Akshay Trikha

I recently negotiated my first contract, and had no idea what I was doing.

TL;DR: We tested whether a plain transformer encoder (no equivariance, no periodic boundary conditions) can learn interatomic forces as data & params scale vs.

Please try out the extension on the VSCode extension marketplace and let me know what you think!

Background A neural network learns a function to map inputs to outputs.

Fig 1.

Recall when beginning to train a neural network its weights might be initialized randomly.

In early 2019 OpenAI published some research showing that large language models, when trained on colossal amounts of text, begin to behave unpredictably intelligently on tasks they weren’t originally trained on.

Overview I wanted to learn how to build and deploy an ML system in the real world and thought this would be a fun place to start.