AI for science & scientific reasoning
Key Questions
What is SciReasoner and what benchmarks has it achieved?
SciReasoner is a multimodal foundation model focused on scientific reasoning. It has reached state-of-the-art performance on 67 out of 86 scientific benchmarks.
How is AI applied at CERN for future discoveries?
CERN is using AI for real-time data filtering and to support the design of future colliders. This approach is expected to accelerate scientific breakthroughs in particle physics.
What role do world models play in physical simulation research?
World models are being developed by organizations such as DeepMind, Runway, World Labs, and AMI to simulate physical environments. They show promise for advancing simulation capabilities while also facing notable limits.
How does the Aurora supercomputer assist with chemistry simulations?
Aurora powers the ChemGraph autonomous platform, which handles complex chemistry simulations. This reduces reliance on specialized expertise and multiple software tools while shortening workflow times.
What contributions did CSE researchers make to ICML 2026?
CSE researchers authored eleven papers presented at ICML 2026. The work covers a range of topics in AI, machine learning, and scientific applications.
Developing: SciReasoner multimodal foundation model achieves SOTA on 67/86 scientific benchmarks. AI model predicts drug release rates using PINNs (94% reduction in experimental time). TESSERA v2 scaling Earth foundation models. OmniTacTune tactile adaptation. ChemGraph on Aurora supercomputer. Eleven papers from CSE researchers at ICML 2026. CERN using AI for real-time data filtering and future collider design. World models for physical simulation (DeepMind, Runway, World Labs, AMI). SPEAR simulator for photorealistic embodied AI research (10x faster rendering, 14K UE functions). GPT-5.6 Sol Pro used to resolve open statistics question on FDR control.