| The Advanced Photon Source (APS) at Argonne organizes an ongoing seminar series, the APS Scientific Computation Seminar Series, focused on various aspects of computation for synchrotron science; talks are related to data analysis, reconstruction, theory, simulation, optimization, machine learning, automation, and visualization. The seminar provides an opportunity to present and learn about state-of-the-art software and the application of computational, AI/ML and mathematical techniques to synchrotron science. |
| Title: | HPC-driven autonomous experiments in action | | Presenter(s): | Ayana Ghosh, Research Scientist, Oak Ridge National Laboratory, Tennessee
Martin Foltin, Senior Principal Researcher, Hewlett Packard Labs, Colorado
Gayathri Saranathan, AI Researcher, Hewlett Packard Labs, Singapore | | Date: | February 17, 2025 | | Time: | 1:00 PM (Central Daylight Savings Time) | | Location: | Join ZoomGov Meeting
https://argonne.zoomgov.com/j/1601444470?pwd=N1phbHZVdCtmcVR5cGh0c1Zhc0…
Meeting ID: 160 144 4470
Passcode: 937918
One tap mobile
+16692545252,,1601444470# US (San Jose) +16468287666,,1601444470# US (New York) Dial by your location
+1 669 254 5252 US (San Jose)
+1 646 828 7666 US (New York)
+1 646 964 1167 US (US Spanish Line) +1 669 216 1590 US (San Jose)
+1 415 449 4000 US (US Spanish Line) +1 551 285 1373 US
Meeting ID: 160 144 4470
Find your local number: https://argonne.zoomgov.com/u/af2crdvQy
| | Abstract: | Recent advancements in algorithms and electron microscopy offer the potential to integrate theoretical models with experiments for solving material science challenges. AI methods excel in extracting atomic features from images, predict physical properties, while being useful to find regions of interest to perform next set of measurements. However, challenges remain in creating an autonomous instrument-computing system, particularly around deployment, novel physics exploration, while refining experimental and theoretical parameters. Issues include instrument specificity, implementation complexity, managing the different latencies of imaging with simulations. This presentation will focus on the development of a multi-surrogate framework on two-dimensional materials that combines deep kernel learning, tree-based models, with Gaussian Mixture Models (GMM) for material property prediction, alongside an autoencoder-decoder for structural reconstruction. In addition, we shall demonstrate how a common metadata framework (CMF) can provide improved model performance via continuous training with dynamic parameter adjustments. The framework is aimed at bringing us closer to time-to-solution by advancing autonomous laboratories through the integration of computational insights with real-time experiments. |
|
