Traditional federated learning (FL) relies on a central aggregator, creating potential performance bottlenecks and privacy risks. Decentralized alternatives remove the server, but repeated local mixing can weaken global information under heterogeneous data and expose peer-to-peer communication patterns. This seminar presents FLTorrent, a BitTorrent-based dissemination layer for serverless FL that preserves FedAvg-style aggregation semantics while improving scalability and source privacy. FLTorrent introduces a short warm-up phase combining pre-round obfuscation, randomized lags, and non-owner-first chunk scheduling before switching to standard BitTorrent swarming. The design provides within-round source unlinkability while keeping the tracker off the data path. Analytical and experimental results show that FLTorrent approaches bandwidth-optimal dissemination, maintains stable warm-up overhead as the network grows, and reduces source-attribution success toward neighborhood-level random guessing. In LLM-scale dissemination stress tests, FLTorrent incurs only 6–10% round-time overhead compared with BitTorrent-only dissemination.

About Naicheng Li

Naicheng Li is a second-year PhD researcher at IMDEA Networks Institute (+UC3M), working under the supervision of Prof. Nikolaos Laoutaris. His research lies at the intersection of distributed systems, networking, and machine learning, with a particular focus on decentralized federated learning and privacy-preserving protocols. Before joining IMDEA Networks, he obtained his master’s degree from Chalmers University of Technology in Sweden.

This event will be conducted in English