IS05 - Artificial Intelligence For Scientific Discovery
Multi-scale coupled physics applications require computational modeling and
simulations at large-scale to attain high precision and accuracy. In situations where the
quantities of interest are defined in a high dimensional space, the curse of dimensionality
pushes the limits of existing high-performance computing (HPC) architectures due to
expensive, and sometimes unaffordable, computational requirements.
To alleviate the computational burden of state-of-the-art approaches and facilitate the
bridging between multiple scales, artificial intelligence (AI) has been shown to provide
significant benefits in several applications ranging from atomic modeling, computational
mechanics, computational fluid dynamics, control, topological optimization, drug
discovery, and material design.
The benefits of AI for scientific discovery are twofold. On the one hand, deep learning
(DL) models accelerate the resolution of forward problem defined in a high dimensional
space, where the goal is to predict target properties from a given set of input features. On
the other hand, reinforcement learning (RL) and generative models (GM) accelerate the
resolution of inverse problems defined in high dimensional spaces, where the goal is to
identify the input features that correspond to a desired, prescribed target property.
In this session, we show how AI can be effectively used to accelerate the resolution of
forward and inverse problems in several computational sciences and engineering
applications and breakthrough existing computational barriers.