Neural Architecture Search using Property Guided Synthesis
This work addresses the bottleneck of manual effort and limited search space in NAS for deep learning practitioners, offering a more efficient approach to architecture design.
The paper tackles the problem of neural architecture search (NAS) being limited to small, structured design spaces by proposing a method that searches in an abstract space of program properties, enabling more efficient exploration. It results in models with significantly fewer parameters and FLOPS while maintaining accuracy, such as 96% fewer parameters on CIFAR-10 and improvements over Vision Transformer, ResNet-50, and EfficientNet on ImageNet.
In the past few years, neural architecture search (NAS) has become an increasingly important tool within the deep learning community. Despite the many recent successes of NAS, however, most existing approaches operate within highly structured design spaces, and hence explore only a small fraction of the full search space of neural architectures while also requiring significant manual effort from domain experts. In this work, we develop techniques that enable efficient NAS in a significantly larger design space. To accomplish this, we propose to perform NAS in an abstract search space of program properties. Our key insights are as follows: (1) the abstract search space is significantly smaller than the original search space, and (2) architectures with similar program properties also have similar performance; thus, we can search more efficiently in the abstract search space. To enable this approach, we also propose a novel efficient synthesis procedure, which accepts a set of promising program properties, and returns a satisfying neural architecture. We implement our approach, $α$NAS, within an evolutionary framework, where the mutations are guided by the program properties. Starting with a ResNet-34 model, $α$NAS produces a model with slightly improved accuracy on CIFAR-10 but 96% fewer parameters. On ImageNet, $α$NAS is able to improve over Vision Transformer (30% fewer FLOPS and parameters), ResNet-50 (23% fewer FLOPS, 14% fewer parameters), and EfficientNet (7% fewer FLOPS and parameters) without any degradation in accuracy.