Vector Institute Unveils CaPC for Secure ML Collaboration
Researchers at the Vector Institute for Artificial Intelligence have developed Confidential and Private Collaborative Learning (CaPC), a new system designed to enable secure collaboration between institutions such as banks and hospitals. This privacy-enhancing technology allows organizations to jointly work with machine learning models without revealing sensitive inputs, training data, or proprietary models to one another. Unlike existing systems that typically offer either privacy or confidentiality, CaPC guarantees both by combining cryptographic tools with privacy research methods. The system aims to improve model accuracy and fairness while strictly adhering to data protection regulations prevalent in the health and finance sectors. Currently a proof of concept, CaPC is being integrated into an industry toolkit featured at Vector’s upcoming PETs Bootcamp. This three-day event will also demonstrate other technologies like Federated Learning and Differential Privacy, aiming to bridge the gap between academic research and industrial application. By showcasing these tools, Vector Institute seeks feedback from industry partners to refine the system for real-world deployment, facilitating safer data sharing and collaborative AI development across regulated industries.
Wire timeline
Vector Institute Unveils CaPC for Secure ML Collaboration
Researchers at the Vector Institute for Artificial Intelligence have developed Confidential and Private Collaborative Learning (CaPC), a new system designed to enable secure collaboration between institutions such as banks and hospitals. This privacy-enhancing technology allows organizations to jointly work with machine learning models without revealing sensitive inputs, training data, or proprietary models to one another. Unlike existing systems that typically offer either privacy or confidentiality, CaPC guarantees both by combining cryptographic tools with privacy research methods. The system aims to improve model accuracy and fairness while strictly adhering to data protection regulations prevalent in the health and finance sectors. Currently a proof of concept, CaPC is being integrated into an industry toolkit featured at Vector’s upcoming PETs Bootcamp. This three-day event will also demonstrate other technologies like Federated Learning and Differential Privacy, aiming to bridge the gap between academic research and industrial application. By showcasing these tools, Vector Institute seeks feedback from industry partners to refine the system for real-world deployment, facilitating safer data sharing and collaborative AI development across regulated industries.
Vector Institute for Artificial Intelligence