In Autumn 2022, the Information Theory Workshop (ITW) hosted the first ITW competition as part of the invited session on "Communication-efficient gradient compression and coding in distributed learning". The task was about training a channel state information (CSI) compressor for massive MIMO in 5G under severe communication constraints. In three subtasks, the competitors trained autoencoders in a federated fashion to cope with communication constraints and provide failure resilience in CSI compression.
The TUM team consisted of the master student Cengizhan Kaya and the doctoral researchers Christoph Hofmeister and Maximilian Egger at the Professorship of Coding and Cryptography and achieved second place (https://itw2022.in/themed-session-announcements/).