CIFAR-10 CNN

I worked on this with a team as part of the Veritas AI program. We trained a CNN on CIFAR-10, which is a dataset of small images, and compared changes to the model and training setup.

Understanding what changed

After experimenting with voice models mostly by trial and error, I wanted to understand more deeply what was happening. This project helped connect learning rates, regularization, and data augmentation to results I could inspect.

Keeping track of the experiments was a substantial part of the work and I believe the presentation brings those comparisons together well.