We are now on summer break - Thank you for another great year!

You can always check out our large backlog of talks on YouTube.

See you next Fall!

FUTURE TALKS

Sept. 3:  Praneeth Vepakomma (MIT, Mohamed bin Zayed University of A. I.)

Modulated learning for private and distributed regression with just a single sample per client device

This talk focuses on the problem of learning from a large number of devices, where each device holds only a single sample of privacy-sensitive data. Several real-world applications exist to this one privacy sensitive sample per client setup up including learning from fitness trackers, data/app usage aggregators, body-worn sensing devices, continuous glucose monitoring, and daily event monitors to name a few. The proposed approach injects a single, carefully calibrated noisy perturbation to transform the sample at each client, followed by a post-processed representation which is shared with the server. These representations aggregated at the server are processed to obtain an unbiased gradient update that in expectation matches the non-private centralized gradient while preserving data privacy. This approach is different than traditional private federated learning, where the communication payloads involve model coefficients as opposed to privately transformed data samples. This method enables devices with extremely limited data to collaborate and learn accurate, privacy-preserving models without requiring large local datasets or sacrificing individual privacy. This is based on work done in https://arxiv.org/pdf/2605.07233 with Praneeth Vepakomma, Amirhossein Reisizadeh, Samuel Horváth and Munther Dahleh.