> For the complete documentation index, see [llms.txt](https://edward-stevinson.gitbook.io/tech-notes/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://edward-stevinson.gitbook.io/tech-notes/papers/other/badger.md).

# Badger Architecture

## Badger Architecture

**Rosa et al. 2019**

An agent is made up of many experts which all share the same communication policy (expert policy) but have different internal memory states.

There are two levels of learning:

1. An *inner loop* in which an agent's behaviour emerges as a result of experts communicating with each other.
2. An *outer loop* in which an expert policy is discovered over generations of agents.

The roles of experts and connectivity among them is assigned dynamically at inference time.

Instead of having a handcrafted expert policy they use a "multi-agent meta-learning" approach
