Safe and Trustworthy Machine Learning

Bhavya Kailkhura | 20-ERD-014

Executive Summary

Through this project we will introduce the notion of certified safety in machine learning systems by developing models with guaranteed robustness and designing a suite of statistical methods to reliably examine and debug trained models. By making a fundamental advance in the field of machine learning, this research will have far-reaching impact across many national security applications, which rely increasingly on artificial intelligence.

Publications, Presentations, and Patents

Bulusu, S., et al. 2020. "Anomalous Example Detection in Deep Learning: A Survey." IEEE Access 8: 13233–132347. doi:10.1109/ACCESS.2020.3010274. LLNL-JRNL-808677

Pan, B., et al. 2020. "Adversarial Mutual Information for Text Generation." International Conference on Machine Learning (online), July 2020. LLNL-CONF-805519

Xu, K., et al. 2020a. "Automatic Perturbation Analysis on General Computational Graphs." International Conference on Machine Learning (online), July 2020. LLNL-CONF-805206

——— 2020b. "Towards an Efficient and General Framework of Robust Training for Graph Neural Networks." IEEE International Conference on Acoustics, Speech, and Signal Processing, Barcelona, Spain, May 2020. LLNL-CONF-794857