MinkowskiPE Encodes Spacetime Displacement for Attention, Cuts Video Prediction MSE 9.9%
Yuhao Li · hf · 2026-10-06
Minkowski Positional Encoding (MinkowskiPE) parameterizes Lorentz transformations on query/key features with joint spatiotemporal coordinates, making attention scores depend only on relative spacetime displacement between tokens — invariant to global translation while keeping the standard dot-product interface.
- Evaluated across microscopic molecular dynamics and macroscopic video prediction
- Best results on all nine multi-trajectory molecular evaluations
- Reduces KTH video-prediction MSE by 9.9% vs the best baseline with roughly one-tenth the parameters
More from Research
- Apple research team opens 2027 PhD internships in video models, 3D/4D reconstruction — HildeKuehne · 2026-10-06
- MemAdapter uses counterfactual reasoning to curb memory-induced sycophancy in LLM agents — Ruqing Ning · 2026-10-06
- Peking University's Code2Games gets coding agents to build playable UE5 game worlds — PekingUniversity · 2026-10-06
- QuantCode: domain pretraining + SFT lifts Qwen trading-code pass from 27.8% to 58.2% — Alexey Chernysh · 2026-10-06
- Subsampling and extrapolation keep the Mandelbrot area estimate unbiased near the boundary — geoffreyirving · 2026-10-06
- Group-Evolving Agents: a new paradigm where the unit of agent self-improvement is a group — xwang_lk · 2026-10-06