Aditi Milind Joshi

Google Summer of Code 2025 Contributor

email: aditij0205@gmail.com

Education: B.Tech in Computer Science and Engineering (AIML), Manipal Institute of Technology, Manipal, India

Ongoing project: Implement and improve an efficient, layered tape with prefetching capabilities
Automatic Differentiation (AD) is a computational technique that enables efficient and precise evaluation of derivatives for functions expressed in code. Clad is a Clang-based automatic differentiation tool that transforms C++ source code to compute derivatives efficiently. A crucial component for AD in Clad is the tape, a stack-like data structure that stores intermediate values for reverse mode AD. While benchmarking, it was observed that the tape operations of the current implementation were significantly slowing down the program. This project aims to optimize and generalize the Clad tape to improve its efficiency, introduce multilayer storage, enhance thread safety, and enable CPU-GPU transfer.

Project Proposal: URL

Mentors: Aaron Jomy, David Lange, Vassil Vassilev

Presentations



Implement and improve an efficient, layered tape with prefetching capabilities, Slides, Team Meeting, 5 June 2025