Federated learning is a recent and rapidly expanding area of machine learning that allows parties to benefit from joint training of models whilst respecting the privacy of each party's data. IBM Research has a broad effort in federated learning comprising novel methods, models and paradigms and offers an enterprise-strength federated learning platform free to use for non-commercial purposes, the IBM Federated Learning Community Edition.

The session will give an overview through a series of 7 short talks on the most exciting new research results from IBM Research in federated learning. Questions shall be collected using the Chat window and addressed after the lightning talks as well as after the live-interactive demo.

In this talk, I will introduce recent works on model fusion, a special case of Federated Learning where only a single communication round is allowed. This setting has a unique feature where it is sufficient for clients to have a pre-trained model, but not the data. Data storage regulations such as GDPR make this setting appealing as the data can be immediately deleted after updating the local model before FL starts. In addition to deep learning methods, I will cover unsupervised settings such as mixture models, topic models, and hidden Markov models.

We present two work targeted at (i) preserving privacy of training data, and (ii) improving adversarial robustness of models, in federated learning (FL).

Privacy of training data is key to FL. Model updates were assumed to be private for a long time, but recent reconstruction attacks (DLG,IDLG, IG) have demonstrated otherwise. Existing techniques for privacy preservation (encryption, secure multiparty computation and addition of statistical noise to model updates) are useful but have drawbacks -- either in terms of performance or model accuracy. We present recent work on effective use of model shuffling combined with trusted execution environments (TEEs) for aggregation.

Adversarial training (AT) in the federated learning setting is challenging given limited communication budget and non-iid data distribution among. We propose FedDynAT, a novel algorithm for efficient AT in federated learning. FedDynAT builds on techniques proposed for preventing catastrophic forgetting in federated learning with non-iid data by augmenting them with a dynamic local AT schedule. FedDynAT improves the convergence time upto a factor 14x for limited communication budget and achieves high accuracy at convergence as compared to other state-of-the-art schemes.