Browse by topic
- LLM (9)
- AI Agents (9)
- Domain LLMs (1)
- Math (6)
- Analysis (1)
- Algebra & Foundations (0)
- Geometry & Topology (1)
- Probability & Finance (5)
- Physics (5)
- Reading (4)
- Books (0)
- Papers (4)
- Robotics & Embodied AI (2)
- Reinforcement Learning (2)
- Conferences (1)
- ICML 2026 (1)
-
Chain-of-Goals: latent chain-of-thought for long-horizon offline RL
A review of CoGHP (arXiv 2602.03389), read as a case study in *internal* chain-of-thought: latent subgoals as reasoning steps inside a single autoregressive control policy — the big problem, the method, the honest numbers, and where it points.
-
VLA models through the lens of adaptation
A survey-first deep dive into vision-language-action models, organized around the one axis that actually measures progress — adaptation across bodies, tasks, action spaces, and distributions — with Qwen-VLA as the worked case.
-
Qwen-VLA — one model for manipulation, navigation, and trajectory
Deep reading note on Qwen-VLA (arXiv 2605.30280): a Qwen3.5-4B VLM backbone + a 1.15B DiT flow-matching action decoder, unified across tasks, environments, and robot embodiments via embodiment-aware prompt conditioning.
-
ICML 2026 — live archive
A running, super-detailed archive of ICML 2026 — day by day, plus workshops, tutorials, and standout papers. Updated live through the conference.
-
One vanishing integral, five lenses: why ∫₀²π (1+2cos t)/(5+4cos t) dt = 0
A single definite integral that evaluates to exactly zero, proved five ways — Weierstrass substitution, residues, cyclotomic/Dirichlet-kernel number theory, winding numbers and de Rham cohomology, and the Green–Stokes generalization. Each lens is written as a settled core plus open threads to expand in a later edit.