Papers

A collection of papers useful for reviewing the Federated Learning space

Paper	Publication	Date	Description
Federated Machine Learning: Concept and Applications	ACM Transactions on Intelligent Systems and Technology	Feb 2019	A short paper which provides some definitions and use cases for Federated Learning. Includes basic definitions, the importance of privacy, and Federated Learning's connection to other related fields
Automatic differentiation in machine learning: a survey	Journal of Machine Learning Research	Feb 2018	Defines Automatic Differentiation, which is the technique computers use to accurately calculate derivatives, especially in long / complex usages of the chain rule, as needed in Deep Neural Networks. Also differentiates between Automatic, Symbolic, & Numeric Differentiation.
PMF: A Privacy-preserving Human Mobility Prediction Framework via Federated Learning	Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies	March 2020	Privacy focused approach. Introduces Differential Privacy: training two types of features (risky/secure) separately. Provides example attack scenario. Makes use of personal adapters that stay local to each edge device. Model uses multi-modal embedding as input to LSTM.
Federated Semi-Supervised Learning with Inter-Client Consistency	International Conference on Learning Representations	March 2021	Introduces FedMatch to learn inter-client consistency in order to maximize agreement between models trained at different clients. Use of Reliability-based Aggregation allows "better" local models to have more influence on global model updates
Advances and Open Problems in Federated Learning	Foundations and Trends in Machine Learning	March 2021	In-depth survey of open problems and overview of Federated learning. Notable components: Setting (section 1), Attacks (section 5), Fairness & Bias (section 6), On-Device Runtime (section 7.4)
Semi-supervised Federated Learning for Activity Recognition	Imperial College London	March 2021	Local clients train autoencoder for unsupervised learning task to learn representation of time-series data. Cloud server incorporates this representation into global supervised learning pipeline. This is a neat way to address the labelling problem, since only the central server needs trusted labelled data.
Federated Evaluation of On-device Personalization	Google Research	October 2019	Google's use of federated learning on smartphones in order to learn keyboard next-word prediction in a privacy preserving and personalized manner.