IRLAS: Inverse Reinforcement Learning for Architecture Search

By Minghao Guo, Zhaobai Zhong, Wei Wu, Dahua Lin, Junjie Yan. 2018

43 citations recorded by the source corpus.

In this paper, we propose an inverse reinforcement learning method for architecture search (IRLAS), which trains an agent to learn to search network structures that are topologically inspired by human-designed network. Most existing architecture search approaches totally neglect the topological characteristics of architectures, which results in complicated architecture with a high inference latency. Motivated by the fact that human-designed networks are elegant in topology with a fast inference speed, we propose a mirror stimuli function inspired by biological cognition theory to extract the abstract topological knowledge of an expert human-design network (ResNeXt). To avoid raising a too strong prior over the search space, we introduce inverse reinforcement learning to train the mirror stimuli function and exploit it as a heuristic guidance for architecture search, easily generalized to different architecture search algorithms. On CIFAR-10, the best architecture searched by our proposed IRLAS achieves 2.60% error rate. For ImageNet mobile setting, our model achieves a state-of-the-art top-1 accuracy 75.28%, while being 2~4x faster than most auto-generated architectures. A fast version of this model achieves 10% faster than MobileNetV2, while maintaining a higher accuracy.

Subfields: Computer Vision and Pattern Recognition, Materials Chemistry, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Materials Chemistry

In Molecular Biology

Citations in the corpus

GreedyNAS: Towards Fast One-Shot NAS With Greedy Supernet (2020, background): “In contrast, neural architecture search (NAS) aims at automatically learning the network architecture to further boost the performance for target tasks [10, 20, 43, 2, 19].”
Online Hyper-Parameter Learning for Auto-Augmentation Strategy (2019, methodology): “Several methods attempt to automatically search architectures with fast inference speed, either explicitly took the inference latency as a constrain [28], or implicitly encoded the topology information [11].”
Overcoming Multi-Model Forgetting in One-Shot NAS With Diversity Maximization (2020, background): “…numerous separate architectures from scratch and utilize an Evolutionary Algorithm (EA) or Reinforcement Learning (RL) to ﬁnd the most promising architectures based on validation accuracy [12, 38, 29], which are highly computational-expensive and impossible for most machine learning practitioners.”
Graph-Guided Architecture Search for Real-Time Semantic Segmentation (2019, methodology): “Most existing architecture search works are based on either reinforcement learning [52, 17] or evolutionary algorithm [37, 11].”
When NAS Meets Robustness: In Search of Robust Architectures Against Adversarial Attacks (2019, methodology): “Representative techniques include reinforcement learning [44, 45, 43, 10, 1], evolution [27, 31] and surrogate model [18], which have been widely adopted to the search process.”
One-Shot Neural Architecture Search: Maximising Diversity to Overcome Catastrophic Forgetting (2020, methodology): “91 Single path RL IRLAS-differential [19] 2.”
MTL-NAS: Task-Agnostic Neural Architecture Search Towards General-Purpose Multi-Task Learning (2020, methodology): “This is realized by various search algorithms including Bayesian optimization [3], reinforcement learning [18, 58, 72], evolutionary algorithms [49, 63], network transformation [11, 17], and gradient descent [1, 33, 61, 70].”
Teacher Guided Architecture Search (2018, methodology): “Several more recent methods [31, 29, 42, 1, 17] proposed to share the trainable parameters across all candidate net-”