Livermore researchers are participating in the 39th annual Conference on Neural Information Processing Systems (NeurIPS) on December 2–7 in San Diego, California. This high-impact event focuses on machine learning and artificial intelligence. Follow @Livermore_Comp on X with the #NeurIPS2025 hashtag. Links below point to session descriptions and poster abstracts on the NeurIPS website.

Workshops

• Closing the Omics Gap: A Benchmark for Unified Modeling for Biomolecular Foundation Models – Joseph Wakim, Vinayak Gupta, Jose Marti, Jonathan Allen, Brian Bartoldson, Bhavya Kailkhura
• Extrapolating Phase-Field Simulations in Space and Time with Purely Convolutional Architectures – Nathan Bieberdorf
• Fourier–Thermodynamic Latent Modeling for Temperature-Dependent Plasma Mixing – Jannik Eisenlohr, Youngsoo Choi
• Generation-Based Multi-Modal Anomaly Detection for Nuclear Fusion Target Polishing –Kshitij Bhardwaj, Sean Hayes, Monika Biener, Suhas Bhandarkar
• Generative Latent Space Dynamics of Electron Density – Youngsoo Choi, Daniel Osei-Kuffuor
• Leveraging Large Language Models to Enhance Machine-Learning-Driven HPC Job Scheduling – Kshitij Bhardwaj, Edgar A. Leon
• Offline Maximizing Minimally Invasive Proper Orthogonal Decomposition for Reduced Order Modeling of Radiation Transport – Jean Ragusa, Youngsoo Choi
• Rollout-LaSDI: Enhancing the Long-Term Accuracy of Latent Space Dynamics – Robert Stephany, Youngsoo Choi
• Sequential Decoder Training for Improved Latent Space Dynamics Identification – William Anderson, Youngsoo Choi, Seung Whan Chung
• Surrogate Modeling for the Design of Optimal Lattice Structures using Tensor Completion – Aldair Gongora, Brian Giera
• Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence – Brian Bartoldson, Bhavya Kailkhura
• Uncertainty Quantification for Reduced-Order Surrogate Models Applied to Cloud Microphysics – Jonas Katona, Emily de Jong, Nipun Gunawardena

Posters

• Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts – Brian Bartoldson, Bhavya Kailkhura
• BOOM: Benchmarking Out-Of-Distribution Molecular Property Predictions of Machine Learning Models – Evan Antoniuk, Tal Ben-Nun, Peggy Li, James Diffenderfer, Tim Hsu, Anna Hiszpanski, Bhavya Kailkhura, Brian Van Essen
• The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text – Brian Bartoldson, Bhavya Kailkhura
• Constrained Discrete Diffusion – Bhavya Kailkhura
• Convergence Rates of Constrained Expected Improvement – Jingyi Wang, Nai-Yuan Chiang, Cosmin Petra
• Polar Sparsity: High Throughput Batched LLM Inferencing with Scalable Contextual Sparsity – Nikoli Dryden
• Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach – Neel Jain, Brian Bartoldson, Bhavya Kailkhura
• Trajectory Balance with Asynchrony: Decoupling Exploration and Learning for Fast, Scalable LLM Post-Training – Brian Bartoldson, James Diffenderfer, Tal Ben-Nun, Bhavya Kailkhura