AI for Science event spotlight: physics-informed losses and quantum chemistry
burny_tech · x · 2026-11-03
@burnytech reports from an AI for Science event, noting heavy use of physics-informed losses and architectural inductive biases, especially in quantum chemistry. The slide shown covers MS-ANI in MLAtom (based on ANAKIN-ME): atom-wise local descriptors ensuring translation-rotation invariance, per-atom energy contributions summed for total energy, and a loss optimizing both state energies and their gap for nonadiabatic MD with SOC. A nice fit for geometric deep learning folks.
More from Research
- Harmonic's Aristotle formally verifies a floating-point SRT divider in 39 minutes, weeks of expert work — satnam6502 · 2026-11-03
- Harmonic to demo AI theorem prover 'Aristotle' for hallucination-free hardware synthesis at REACH 2026 — satnam6502 · 2026-11-03
- Zeroing one weight column in Qwen3.5-0.8B turns broken generated chair code into a correct chair — victormustar · 2026-11-03
- AlphaFold3 guided by NMR, X-ray and cryo-EM data resolves protein ensembles — mmbronstein · 2026-11-03
- Xiaomi open-sources Robotics-1, a robot foundation model trained on 100K hours of trajectories — CyberRobooo · 2026-11-03
- Honnibal: LLM weight-editing paper abstract reads like stage magic — each result uses a different method — honnibal · 2026-11-03