FIU Researchers Develop AI Protection Against Data Poisoning

Researchers at Florida International University have developed a novel method to safeguard AI systems from data poisoning attacks, announced in a press release. This innovative approach integrates federated learning and blockchain technology to detect and eliminate malicious data before it can compromise AI models.

Data poisoning involves inserting false information into datasets used for training AI, potentially leading to dangerous outcomes such as autonomous vehicles ignoring traffic signals. The FIU team, led by Hadi Amini, has addressed this threat by using federated learning, which allows AI models to train across multiple devices without centralizing sensitive data. However, federated learning alone is vulnerable to poisoned updates.

To enhance security, the researchers incorporated blockchain technology, known for its tamper-proof verification capabilities. This addition helps flag and discard outliers in data updates, preventing potential threats from reaching the global model. The research is being further developed with partners at the National Center for Transportation Cybersecurity and Resiliency, aiming to integrate quantum encryption for even stronger data protection.