Approximating optimal Broadcast in Wireless Mesh Networks with Machine Learning

With the growth of IoT, efficient broadcast is required for many applications. Yet, current protocols use primitive mechanisms based on heuristics. Multi-agent reinforcement learning is applied to approximate optimal broadcast in Wireless Mesh Networks. One of the proposed fully distributed algorithms, using Bayesian Neural Networks, outperforms MORE multicast and BATMAN, improving airtime up to 20%, e2e delay up to 30%, and satisfying timeout constraints in over the 97% of the cases.