2026 09 27

(Sep 2026) 2 papers are accepted to NeurIPS 2026: "MILES: Modular Instruction Memory with Learnable Selection for Self-Improving LLM Reasoning " (self-improving LLM with modular memory of re-usable sub-skills/instructions + learned reading); "Towards Identifiable Latent Additive Noise Models" (causal representation learning). Congratulations to Ruilin and Yuhang!