openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23
Resource-Efficient Routing in Mixture-of-Experts Models Based on Multi-Layer Reversible Cellular Automata
The Mixture of Experts (MoE) architecture has demonstrated high efficiency in scalinglarge neural network models. However, deploying sparse MoE models on computationalnodes with strict resource constraints entails critical latencies, high memory consumption,and computational redu…