Accelerating Moment Tensor Potentials through Post-Training Pruning

TL;DR

This paper presents a post-training pruning method for Moment Tensor Potentials that reduces computational cost significantly without sacrificing accuracy, making MTPs more efficient for practical use.

Contribution

The authors introduce a cost-aware pruning strategy for MTPs that is data-agnostic and compatible with existing implementations, improving efficiency without additional data.

Findings

Models up to seven times faster after pruning

No additional data needed for pruning process

Maintains accuracy while reducing computational cost

Abstract

Moment Tensor Potentials (MTPs) are machine-learning interatomic potentials whose basis functions are typically selected using a level-based scheme that is data-agnostic. We introduce a post-training, cost-aware pruning strategy that removes expensive basis functions with minimal loss of accuracy. Applied to nickel and silicon-oxygen systems, it yields models up to seven times faster than standard MTPs. The method requires no new data and remains fully compatible with current MTP implementations.