XferNAS: Transfer Neural Architecture Search

The term Neural Architecture Search (NAS) refers to the automatic optimization of network architectures for a new, previously unknown task. Since testing an architecture is computationally very expensive, many optimizers need days or even weeks find suitable architectures. However, this search time can be significantly reduced if knowledge from previous searches on different tasks reused. In work, we propose generally applicable framework that introduces only minor changes existing leverage feature. As example, select optimizer and demonstrate complexity integration as well its impact. experiments CIFAR-10 CIFAR-100, observe reduction in 200 6 GPU days, speed up by factor 33. addition, new records 1.99 14.06 NAS CIFAR benchmarks, respectively. separate study, analyze impact amount source target data. Empirically, proposed gives better results and, worst case, just good unmodified optimizer.