AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transformations rather than Data
Liheng Zhang ♭ , Guo-Jun Qi ♭♮ , Liqiang Wang♯ , Jiebo Luo§ ♭Laboratory for MAchine Perception and LEarning (MAPLE) http://maple-lab.net/ ♮Huawei Cloud, ♯University of Central Florida, §University of Rochester https://github.com/maple-research-lab/AET The work was done while L. Zhang was interning at Huawei Cloud.Corresponding author: G.-J. Qi, email:
Abstract
The success of deep neural networks often relies on a large amount of labeled examples, which can be difficult to obtain in many real scenarios. To address this challenge, unsupervised methods are strongly preferred for training neural networks without using any labeled data. In this paper, we present a novel paradigm of unsupervised representation learning by Auto-Encoding Transformation (AET) in contrast to the conventional Auto-Encoding Data (AED) approach. Given a randomly sampled transformation, AET seeks to predict it merely from the encoded features as accurately as possible at the output end. The idea is the following: as long as the unsupervised features successfully encode the essential information about the visual structures of original and transformed images, the transformation can be well predicted. We will show that this AET paradigm allows us to instantiate a large variety of transformations, from parameterized, to non-parameterized and GAN-induced ones. Our experiments show that AET greatly improves over existing unsupervised approaches, setting new state-of-the-art performances being greatly closer to the upper bounds by their fully supervised counterparts on CIFAR
中文速览
当标注数据难以获取时，如何让神经网络学到有用的视觉特征，是计算机视觉领域的核心难题。研究者提出了一种新的无监督表示学习框架——自编码变换（Auto-Encoding Transformation，AET），其核心思路与传统"重建图像像素"的自编码器截然不同：随机对图像施加一个变换，然后让网络仅凭原图和变换后图像的编码特征，尽可能准确地"猜出"这个变换是什么。这一目标倒逼编码器去捕捉图像真正的视觉结构，而非死记硬背像素细节。AET框架高度灵活，可容纳仿射、投影等参数化几何变换、GAN生成的语义级变换以及各类非参数化变换。在CIFAR-10、ImageNet和Places等标准基准上，AET大幅超越此前最优无监督方法，性能显著逼近有监督训练的上界，为无标注场景下的深度特征学习提供了一条更有效的新路径。
原文 arXiv:1901.04596;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1901.04596v2