Skip to content

The aim of this study is to conduct topological analysis on a FCNN trained using the MNIST dataset. Various clustering techniques were applied to identify the network's critical neurons, and the results were analyzed by examining their corresponding weights.

Notifications You must be signed in to change notification settings

alessimichele/Topological-Analysis-of-Neural-Networks

Repository files navigation

Topological-Analysis-of-Neural-Networks

Final project for Advanced Topics in Machine Learning course.

The project's objective is to investigate topological properties, and potentially, topological invariants, related to both the weights and activations within a neural network.

This is achieved through a two-step process: first, clustering the final network activations and then examining the neurons that exhibit the highest activation within each cluster.

Subsequently, after identifying these "most significant neurons" in the network, a detailed analysis is conducted, focusing on their respective weights. Here you can find the outline of the project.

About

The aim of this study is to conduct topological analysis on a FCNN trained using the MNIST dataset. Various clustering techniques were applied to identify the network's critical neurons, and the results were analyzed by examining their corresponding weights.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published