Mixtures of Neural Operators Reduce Active Complexity in Operator Learning
Researchers have developed a new method called Mixtures of Neural Operators (MoNOs) to improve the efficiency of operator learning systems. This approach routes input functions to specific 'expert' neural operators, reducing the computational complexity for each query. The study demonstrates that MoNOs can approximate operators with smaller active expert sizes compared to traditional single-operator constructions, with theoretical guarantees on approximation accuracy. AI