Evolutionary NAS with Gene Expression Programming of Cellular Encoding
This work addresses the problem of inefficient encoding in evolutionary NAS for researchers and practitioners in computer vision, though it is incremental as it builds on prior genetic methods.
The paper tackled the scalability and complexity issues in neural architecture search (NAS) for convolutional neural networks (CNNs) by introducing a symbolic linear generative encoding (SLGE) scheme, which improved performance on CIFAR-10 and CIFAR-100 tasks and achieved competitive error rates with less GPU resources compared to existing NAS methods.
The renaissance of neural architecture search (NAS) has seen classical methods such as genetic algorithms (GA) and genetic programming (GP) being exploited for convolutional neural network (CNN) architectures. While recent work have achieved promising performance on visual perception tasks, the direct encoding scheme of both GA and GP has functional complexity deficiency and does not scale well on large architectures like CNN. To address this, we present a new generative encoding scheme -- $symbolic\ linear\ generative\ encoding$ (SLGE) -- simple, yet powerful scheme which embeds local graph transformations in chromosomes of linear fixed-length string to develop CNN architectures of variant shapes and sizes via evolutionary process of gene expression programming. In experiments, the effectiveness of SLGE is shown in discovering architectures that improve the performance of the state-of-the-art handcrafted CNN architectures on CIFAR-10 and CIFAR-100 image classification tasks; and achieves a competitive classification error rate with the existing NAS methods using less GPU resources.