MIT team proposes attention-based foundation model for quantum states, mapping t-V phase diagram from 18 parameters
jwt0625 · x · 2026-10-18
An arXiv paper by Timothy Zaklama, Daniele Guerci and Liang Fu introduces an attention-based foundation model architecture that learns and predicts quantum states across Hamiltonian parameters, system sizes and physical systems.
- Inputs are just basis configurations and physical parameters; the network outputs highly accurate ground-state wavefunctions
- It builds the full phase diagram of the 2D square-lattice t-V model from only 18 parameters (V/t, N)
- Key insight: physics-informed training-data selection is crucial for generalization, and success relies on a smooth mapping from parameter space to wavefunction space
- The authors frame it as a basis for a universal foundation model for quantum matter
Related event: MIT Proposes Attention-Based Foundation Model for Quantum States(2 posts)→
More from Research
- Tohoku U and Harvard Yenching host free AI digital humanities workshop, Dec 2026 — SebastianNehrd2 · 2026-10-18
- How voice AI hears: from 16,000 samples per second to spectrograms — prdeepakbabu · 2026-10-18
- IEEE STEM Summit 2026 talk: how do voice assistants hear us? — prdeepakbabu · 2026-10-18
- CoRL 2026 debuts late-breaking results track on agentic robotics and Real2Sim2Real — GeorgiaChal · 2026-10-18
- Obligatory Contour Principle offers statistical evidence Linear A maps to Linear B values — PontiEdoardo · 2026-10-18
- Gemini 4 Argon takes top spot on Edison Sci's long-horizon bioinformatics benchmark BixBench3 — divy93t · 2026-10-18