that explicitly models correlations among low-rank factors to improve parameter-efficient fine-tuning of convolutional networks.