Federated learning for the real world at the Rochester Security Summit
Dmitrii Korobeinikov, AJ Barea, and Dr. Leon Reznik presented a decision framework for federated learning applications that detect and respond to attacks.
On October 14, Dmitrii Korobeinikov, AJ Barea, and Dr. Leon Reznik presented “Federated Learning for the Real World: Privacy, Security, and Cost, Without the Trade-offs” at the Rochester Security Summit 2026, the 20th edition of the region’s security conference, run by the Rochester chapter of ISSA.
Federated learning keeps customer data on the devices that hold it, which helps with privacy and compliance. Keeping data local does not make a system safe: an attacker who takes over a few devices can corrupt the shared model. The talk gave practitioners a decision framework for federated learning applications that detect and respond to attacks, weigh what each defense costs against model performance, and adapt as conditions change, without a dedicated research team.
For the lab’s research in this area, see InteFL, whose experiment record is on its project page, and our IEEE Intelligent Systems paper on federated learning applications that monitor and adapt to changing conditions.
Filed under Talks, Federated learning, Security