Boosting Axon's Training Capabilities Toward State of the Art Image Models

[ElixirConf]

Boosting Axon’s Training Capabilities Toward State of the Art Image Models

Boosting Axon's Training Capabilities Toward State of the Art Image Models

ElixirConf US 2024 - Scott Mueller

Welcome to ElixirConf

Publicly shared Axon/Bumblebee image training examples don’t produce models that are as accurate as some training approaches in other languages. We will demonstrate new Nx related libraries and techniques that can boost your Elixir trained model’s capabilities. Using our solution to the elixirfashionmlchallenge, #ElixirFashionMLChallenge General Discussion/Questions, we’ll teach the key training techniques. Then with just a few images, say 400, we’ll create our own custom model. After the presentation, you can use the Livebook notebook for producing your own custom model.

With robust fundamentals demonstrated, we’ll then explore techniques that prepare us for advanced training of language models.

Comments welcome! View the elixirconf tag for more ElixirConf talks!