Machine Learning Theory
Channel: #ml-theory
Co-leads:Â
Anier - @Anier Velasco Sotomayor:) on Discord, @aniervs on Twitter
Martina - @Martina Vilas on Discord, @martinagvilas on Twitter
Goal
In my opinion, ML Theory is all about making Machine Learning a more rigorous science, and less of a set of fancy techniques that turned out to work nicely. To give a few examples, ML Theory tries to answer questions like: why is Adam the de facto optimizer in Neural Networks? what's the underlying theory behind all modern architectures? can we predict the behaviour of a Neural Network once it's trained? can we do it without even training it?
The learning group aims to provide a domain/task/modality independent approach to studying Machine Learning theory.
The topic aims to cover different levels (introductory to advanced) of ML theory, in order to make topics in the field more accessible and also encourage a deeper exploration.
Create a space for discussions with an aim also to bridge our gap of understanding related to empirical phenomenon and theoretical predictions.
Logistics:
Bi-weekly sessions, ideally a topic per meeting, but sometimes a topic takes a bunch of meetings. The topics are decided by votes in the discord chat.
Anyone can present!
Pre-requisites in terms of reading material will be provided at least a week in advance, basic introduction will also be provided by the user.
Around 20-25 minutes talk, 5-10 minutes for questions (2x per meeting).
Discussion can take place offline on Discord after!
C4AI ML Theory Resources | Paper ListÂ
Occurrences:Â Bi-weekly, Thursdays 9a PT
Recent Presentations
May 9, 2024
Guillem Simeon - TensorNet: Cartesian Tensor Representations for Efficient Learning of Molecular Potentials
March 14, 2024
February 29, 2024
February 15, 2024
February 7, 2024
January 18, 2024
Materials from all past sessions
April 8, 2023 - Yaroslav Bulatov presentes generating functions approach to gradient descent analysisÂ
March 4, 2023 - Gradient Descent Proofs
February 18, 2023 - Privacy of Noisy SGD
February 4, 2023 - Neural Operators
January 21, 2023 - 2023 New Year Catch-Up