Stretchable Cells Help DARTS Search Better
This work provides an incremental improvement to DARTS, a neural architecture search method, by addressing its tendency to produce wide and shallow cells, which is a problem for researchers and practitioners using DARTS.
The paper addresses the issue of Differentiable Neural Architecture Search (DARTS) methods tending to produce wide and shallow cells, leading to sub-optimal performance, especially with fewer layers. They introduce stretchable cells with explicit topological variables and a combinatorial probabilistic distribution to model target topology, allowing for more diverse and complex topologies. This approach improves DARTS by 0.28% accuracy on CIFAR-10 with 45% fewer parameters or 2.9% on ImageNet with similar FLOPs.
Differentiable neural architecture search (DARTS) has gained much success in discovering flexible and diverse cell types. To reduce the evaluation gap, the supernet is expected to have identical layers with the target network. However, even for this consistent search, the searched cells often suffer from poor performance, especially for the supernet with fewer layers, as current DARTS methods are prone to wide and shallow cells, and this topology collapse induces sub-optimal searched cells. In this paper, we alleviate this issue by endowing the cells with explicit stretchability, so the search can be directly implemented on our stretchable cells for both operation type and topology simultaneously. Concretely, we introduce a set of topological variables and a combinatorial probabilistic distribution to explicitly model the target topology. With more diverse and complex topologies, our method adapts well for various layer numbers. Extensive experiments on CIFAR-10 and ImageNet show that our stretchable cells obtain better performance with fewer layers and parameters. For example, our method can improve DARTS by 0.28\% accuracy on CIFAR-10 dataset with 45\% parameters reduced or 2.9\% with similar FLOPs on ImageNet dataset.