URAP

Shiladitya Dutta

Stability Driven Interpretation of Low-rank Compression Methodologies for CNNs

Deep neural networks achieve state-of-the-art performance in many tasks such as computer vision and natural language processing, however many compression/acceleration methods for these models yield unstable or unreliable results. As such, I am using statistical methodologies to investigate the properties of low-rank compression methodologies, with a particular focus on SVD-based compression. In particular, I am focusing on how SVD compression can be used in compression with various filter-level visualization methods (i.e. saliency based methods) in order to reveal the internal “decision-making” structure of particular sections of these highly complex models. This will hopefully yield insights into how these compression methods could be used for the visualizations and interpretation of Neural Networks, something that is particularly relevant since utilizing “Black Box” algorithms poses a significant ethical dilemma in fields like medicine.

Message to Sponsor

I would like to extend my sincerest thanks to my URAP donor. It truly cannot be overstated how much this opportunity would allow me to deepen my experience in the field of computer science research, both in terms of general exposure to academia and specific learning in regards to the field of AI. URAP as a program is equipping me, along with so many others, with the tools and knowledge to continue their academic passion through the arena of research, hopefully extending into a fruitful academic or industry career. With the work I do here, I am setting the experiential foundations for any future work I do within the fields of AI and even Computer Science in general, something whose value is truly immense. Beyond myself, over the course of the coming Summer months I aim to formulate my research into a publication so that the fruits of my experience here can be shared with the rest of the world. Once again, I'd like to thank you for supporting research at UC Berkeley/UCSF and for providing me with this truly unique educational opportunity.
  • Major: Electrical Engineering & Computer Science (EECS) and Business Administration
  • Mentor: Reza Abbasi-Asl, UCSF