[论文解读] AlphaX: eXploring Neural Architectures with Deep Neural Networks and Monte Carlo Tree Search

Linnan Wang, Yiyang Zhao+3 more|arXiv (Cornell University)|Mar 26, 2019

Advanced Neural Network Applications参考文献 57被引用 57

一句话总结

AlphaX 使用分布式蒙特卡罗树搜索配合 Meta-DNN 自动化神经架构搜索，在 CIFAR-10 和 ImageNet 上实现快速全局优化和强精度，在 NASBench-101 上具有可扩展性能。

ABSTRACT

Neural Architecture Search (NAS) has shown great success in automating the design of neural networks, but the prohibitive amount of computations behind current NAS methods requires further investigations in improving the sample efficiency and the network evaluation cost to get better results in a shorter time. In this paper, we present a novel scalable Monte Carlo Tree Search (MCTS) based NAS agent, named AlphaX, to tackle these two aspects. AlphaX improves the search efficiency by adaptively balancing the exploration and exploitation at the state level, and by a Meta-Deep Neural Network (DNN) to predict network accuracies for biasing the search toward a promising region. To amortize the network evaluation cost, AlphaX accelerates MCTS rollouts with a distributed design and reduces the number of epochs in evaluating a network by transfer learning guided with the tree structure in MCTS. In 12 GPU days and 1000 samples, AlphaX found an architecture that reaches 97.84\% top-1 accuracy on CIFAR-10, and 75.5\% top-1 accuracy on ImageNet, exceeding SOTA NAS methods in both the accuracy and sampling efficiency. Particularly, we also evaluate AlphaX on NASBench-101, a large scale NAS dataset; AlphaX is 3x and 2.8x more sample efficient than Random Search and Regularized Evolution in finding the global optimum. Finally, we show the searched architecture improves a variety of vision applications from Neural Style Transfer, to Image Captioning and Object Detection.

研究动机与目标

• Motivate reducing human manual effort in neural architecture design and accelerate NAS search efficiency.
• Propose a scalable NAS framework that combines MCTS with a predictive Meta-DNN to guide sampling and evaluation.
• Demonstrate end-to-end learning of Meta-DNN using data generated during architecture search.
• Show transfer learning to accelerate network evaluations and enable distributed, large-scale search.

提出的方法

• Use Monte Carlo Tree Search to guide architecture search with online statistics N(s,a) and Q(s,a).
• Integrate Meta-DNN to predict architecture performance and speed up rollouts; update Meta-DNN with data gathered during search (end-to-end learning).
• Encode NASNet and NASBench architectures into fixed-length vectors for Meta-DNN input.
• Incorporate transfer learning by reusing weights from parent architectures when training child architectures.
• Distribute the search across a master scheduler and multiple GPU trainers, with a two-phase backpropagation using predicted and true accuracies.

实验结果

研究问题

• RQ1 Can a scalable MCTS-based agent outperform random search and greedy methods in NAS tasks?
• RQ2 Does a Meta-DNN accurately predict unseen architecture performance to accelerate search?
• RQ3 How does transfer learning affect evaluation speed and final architecture quality in NAS?
• RQ4 What is the impact of distributed training on NAS search efficiency and scalability?

主要发现

• AlphaX is up to 3x faster than Random Search and Regularized Evolution in finding the best architecture on NASBench-101.
• In CIFAR-10, AlphaX achieves high accuracy and demonstrates competitive results on large-scale datasets like ImageNet.
• The Meta-DNN improves search efficiency by providing accurate performance estimates and enabling more informed rollouts.
• Transfer learning significantly reduces epochs required to reach target accuracies for child architectures.
• Distributed AlphaX enables scalable NAS with a master node and multiple GPU clients, maintaining/searching effectively across large spaces.