AI Surrogates for Enhancing Traditional HPC Applications

Topic: AI Surrogates as Adjuncts to Traditional HPC Simulations

Join us for a session on how AI surrogates can enhance traditional HPC applications, presented by Alessandro Fanfarillo, Senior Member of Technical Staff at AMD – Frontier Center of Excellence. This event will showcase practical examples of using AI surrogates as complementary tools to established HPC simulation workflows. While the session is open to all, it will be particularly valuable for researchers looking to integrate AI into their HPC applications to improve performance and accuracy.

This event will count toward a review score bonus point for research teams planning to apply for the April Frontier AI Hackathon.

What You’ll Learn:

1. Generating Initial Condition Data with Surrogates:
   • Alessandro will demonstrate how a Generative Variational Autoencoder (CVAE), trained on real data, can create temperature distributions for a climate model.
   • Benefit: Significant memory savings by replacing large, pre-stored databases with a compact surrogate capable of dynamically generating the required data.
2. Replacing Fixed Parametrization with AI Estimates:
   • See how a Deep Neural Network (DNN), trained on real-world data, can replace fixed parameterizations in HPC models.
   • Example: Predicting snow density with greater accuracy compared to traditional parameterizations, leading to results that align more closely with observed natural phenomena.
3. Learn Practical Strategies for Incorporating AI Surrogates into HPC Workflows:
   • Explore readable Python implementations of both examples, which you can adapt for your own work.
   • Discover how AI surrogates can improve efficiency and accuracy in computational simulations.

There will be dedicated time for discussion, and attendees are encouraged to bring descriptions and questions of their codes that could benefit from AI modeling or augmentation.

Don’t miss this opportunity to deepen your understanding of AI’s role in HPC and gain hands-on insights into integrating AI surrogates into your research.