Manifold

Model-agnostic ML visualization tool


Manifold is a tool for visualizing machine learning model predictions, understanding their performance, and comparing them. It is independent of the specifics of the model and only cares about inputs and outputs, which makes it widely applicable.

I worked on a prototype of Manifold that was described in a paper published in the IEEE Transactions on Visualization and Computer Graphics in 2019.

Yang Wang gave a presentation at IEEE VIS 2018 in Berlin. Lezhi Li and Yang Wang then developed the prototype further into a fully fledged model debugging tool described in a blogpost, and eventually open-sourced it, making the code available on GitHub.

The tool received a significant amount of press coverage:


Collaborators

Lezhi Li

Yang Wang

Jiawei Zhang