Federated Learning and Cooperative NN represent emerging paradigms for thinking and designing learning scenarios and Neural Networks able to exploit the growing computational capabilities of the modern distributed networks, that generating a wealth of data each day. Federated and Cooperative Learning is promising in mitigating the issues concerning privacy protection and private information sharing. It is increasingly attractive to store data locally and push network computation to the edge devices. Federated learning has emerged as a training paradigm in such settings but it raises new questions at the intersection of machine learning and systems and requires fundamental advances in privacy, large-scale machine learning, and distributed optimization areas. Federated learning and classical distributed learning share a similar goal of minimizing the empirical risk over distributed entities. However, solving this objective in the federated settings poses some fundamental challenges as Privacy Concerns, Systems Heterogeneity, Statistical Heterogeneity, Expensive Communication, Trust in Federation Membership. According to my latest experience (and inspired by the trends addressed by the most authoritative players in the AI field (Facebook AI, Google AI) and the availability of specific frameworks (cfr. TensorFlow Federated, e.g.), Federated and Cooperative Learning are among the most recent and challenging paradigm related to machine learning and to its implementations by NNs, also considering that such kinds of distributed learning systems are quite vulnerable to security attacks (data or model poisoning attacks), thus offering several open issues and multiple directions of investigation.
The IJCNN 2021 organizers and this Special Session Technical Program Committee cordially invite internationally recognized experts to propose their novel and yet unpublished contribution to this Special Sessions mainly focusing on Federated Learning and Cooperative Neural Networks issues, challenges, frontiers, current and future research and applicative directions, also extending the contributions within the general scope and topics of the main conference in both classic, contemporary, and emerging topics of interest to the Neural Networks, Machine Intelligence, Cognitive Neuroscience, Artificial Intelligence, and Data Analytics communities.
Papers submitted for special sessions will be peer-reviewed in the same way as submissions to the regular sessions for the main conference.