PhD student, Massachusetts Institute of Technology
2 papers at NeurIPS 2025
We introduce a bicycle design benchmark that evaluates multiphysics performance, constraint satisfaction, and adherence to human preferences, and we benchmark LLMs, tabular generative models, and design optimization algorithms side by side.
OAT introduces the first step towards foundation models for topology optimization by combining a neural field auto-encoder and latent diffusion models with large scale training on a new general dataset of 2M optimized topologies called OpenTO.
