ASPIRE: Saddle-Point Discovery through Set Prediction and Physical Refinement

Name: ASPIRE: Saddle-Point Discovery through Set Prediction and Physical Refinement
Uploaded: 2026-10-15T06:55:32.311119+00:00
Duration: 1352 s
Description: ASPIRE predicts multiple atomic saddle points from a single starting configuration using an equivariant set predictor called Ev-Quiformer. These candidates are then refined via Dimer searches on the original potential. The framework achieves 77.20% reference-event coverage, significantly outperforming baselines while requiring only half the Dimer force evaluations.

ASPIRE predicts multiple atomic saddle points from a single starting configuration using an equivariant set predictor called Ev-Quiformer. These candidates are then refined via Dimer searches on the original potential. The framework achieves 77.20% reference-event coverage, significantly outperforming baselines while requiring only half the Dimer force evaluations.

Open the full episode