Want to Understand Neural Networks? Think Elastic Origami!

This article summarizes a podcast interview with Professor Randall Balestriero, focusing on the geometric interpretations of neural networks. The discussion covers key concepts like neural network geometry, spline theory, and the 'grokking' phenomenon related to adversarial robustness. It also touches upon the application of geometric analysis to Large Language Models (LLMs) for toxicity detection and the relationship between intrinsic dimensionality and model control in RLHF. The interview promises to provide insights into the inner workings of deep learning models and their behavior.

• •Exploration of neural network geometry and its connection to spline theory.
• •Discussion of 'grokking' and adversarial robustness in deep learning.
• •Application of geometric analysis to LLMs for toxicity detection and RLHF.