RADARS: Memory Efficient Reinforcement Learning Aided Differentiable Neural Architecture Search
This addresses memory efficiency for researchers and practitioners in neural architecture search, enabling exploration of large search spaces on standard GPUs, though it is incremental as it combines existing RL and DNAS methods.
The paper tackles the memory explosion problem in differentiable neural architecture search (DNAS) by proposing RADARS, a scalable RL-aided framework that prunes candidates with RL and applies DNAS to promising subspaces, achieving up to 3.41% higher accuracy with 2.5X faster search on CIFAR-10 and ImageNet while using bounded memory.
Differentiable neural architecture search (DNAS) is known for its capacity in the automatic generation of superior neural networks. However, DNAS based methods suffer from memory usage explosion when the search space expands, which may prevent them from running successfully on even advanced GPU platforms. On the other hand, reinforcement learning (RL) based methods, while being memory efficient, are extremely time-consuming. Combining the advantages of both types of methods, this paper presents RADARS, a scalable RL-aided DNAS framework that can explore large search spaces in a fast and memory-efficient manner. RADARS iteratively applies RL to prune undesired architecture candidates and identifies a promising subspace to carry out DNAS. Experiments using a workstation with 12 GB GPU memory show that on CIFAR-10 and ImageNet datasets, RADARS can achieve up to 3.41% higher accuracy with 2.5X search time reduction compared with a state-of-the-art RL-based method, while the two DNAS baselines cannot complete due to excessive memory usage or search time. To the best of the authors' knowledge, this is the first DNAS framework that can handle large search spaces with bounded memory usage.