Building Optimal Neural Architectures using Interpretable Knowledge
This addresses the problem of expensive architecture search for researchers and practitioners in deep learning, offering a more efficient alternative.
The paper tackles the high cost of Neural Architecture Search by proposing AutoBuild, which learns to assign interpretable importance scores to architecture modules based on performance, enabling the construction of high-quality neural networks without search. It demonstrates that AutoBuild outperforms original labeled architectures and search baselines in image classification, segmentation, and Stable Diffusion models.
Neural Architecture Search is a costly practice. The fact that a search space can span a vast number of design choices with each architecture evaluation taking nontrivial overhead makes it hard for an algorithm to sufficiently explore candidate networks. In this paper, we propose AutoBuild, a scheme which learns to align the latent embeddings of operations and architecture modules with the ground-truth performance of the architectures they appear in. By doing so, AutoBuild is capable of assigning interpretable importance scores to architecture modules, such as individual operation features and larger macro operation sequences such that high-performance neural networks can be constructed without any need for search. Through experiments performed on state-of-the-art image classification, segmentation, and Stable Diffusion models, we show that by mining a relatively small set of evaluated architectures, AutoBuild can learn to build high-quality architectures directly or help to reduce search space to focus on relevant areas, finding better architectures that outperform both the original labeled ones and ones found by search baselines. Code available at https://github.com/Ascend-Research/AutoBuild