Memristive Neural Networks - PhD Dissertation Defense

Zoom recording from my PhD Dissertation Defense held on January 30, 2025 at SEH 2000, Science and Engineering Hall, The George Washington University.
The title of my dissertation was "Memristive Neural Networks: Modeling, Prototyping, and Hardware-Software Co-Design". I worked on robust algorithms for accurately performing neural network training and inference operations on noisy hardware accelerators. The work is verified on a variety of network architectures, ranging from small multi-layer perceptron networks all the way to transformer-based large language models (LLMs). A key contribution is the development of a complete deep learning library that extends PyTorch and implements various noise sources, fault-tolerance algorithms, and low-rank training algorithms for efficiently training memristive neural networks under resource limitations, system non-idealities, and other hardware constraints.
Unfortunately, this recording starts after the introduction. You can view the full slide deck here: drive.google.c...

[@MrJoeboo123]

Quality content! Excellent work Dr. Yousuf 🤩