Communication-Efficient Distributed Cooperative Learning With Compressed Beliefs

In this article, we study the problem of distributed cooperative learning, where a group agents seeks to agree on set hypotheses that best describes sequence private observations. scenario is large, propose belief update rule share compressed (either sparse or quantized) beliefs with an arbitrary positive compression rate. Our algorithm leverages unified communication enables access wide-ranging operators as black-box modules. We prove almost sure asymptotic convergence optimal hypotheses. Additionally, show nonasymptotic, explicit, and linear concentration rate in probability hypothesis set. provide numerical experiments illustrate benefits our method. The simulation results number transmitted bits can be reduced 5%–10% noncompressed method studied scenarios.