Kognitivt federerat lärande
- Diarienummer
- SAB19-0005
- Projektledare
- Chen, Jiajia
- Start- och slutdatum
- 200101-221231
- Beviljat belopp
- 1 728 238 kr
- Förvaltande organisation
- Chalmers University of Technology
- Forskningsområde
- Informations-, kommunikations- och systemteknik
Summary
The main objective of this proposal is to explore a revolutionary, highly scalable distributed machine learning (ML) paradigm, where network intelligence is introduced to boost learning performance. In federated learning (FL), multiple data owners collaboratively train a model but do not need to expose their data to each other. Compared to the classical ML, where all data is accessed by the centralized cloud, the FL perfectly fits the Internet based services and assures data privacy, but its limited access to data could cause severe learning performance degradation. In this regard, we propose a new learning paradigm, cognitive federated learning, which greatly extends the FL with cognitive control of learning process and fully explores network intelligence to enhance learning performance. Research problems in geographical, temporal and data domains to control the FL process will be addressed by developing network automation strategies for cognitive device/edge selection, synchronization and balance of data/model sharing, respectively. Finally, a joint optimization of learning performance will be carried out by integrating all three aspects. The applicant has demonstrated strong competence in communication networks and will address core problems in ML. One-year sabbatical in California Institute of Technology enables her to initiate a strategic move towards a new research area and foster excellent interdisciplinary research on combining communication networks with ML.
Populärvetenskaplig beskrivning
Artificial intelligence (AI) techniques become essential for everyday life. The global AI software market is expected to experience a massive growth, with a forecast annual revenue increase over 40% in the coming years. Machine learning (ML) is a key branch of AI. Classical ML approaches require centralizing the training data on one machine or a datacenter. Driven by popularity of 5G as well as cloud/edge computing, federated learning (FL) is introduced that supports mobile users to collaboratively learn a shared prediction model while keeping all the training data on device, decoupling the ability to do ML from the need to store the data in the cloud. Such a distributed manner well fits many Internet based services, e.g., connected vehicles and Internet of things, and attracts many interests from various industries, such as automotive manufacturers and ICT system vendors. Besides, the FL is also an excellent learning paradigm to address data privacy and ownership, which are essential for society. However, the FL suffers severe performance degradation in terms of learning accuracy and efficiency, particularly when mobile users are highly heterogeneous. In this regard, the strategic relevance of this project stems from the need of a new distributed learning paradigm that can address data privacy and ownership issues but also able to achieve efficient learning performance. The applicant, Prof. Jiajia Chen at Chalmers University of Technology (Chalmers), and the host, Prof. Animashree Anandkumar in California Institute of Technology (Caltech), USA, are already highly recognized scientists in the areas of communication networks and ML algorithms, respectively. One-year sabbatical leave in Caltech enables the applicant to initiate a strategic bold move towards a new research area and foster excellent interdisciplinary research on combining communication networks with ML. Known for its strength in natural science and engineering, Caltech is often ranked as one of the world's top-ten universities. ML is a core area in the Computing and Mathematical Science discipline at Caltech. Caltech is particularly interested in pursuing emerging connections between ML and other disciplines, which will stimulate the development of the next generation of ML methods. Therefore, we strongly believe the project will bring mutual benefits and enable long-term cooperation for two leading technical universities in the USA and Sweden.