CliqueAI
SN83AI network that solves the maximum clique problem, a fundamental challenge in graph theory
Graph theory meets decentralized compute. CliqueAI tasks a network of miners with solving one of combinatorics' hardest problems, using a four-stage mechanism that matches problems to miners, scores solutions, and continuously refines performance.
// AI-powered maximum clique solving on Bittensor.
CliqueAI is a Bittensor subnet (SN83) that distributes maximum clique problems across a network of AI-powered miners. The maximum clique problem asks: given a graph (a network of connected nodes), what is the largest group of nodes where every pair is directly connected? It's a classic NP-hard problem with applications in drug discovery, network analysis, and social network mapping.
The simple version: It's like a decentralized competition where miners race to find the densest connected cluster in a complex network of relationships. Harder and larger graphs are worth more.
Centralized equivalent: No direct single-platform equivalent. Similar compute problems are typically solved by academic supercomputers or specialized research tools.
How it works:
- Miners receive graph problems, apply AI algorithms to find the maximum clique, and submit solutions with both the result and the methodology
- Validators score submissions using a dual-metric system that rewards both solution optimality and algorithmic diversity
- The problem it solves: Maximum clique and related graph optimization problems are computationally expensive and difficult to parallelize on centralized infrastructure. A distributed network of specialized miners can tackle more variety simultaneously.
- The opportunity: Applications span bioinformatics (identifying protein interaction clusters), fraud detection (finding tightly connected fraud rings), and social network analysis.
- The Bittensor advantage: Rewarding both optimality and diversity of approach encourages genuinely different algorithmic strategies, rather than convergence on a single method.
- Traction signals: 245 active miners, making it one of the more populated subnets. Price is up 8.7% over 7 days and 34.6% over 90 days. Net buy volume of 30.6 TAO over 24 hours.
Category: Reinforcement Learning | Centralized Competitor: No direct equivalent (academic HPC clusters, specialized research tools)
Graph optimization problems have real-world value across several industries, but they are also genuinely hard. The maximum clique problem is NP-hard, meaning the search space grows exponentially with graph size. Distributing this work across a subnet of specialized miners creates a competitive market for solving these problems efficiently.
Mechanism:
According to the CliqueAI GitHub README, the network uses a four-stage mechanism. First, an AI system curates graph problems from a distributed database, categorizing them by difficulty, structure, and computational requirements. Second, a smart allocation engine distributes problems to miners based on historical performance and experience level, with stake-weighted distribution. Third, validators score solutions using a dual-metric system that rewards both solution optimality and algorithmic diversity. Fourth, an exponential moving average algorithm continuously updates miner reputation scores, influencing future problem allocation.
The subnet has 16 commits from 1 contributor, with the most recent commit in March 2026, showing recent active development. Emission share is around 0.36%, with root proportion at 0.27, indicating organic staking is the dominant pool component. The Gini coefficient of 0.77 suggests concentrated holder distribution.
- Single contributor: The GitHub repo shows 1 active contributor. Key-person risk is high if that person steps back.
- Concentration: Gini of 0.77 and HHI of 0.12 indicate a concentrated holder base with limited organic breadth.
- Niche use case: Maximum clique solving is a real problem space but a narrow one. Broader demand for this as a paid service is unproven.
- Low emission share: At 0.36%, the subnet receives a small fraction of network emissions. Any sustained net outflow would push it toward zero under Taoflow.
- Development depth: 16 commits is a thin codebase for a complex distributed problem-solving system. Details of the AI problem curation system are not fully documented publicly.