Topic: ML Theory

LLNL researchers have posters and workshop papers accepted to the 13th International Conference on Learning Representations on April 24–28.

The February 28 event brought together over 1,400 Department of Energy scientists across multiple sites to explore how cutting-edge AI models could transform scientific research.

Highlights include ML techniques for computed tomography, a scalable Gaussian process framework, safe and trustworthy AI, and autonomous multiscale simulations.

Drawing more than 300 attendees, this year’s “D3” workshop focused on tackling pressing data challenges in nuclear security, energy and collaborative scientific discovery, and featured a host of talks, presentations and panels.

A CASC researcher and collaborators study model failure and resilience in a paper accepted to the 2024 International Conference on Machine Learning.

LLNL researchers study model robustness in a paper accepted to the 2024 International Conference on Machine Learning.

In two papers from the 2024 International Conference on Machine Learning, Livermore researchers investigate how LLMs perform under measurable scrutiny.

MuyGPs helps complete and forecast the brightness data of objects viewed by Earth-based telescopes.

Using explainable artificial intelligence techniques can help increase the reach of machine learning applications in materials science, making the process of designing new materials much more efficient.

New research debuting at ICLR 2021 demonstrates a learning-by-compressing approach to deep learning that outperforms traditional methods without sacrificing accuracy.

Three papers address feature importance estimation under distribution shifts, attribute-guided adversarial training, and uncertainty matching in graph neural networks.

Highlights include perspectives on machine learning and artificial intelligence in science, data driven models, autonomous vehicle operations, and the OpenMP standard 5.0.