Trajectory inference via Acceleration Matching
Acceleration Matching trains trajectory inference from positional snapshots without simulating trajectories during training.
The paper lifts interpolation into phase space, then regresses an explicit conditional acceleration field. That field is meant to generate random smooth trajectories matching the given time-point marginals. The authors say the method avoids both smoothness preprocessing and simulation-based training objectives. They report numerical evidence that it is competitive with or better than existing benchmark methods. ArXiv · AI/CL/LG's note
The paper lifts interpolation into phase space, then regresses an explicit conditional acceleration field. That field is meant to generate random smooth trajectories matching the given time-point marginals. The authors say the method avoids both smoothness preprocessing and simulation-based training objectives. They report numerical evidence that it is competitive with or better than existing benchmark methods. ArXiv · AI/CL/LG's note
score 4