A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Unknown
Introduces Lifelongagentbench, the first systematic benchmark for evaluating LLM agents as lifelong learners, revealing limitations of experience replay.
Nossa Iyamu
A deterministic, zero-model pipeline compiles passively captured screen activity into typed activity frames, reducing raw capture 86x and enabling 98.4% question-answering accuracy, while also measuring agent cost parameters.
Stefan Elfwing, Eiji Uchibe, Kenji Doya
This paper proposes SiLU and dSiLU activation functions and demonstrates that on-policy Sarsa(λ) with softmax action selection can outperform DQN in Atari games without experience replay or target networks.
Baohao Liao, Hanze Dong, Christof Monz, et al.
ReOPD enables efficient multi-turn on-policy distillation for LLM agents by reusing teacher trajectories as replayed prefixes, avoiding costly environment interactions.