Surrogate Neural Architecture Codesign Package (SNAC-Pack)
Published in NeurIPS 2025 Machine Learning and the Physical Sciences Workshop, 2025
SNAC-Pack is an integrated framework that automates the discovery and optimization of neural networks with a focus on FPGA deployment. By leveraging surrogate models, it eliminates the need for time-intensive synthesis during the neural architecture search process, enabling rapid exploration of the design space while maintaining accurate predictions of hardware performance metrics.