federated learning

federated learning
MarcT0K · CC BY-SA 4.0 · source
Name	Federated learning

Contents

federated learning Federated learning is a collaborative machine learning paradigm that enables multiple participants to train shared models without exchanging raw data. It contrasts with centralized approaches practiced by organizations such as Google, Apple Inc., IBM, and Microsoft by keeping data local to devices or institutions. The approach has drawn attention from research communities associated with Massachusetts Institute of Technology, Stanford University, Carnegie Mellon University, University of California, Berkeley, and University of Oxford as well as industry labs like DeepMind, OpenAI, Facebook AI Research, and NVIDIA.

Overview

Federated learning emerged from intersections of distributed systems research at institutions such as Bell Labs, Xerox PARC, and AT&T Labs and statistical learning theories advanced at Princeton University, Harvard University, and University of Cambridge. Early driving applications included mobile telemetry from products by Samsung Electronics, Google Pixel, and services by Mozilla Foundation and WhatsApp owned by Meta Platforms, Inc.. Research funding and policy interest have involved agencies and programs like the National Science Foundation, European Commission, Defense Advanced Research Projects Agency, and collaborations with firms such as Siemens and Philips. Ethical, legal, and regulatory considerations intersect with institutions such as the European Union and statutes like the General Data Protection Regulation debated in venues including the United Nations and national data protection authorities.

Technical approaches draw on optimization techniques from traditions linked to Bell Labs, algorithms from researchers at Courant Institute of Mathematical Sciences and Turing Institute, and cryptography advances from RSA Laboratories and scholars associated with Stanford Law School and University of Waterloo. Core methods include synchronous and asynchronous aggregation inspired by protocols used in Apache Hadoop and Apache Spark ecosystems and parameter server designs similar to those in Google Brain. Secure aggregation protocols leverage primitives pioneered by cryptographers at MIT Computer Science and Artificial Intelligence Laboratory, Weizmann Institute of Science, and École Polytechnique Fédérale de Lausanne. Approaches vary between cross-device settings exemplified by deployments on devices from Apple Inc. and Samsung Electronics and cross-silo settings used by institutions like Mayo Clinic and Cleveland Clinic or consortia involving HSBC, JPMorgan Chase, and Goldman Sachs for privacy-preserving analytics. Model architectures used range from convolutional networks developed at Yann LeCun’s lineage to transformer variants associated with research at Google Research and OpenAI.

Privacy-preserving techniques integrate cryptographic mechanisms such as secure multi-party computation advanced at University of Maryland and Microsoft Research, and differential privacy frameworks originating from scholars at Microsoft Research and Harvard University. Robustness and adversarial defenses relate to work by labs at Berkeley Artificial Intelligence Research Lab and Caltech. Threat models consider insider attacks, poisoning, and backdoor strategies studied in security venues including conferences like USENIX, IEEE Symposium on Security and Privacy, and ACM CCS. Fairness concerns connect to ethics research at Oxford Internet Institute, policy studies at Brookings Institution and RAND Corporation, and standards discussions at International Organization for Standardization and National Institute of Standards and Technology. Legal assessments engage law faculties at Yale Law School and Columbia Law School regarding compliance with statutes like California Consumer Privacy Act.

Applications span healthcare collaborations between Johns Hopkins Hospital, Mayo Clinic, and Cleveland Clinic for clinical prediction models; finance use cases with firms like Visa, Mastercard, and Goldman Sachs for fraud detection; and telecommunications deployments by Verizon Communications and AT&T Inc. for network optimization. Mobile personalized services have been implemented in products by Google, Apple Inc., and Samsung Electronics for on-device keyboards and recommendations. Cross-institution scientific collaborations include projects with CERN and astronomy groups at European Southern Observatory and National Aeronautics and Space Administration centers. Industrial use includes predictive maintenance in companies like General Electric and Siemens and smart-city initiatives involving municipalities such as New York City and Singapore.

Benchmarking efforts reference datasets and platforms maintained by organizations like UCI Machine Learning Repository, Kaggle, ImageNet initiatives linked to research at Stanford University and Princeton University, and biomedical repositories managed by National Institutes of Health. Evaluation metrics build on statistical practices developed at American Statistical Association and machine learning benchmarks from NeurIPS, ICML, CVPR, and ACL communities. Open-source frameworks used for reproducibility include projects hosted by GitHub, integrations with TensorFlow from Google, libraries from PyTorch associated with Meta Platforms, Inc. research, and orchestration work from Kubernetes and Docker ecosystems supported by Cloud Native Computing Foundation.

Key challenges involve systems-scale issues studied at MIT CSAIL and CMU for heterogeneous device participation, communication constraints reminiscent of problems addressed by Cisco Systems and Qualcomm, and economic incentives explored by scholars at London School of Economics and Harvard Business School. Future directions intersect with quantum-resilient cryptography researched at IBM Research and NIST, regulatory frameworks shaped by the European Commission and national legislatures, and interdisciplinary collaborations with public health agencies such as World Health Organization and disaster-response organizations like International Federation of Red Cross and Red Crescent Societies. Emerging topics include integration with edge computing paradigms promoted by Intel Corporation and ARM Holdings, governance models studied at The Brookings Institution and Council on Foreign Relations, and standards development within bodies like IEEE and ISO.