About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI-Enabled Materials Processing: Integrating Accelerated Experimental Workflows and Processing-Aware Machine Learning
|
| Presentation Title |
Uncertainty-Propagating Active Learning Accelerates High-Entropy Oxide Discovery via Pulsed Electrodeposition for Hydrogen Evolution |
| Author(s) |
Shakti Prasad Padhy, Neha Mishra, Juan Florez-Coronel, Subhajit Das, Ankur Srivastava, Doğuhan Sarıtürk, Suhash Ranjan Dey, Douglas Allaire, Raymundo Arróyave |
| On-Site Speaker (Planned) |
Shakti Prasad Padhy |
| Abstract Scope |
Discovering high-entropy oxide (HEO) electrocatalysts for the hydrogen evolution reaction demands navigating a coupled design space where pulsed electrodeposition process parameters determine composition stochastically, and composition governs catalytic performance. Standard active learning approaches that treat composition as a direct input discard this coupling uncertainty, leading to overconfident predictions in sparse data regimes. We present a closed-loop Bayesian optimization framework for FeCoNiCuZn oxides synthesized via pulsed electrodeposition, operating natively in process space and propagating compositional uncertainty to overpotential predictions via a chained two-stage surrogate: a multi-output Gaussian process maps deposition parameters to composition, and a noisy-input Gaussian process (NIGP) ingests the resulting variance as structured input noise. A batch Bayesian optimization strategy with snap-and-refine proposes candidates on the discrete parameter grid. Initialized on 87 samples, the pipeline targets overpotential improvements with calibrated uncertainty. Best-performing candidates investigated through XRD, XPS, and TEM inform computational prediction of local structural motifs and activity descriptors. |
| Proceedings Inclusion? |
Planned: |