Blog

On some elementary properties of Hopfield networks

Matrices

In this post we derive some basic mathematical properties of asynchronous Hopfield networks from first principles.

A simple neural network with a benign loss landscape

Neural Networks

Optimization

We look at an example of a simple network whose loss landscape, albeit non-convex, is relatively benign from an optimization perspective. This post is based on Theorem 2.1 of …

Global optimization guarantees for shallow neural networks via the Perceptron algorithm

Neural Networks

Optimization

In this post we will…

Optimization guarantees for neural networks via the NTK

Neural Networks

Optimization

In this post we will look at how the Neural Tangent Kernel (NTK) can be used to derive training guarantees for sufficiently overparameterized neural networks.

Concentration of Gibbs measures onto minimizers of objective functions

Probability

Under what conditions do Gibbs measures concentrate on the minimizers of an associated objective function in the inverse temperature limit?