Table of Contents:
Nous Research Secures $50 Million for Decentralized AI Training
Nous Research, a decentralized artificial intelligence startup, has announced the successful completion of a $50 million early-stage funding round, as reported by SiliconANGLE. The round was led almost entirely by the crypto-focused investment firm Paradigm. This Series A round follows previous seed rounds totaling $20 million, with investors such as Distributed Global, North Island Ventures, and Delphi Digital. The latest funding brings the company's crypto token valuation to $1 billion.
Nous Research utilizes the Solana blockchain network to enable distributed training for AI models. This approach allows users to contribute idle compute resources from both consumer- and industrial-grade graphics processing units (GPUs). The company aims to challenge the dominance of centralized AI development by major players like DeepSeek and OpenAI, instead fostering a network where numerous participants can join and contribute resources.
Although blockchain was not part of the company's original vision, it was later adopted to facilitate decentralized model training and to reward participants with blockchain-based cryptocurrency for their unused GPU compute. According to Nous cofounder Karan Malhotra, the incentive mechanism behind crypto encourages people to utilize their idle compute as a transaction rather than a donation. Blockchain also provides a transparent historical audit trail of all data committed to the system, which is crucial for detecting and preventing data poisoning attacks.
Malhotra acknowledged skepticism surrounding blockchain and cryptocurrency, particularly regarding scams and hacking, but emphasized that the research team was initially hesitant about blockchain until a recent breakthrough in decentralized training. The company recently launched Psyche, a network on the Solana blockchain designed for fault-tolerant, globally distributed AI model training. In December 2024, Nous successfully tested a 15 billion parameter model, scaling the network through 11,000 steps. The core of Psyche is Nous Distro, a component that allows each node to train independently, significantly reducing the bandwidth required compared to centralized training.
| Funding Round | Amount Raised | Company Valuation | Key Technology |
|---|---|---|---|
| Series A | $50 million | $1 billion (crypto token valuation) | Solana blockchain, decentralized AI training |
- Nous Research raised $50 million in Series A funding led by Paradigm.
- The company’s total funding, including seed rounds, is $70 million.
- Valuation of the company’s crypto token reached $1 billion.
- Psyche network enables distributed AI training with 15 billion parameter model tested over 11,000 steps.
Key Takeaway: Nous Research is leveraging blockchain technology to decentralize AI training, offering transparency, incentives for resource sharing, and enhanced security against data poisoning, with a significant $50 million investment and a $1 billion valuation. (Source: SiliconANGLE)
Preventing Data Poisoning in AI: FIU Researchers Combine Federated Learning and Blockchain
A team of cybersecurity researchers at Florida International University (FIU) has developed a novel approach to defend AI systems against data poisoning attacks, according to FIU News. Data poisoning involves cyber attackers inserting false or misleading information into AI training sets, potentially causing models to behave unpredictably or dangerously. The consequences can range from malfunctioning chatbots to critical failures in self-driving cars or power grid disruptions.
To address this threat, the FIU team combined federated learning and blockchain technologies. Federated learning allows AI models to be trained directly on user devices, sharing only updates with a central server, thus preserving privacy. However, this method remains vulnerable to data poisoning, as verifying the honesty of user data before it reaches the model is challenging.
Blockchain, known for its role in cryptocurrency, offers a tamper-proof, distributed database where data is stored in blocks linked chronologically. The FIU researchers leveraged blockchain’s structure to vet and compare block updates, identifying and discarding potentially poisonous data before it could compromise the training dataset. Their approach, detailed in the IEEE Transactions on Artificial Intelligence, successfully detected and removed dishonest data.
“We’ve built a method that can have many applications for critical infrastructure resilience, transportation cybersecurity, healthcare and more,” said Hadi Amini, lead researcher and FIU assistant professor in the Knight Foundation School of Computing and Information Sciences.
The team is now collaborating with the National Center for Transportation Cybersecurity and Resiliency to integrate quantum encryption for further protection. Their ongoing research aims to ensure the safety and security of America’s transportation infrastructure while harnessing advanced AI to enhance these systems. The research is partly funded by the Advanced Education and Research for Machine Learning-driven Critical Infrastructure Resilience (ADMIRE) Center and the National Center for Transportation Cybersecurity and Resiliency.
- Data poisoning can cause AI models to malfunction, with real-world risks for critical infrastructure.
- FIU’s approach combines federated learning and blockchain to detect and remove dishonest data.
- The method has potential applications in transportation, healthcare, and infrastructure security.
- Research is supported by national centers and aims to integrate quantum encryption for added security.
Key Takeaway: FIU researchers have demonstrated that combining federated learning with blockchain can effectively detect and prevent data poisoning in AI systems, with broad implications for the security of critical infrastructure. (Source: FIU News)
Sources:
