1 min to read

Mapping inference to physics - Part 2

Update on a work in progress

Featured image

A couple months ago I shared a post about a project we are working on in the lab. I got a lot of great feedback and I incorporated it into this new version of the paper.

The idea of this project is to derive the physics of inference, in other words we want to try to derive equations for artificial general intelligence (AGI) models from first principles. The cool part about this project is that we are showing that model parameters of such a general inference algorithm can be mapped to physical quantities in such a way that when a model is maximially predictive, its parameters evolve with dynamics similar to a system of charged particles. The hope is that modeling models in this way may help us to build more efficient and robust models.

You can check out the latest version of the manuscript here: Information Physics. Keep in mind that this is still a work in progress, so there may be typos, errors, and I clearly havent put in enough citations yet. But I hope that most readers will be able to ignore these and focus on the big picture. Please feel free to leave comments and suggestions, your feedback is greatly appreciated!