Automatic Differentiation

This is a small presentation I gave in lab meeting, introducing principles of automatic differentiation. Among various AD frameworks, Autograd, Autodiff, and Enzyme were benchmarked over Stillinger-Weber potential of Si atoms. Stillinger-Weber was chosen as most benchmarks and frameworks focus on traditional linear algebra kind problems, where as I am more interested on ML applications on more scientific problems. All the code would be uploaded in github soon. Presentation can be downloaded here. Sources for figures and content are given in the end.

---