Data-Efficient Image Recognition with Contrastive Predictive Coding
Olivier J. Hénaff Aravind Srinivas Jeffrey De Fauw Ali Razavi Carl Doersch S. M. Ali Eslami Aaron van den Oord
Abstract
Human observers can learn to recognize new categories of images from a handful of examples, yet doing so with artificial ones remains an open challenge. We hypothesize that data-efficient recognition is enabled by representations which make the variability in natural signals more predictable. We therefore revisit and improve Contrastive Predictive Coding, an unsupervised objective for learning such representations. This new implementation produces features which support state-of-the-art linear classification accuracy on the ImageNet dataset. When used as input for non-linear classification with deep neural networks, this representation allows us to use 2–5 $\times$ less labels than classifiers trained directly on image pixels. Finally, this unsupervised representation substantially improves transfer learning to object detection on the PASCAL VOC dataset, surpassing fully supervised pre-trained ImageNet classifiers.
中文速览
用自监督方式预训练的视觉表示（representation）能否大幅降低图像识别所需的标注数据量，是人工智能领域长期未解决的难题。研究者基于"好的表示应能让自然信号的变化变得更可预测"这一思路，对对比预测编码（Contrastive Predictive Coding，CPC）方法的网络架构和训练流程进行了系统性改进，将其在 ImageNet 线性分类上的准确率从 48.7% 大幅提升至 71.5%。在此表示之上接一个非线性分类器后，仅用 1% 的 ImageNet 标注数据就能达到 78% 的 Top-5 准确率，比直接在像素上训练提高了 34 个百分点，也超过了此前所有半监督方法；即便使用全量标注，该模型仍比监督训练的 ResNet 高出 3.2%。更关键的是，这种无监督表示迁移到 PASCAL VOC 目标检测任务时同样超越了有监督的 ImageNet 预训练模型，说明可预测性表示不仅省标注，还具备更强的泛化能力，为弥合人类与机器在少样本学习上的巨大差距提供了切实可行的方向。
原文 arXiv:1905.09272;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1905.09272v3