Summer Undergrad Award
Image
Fine-tuning– Pretrained a VAE on 2D unit cell dataset to guide a conditional GAN, aiming to stabilize training and improve acoustic design quality – Used weights from a previously trained GAN model; reduced training time by 40%, and increased viable unit cell yield from 20% to 65% – Diagnosed and fixed discriminator overpowering in WGAN-GP by tuning hyper-parameters ; cut convergence time by 40 – Automated manufacturability scoring for over 2800 AI-designed unit cells via Python and PrusaSlicer’s CLI; cut manual effort by 95% – Performed COMSOL-based acoustic simulations on selected 2D unit cells to evaluate performance across target frequency ranges
2025 - 2025