TopoAct - Visually Exploring the Shape of Activations in Deep Learning

TopoAct - Visually Exploring the Shape of Activations in Deep Learning

Explore the shape of activations in deep learning through a comprehensive lecture on TopoAct, a visual exploration system. Delve into the organizational principles behind neuron activations in deep neural networks like GoogLeNet and ResNet. Discover how topological data analysis tools are applied to study topological summaries of activation vectors within and across layers. Gain valuable insights into learned representations of image classifiers through visual exploration scenarios. Examine the joint work of Bei Wang, Archit Rathore, Nithin Chalapathi, and Sourabh Palande, which investigates the space of activations and their relationships within and across network layers.