AI for Discovery in the Sciences
ML curriculum and tutorial notebooks for PhD students and industry professionals.
As an undergraduate research assistant for USC’s AI for Discovery in the Sciences workshop, I co-developed and taught machine-learning tutorial notebooks for PhD students and industry professionals who had science research experience but less ML background.
The notebooks covered generative diffusion models and nonlinear phase-transition detection through the Learning by Confusion method. I reviewed relevant AI-for-science literature, translated methods into hands-on Colab exercises, and helped make the material accessible to researchers who wanted to use ML without treating it as a black box.
For the diffusion notebook, I implemented a custom UNet2DModel and pipeline, explored noise schedulers, and trained classifiers to detect generated samples. For the Learning by Confusion notebook, I created and augmented a sunrise image dataset from a GoPro time-lapse to demonstrate how the method can detect phase-transition-like structure.
The workshop is not an AI safety project directly, but it remains important to how I think about research. Teaching forced me to diagnose what someone understands, design probes that reveal confusion, and build interventions that actually help. That loop is close to how I now think about evaluations and interpretability.