Axiom Distributed AI
Axiom Distributed AI
Axiom Distributed AI is an experimental volunteer distributed computing project built on the BOINC platform. The project explores a new approach to artificial intelligence training using distributed, biologically inspired learning techniques rather than conventional centralized AI infrastructure.
Axiom invites volunteers from around the world to donate spare CPU and GPU processing power to help train a massive neural network collaboratively across thousands of computers.
Official website: Axiom Distributed AI Open source development: GitHub Repository
Overview
Unlike most modern AI systems that rely on centralized datacenters and expensive GPU clusters, Axiom experiments with a decentralized learning architecture inspired by Hebbian learning — often summarized as:
"Neurons that fire together wire together."
The project distributes training tasks across volunteer computers worldwide using the BOINC infrastructure. Each participant contributes small amounts of computation toward a larger shared neural model.
Axiom is currently considered a proof-of-concept research project exploring whether distributed Hebbian learning can scale effectively in a real-world volunteer computing environment.
Why Axiom Distributed AI?
Traditional AI systems typically depend on:
- Massive centralized datacenters
- High energy consumption
- Large-scale backpropagation training
- Expensive specialized hardware
Axiom instead investigates whether learning can emerge from:
- Distributed volunteer computation
- Correlation-based learning
- Decentralized pattern recognition
- Biologically inspired neural adaptation
The long-term vision is to explore alternative methods for building AI systems that are:
- More decentralized
- More energy efficient
- Community driven
- Resistant to centralized control
Goal
The immediate goal of the project is to demonstrate that:
Distributed Hebbian learning works at scale.
If successful, the research could contribute toward new approaches in:
- Neural network training
- Decentralized artificial intelligence
- Distributed machine learning
- Collaborative AI infrastructure
- Emergent pattern recognition systems
In the long term, models trained on extremely diverse real-world data could potentially become foundational systems for future AI applications.
At present, the project primarily serves as an experimental research platform and proof of concept.
Methods
Axiom trains neural systems to recognize patterns in data using the spare processing power of volunteer computers.
Instead of conventional language-model training, Axiom currently focuses on learning:
- Bit-level compression patterns
- Sequence prediction
- Correlations within data streams
- Generalized pattern recognition
Key Technical Concepts
- Not an LLM
- Axiom is not currently training a Large Language Model (LLM).
- The system instead attempts to learn predictive patterns at the bit-stream level.
- Distributed Learning
- Training workloads are distributed across volunteers worldwide through BOINC.
- Each computer contributes partial learning updates to the shared model.
- Correlation-Based Credit System
- Random or low-quality data naturally contributes less to the model.
- Useful learning signals correlate more strongly with other contributors' gradients.
- This creates a natural weighting system favoring meaningful patterns.
- Noise Reduction
- Random noise generally fails to correlate with broader learning patterns.
- As a result, noisy data contributes less to overall model development.
- Generalization Through Diversity
- Diverse datasets may improve the model's ability to generalize across different data types.
- Text, code, images, documents, and structured data may all contribute useful patterns.
Monitoring Learning Progress
The Axiom website provides live statistics and progress monitoring.
One key metric is:
Bit loss
When bit loss trends downward over time, it suggests the neural model is learning meaningful structures and predictive relationships rather than memorizing random noise.
Volunteers can monitor:
- Model learning progress
- Training statistics
- Network participation
- Workunit activity
- Validation metrics
Contributing Data
Volunteers may optionally contribute local files to help train the system.
Windows
Place files in:
%USERPROFILE%\Axiom\contribute\
Linux Mint 22.3
Place files in:
\boinc\Axiom\contribute\
Supported Contribution Types
Users may contribute many different file types, including:
- Documents
- Images
- Source code
- PDFs
- Logs
- Structured datasets
- Experimental real-time data streams
The project encourages experimentation and creative uses of distributed training data.
Open Source Development
Axiom Distributed AI is open source software.
Source code and development discussions are available on GitHub:
The project welcomes:
- Developers
- Testers
- Researchers
- BOINC volunteers
- AI enthusiasts
Project Team / Sponsors
- PyHelix
- Axiom Project