# 🧠 Federated inference and belief sharing

**Karl J. Friston, Thomas Parr, Conor Heins, Axel Constant, Daniel Friedman, Takuya Isomura, Chris Fields, Tim Verbelen, Maxwell Ramstead, John Clippinger, Christopher D. Frith** (2024) · *Neuroscience & Biobehavioral Reviews*

[![DOI](https://img.shields.io/badge/DOI-10.1016%2Fj.neubiorev.2023.105500-blue)](https://doi.org/10.1016/j.neubiorev.2023.105500)

---
<!-- Schema.org structured data for search engines -->
<!--
{"@context":"https://schema.org","@type":"ScholarlyArticle","headline":"FederatedInference","abstract":"This paper formulates federated inference and belief sharing as a principled approach to distributed intelligence. By extending Active Inference to multi-agent settings, agents maintain local generative models while sharing beliefs through message passing to achieve collective inference. The framework addresses how agents can coordinate without sharing raw observations, preserving privacy while enabling collective cognition. This speaks to multi-scale inference architectures and distributed deci","keywords":["federated inference","belief sharing","Active Inference","distributed intelligence","multi-agent systems","message passing","collective cognition","privacy-preserving inference"],"author":{"@type":"Person","name":"Daniel Ari Friedman","url":"https://danielarifriedman.com/"}}
-->


## Abstract

> This paper explores federated inference and belief-sharing among agents in a shared environment, demonstrating how communication and language can emerge from minimizing free energy. Through simulations, it illustrates the processes of inference, learning, and model selection that facilitate collective intelligence and the development of language.

## Keywords

`federated inference` · `belief sharing` · `Active Inference` · `distributed intelligence` · `multi-agent systems` · `message passing` · `collective cognition` · `privacy-preserving inference`

## Key Contributions

- Introduces the concept of federated inference through belief-sharing among agents.
- Demonstrates the emergence of language as a byproduct of free energy minimization.
- Provides a framework for understanding collective intelligence in ensembles of agents.
- Explores the implications of belief-sharing for cultural niche construction and complexity in self-organizing systems.

## Methods

- Utilizes numerical simulations to model belief-sharing and active inference among synthetic agents.
- Applies the free energy principle to describe the dynamics of belief updating and communication.
- Examines the acquisition of language through active learning and the transmission of beliefs across generations.
- Employs Bayesian model selection to illustrate structure learning in generative models.

## 🎯 Consulting & Tutoring

**Available for AI Research Consulting and Tutoring.** [Contact Daniel Ari Friedman, PhD](https://danielarifriedman.com/) for collaboration on Active Inference, Bayesian modeling, and computational biology.

## Citation

```bibtex
@article{2024_FederatedInference,
  author = {Karl J. Friston, Thomas Parr, Conor Heins, Axel Constant, Daniel Friedman, Takuya Isomura, Chris Fields, Tim Verbelen, Maxwell Ramstead, John Clippinger, Christopher D. Frith},
  title = {{Federated inference and belief sharing}},
  journal = {Neuroscience & Biobehavioral Reviews},
  year = {2024},
  doi = {10.1016/j.neubiorev.2023.105500},
}
```

## File Inventory

- `AGENTS.md` (1,816 bytes)
- `2024_FederatedInference.pdf` (4,200,620 bytes)
- `README.md` (1,546 bytes)
- `SKILL.md` (1,756 bytes)