# Learning atomic forces from uncertainty-calibrated adversarial attacks

arXiv: https://arxiv.org/abs/2502.18314 · HTML: https://mlsanity.com/paper/2502.18314/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 6 of its 46 references that are in the map.
- First posted on arXiv 25 Feb 2025.

## Builds on (most cited first, 6 of 6)

- [E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials](https://mlsanity.com/paper/2101.03164/) (2021)
- [Learning Local Equivariant Representations for Large-Scale Atomistic Dynamics](https://mlsanity.com/paper/2204.05249/) (2022)
- [Forces are not Enough: Benchmark and Critical Evaluation for Machine Learning Force Fields with Molecular Simulations](https://mlsanity.com/paper/2210.07237/) (2022)
- [Less is more: sampling chemical space with active learning](https://mlsanity.com/paper/1801.09319/) (2018)
- [Differentiable sampling of molecular geometries with uncertainty-based adversarial attacks](https://mlsanity.com/paper/2101.11588/) (2021)
- [Uncertainty-biased molecular dynamics for learning uniformly accurate interatomic potentials](https://mlsanity.com/paper/2312.01416/) (2023)

## Built on by (most cited first, 2 of 2)

- [Proof-Carrying Materials: Falsifiable Safety Certificates for Machine-Learned Interatomic Potentials](https://mlsanity.com/paper/2603.12183/) (2026)
- [MLIP Detective: Active Failure Mode Discovery Beyond Benchmark Scores for Machine-Learning Interatomic Potentials](https://mlsanity.com/paper/2609.08399/) (2026)

## Abstract (arXiv)

Adversarial approaches, which intentionally challenge machine learning models by generating difficult examples, are increasingly being adopted to improve machine learning interatomic potentials (MLIPs). While already providing great practical value, little is known about the actual prediction errors of MLIPs on adversarial structures and whether these errors can be controlled. We propose the Calibrated Adversarial Geometry Optimization (CAGO) algorithm to discover adversarial structures with user-assigned errors. Through uncertainty calibration, the estimated uncertainty of MLIPs is unified with real errors. By performing geometry optimization for calibrated uncertainty, we reach adversarial structures with the user-assigned target MLIP prediction error. Integrating with active learning pipelines, we benchmark CAGO, demonstrating stable MLIPs that systematically converge structural, dynamical, and thermodynamical properties for liquid water and water adsorption in a metal-organic framework within only hundreds of training structures, where previously many thousands were typically required.