ABSTRACT This paper proposes a distributed online optimization algorithm for multiagent networks where communication is limited to quantized data. Each agent converts its current estimate of the optimal strategy into a quantized value through a uniform quantizer. These quantized values are then sent to neighboring agents. After receiving the quantized data from neighbors, each agent reconstructs real‐valued estimates by applying a decoding process. Using this decoded information, the agent updates its estimate by applying a distributed adaptive gradient descent method. Theoretical analysis demonstrates that a sublinear regret bound can be achieved. Numerical experiment shows that the algorithm achieves a comparable performance to that with unquantized communication while reducing the communication load.
Yamamoto et al. (Sun,) studied this question.