ShuffleNASNets: Efficient CNN models through modified Efficient Neural Architecture Search
This work addresses efficiency and complexity issues in automated neural architecture search for computer vision tasks, representing an incremental improvement by combining existing methods.
The paper tackled the problem of complex and slow neural architectures from automated search by integrating expert design principles from ShuffleNet V2 into Efficient Neural Architecture Search (ENAS), resulting in models that maintain a 2.85% test error on CIFAR-10 while being two times faster and requiring fewer parameters.
Neural network architectures found by sophistic search algorithms achieve strikingly good test performance, surpassing most human-crafted network models by significant margins. Although computationally efficient, their design is often very complex, impairing execution speed. Additionally, finding models outside of the search space is not possible by design. While our space is still limited, we implement undiscoverable expert knowledge into the economic search algorithm Efficient Neural Architecture Search (ENAS), guided by the design principles and architecture of ShuffleNet V2. While maintaining baseline-like 2.85% test error on CIFAR-10, our ShuffleNASNets are significantly less complex, require fewer parameters, and are two times faster than the ENAS baseline in a classification task. These models also scale well to a low parameter space, achieving less than 5% test error with little regularization and only 236K parameters.