Action learning: score whole trajectories with no forces or simulator, 2000× lower error
LeoTZ03 · x · 2026-10-24
Sarah Liaw, with Yilun Du and Yuanqi Du, introduces action learning — learning a stochastic action that scores how plausible an entire trajectory is, directly from observed trajectories, with no known forces, potential energy function, or simulator likelihood. Results: over 2,000× lower error than local regression recovering dynamics in a double well system; detects a hidden 0.01% perturbation in the Sun's mass from a 31-body orbital system; a single learned molecular action reuses across trajectory completion, steering, constrained optimization, and temperature transfer. Framed as a new paradigm for reasoning about trajectory data.
More from Research
- Quant hallucination forensics: fabricated prompt injections, pirate puzzles and a phantom Windows attack — Consumerbot37427 · 2026-10-24
- AgentVQA: 1,000 human-verified examples to evaluate visual agents offline across 5 domains — liuzhuang1234 · 2026-10-24
- Science study: STAT2 intron retention predicts lung cancer immunotherapy response (HR 0.49) — EricTopol · 2026-10-24
- Berkeley's Shreya Shankar: best data agent passes only 61% of queries, semantic layers help a lot — sh_reya · 2026-10-24
- New paper introduces "bottleneck pressure" to measure where work AI can't do gains value — soumitrashukla9 · 2026-10-24
- Full 510GB DeepSeek-V4.1-Flash on a 64GB RAM + RTX 3060 PC: 7.4 tok/s and 98.3% agreement — Muted_Context_1999 · 2026-10-24