Deep Co-Training for Semi-Supervised Image Recognition

By Siyuan Qiao, Wei Shen, Zhishuai Zhang, Alan L. Yuille. 2018

539 citations recorded by the source corpus.

In this paper, we study the problem of semi-supervised image recognition, which is to learn classifiers using both labeled and unlabeled images. We present Deep Co-Training, a deep learning based method inspired by the Co-Training framework. The original Co-Training learns two classifiers on two views which are data from different sources that describe the same instances. To extend this concept to deep learning, Deep Co-Training trains multiple deep neural networks to be the different views and exploits adversarial examples to encourage view difference, in order to prevent the networks from collapsing into each other. As a result, the co-trained networks provide different and complementary information about the data, which is necessary for the Co-Training framework to achieve good results. We test our method on SVHN, CIFAR-10/100 and ImageNet datasets, and our method outperforms the previous state-of-the-art methods by a large margin.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Self-Training With Noisy Student Improves ImageNet Classification (2019, background): “Apart from self-training, another important line of work in semi-supervised learning [9, 85] is based on consistency training [6, 4, 53, 36, 70, 45, 41, 51, 10, 12, 49, 2, 38, 72, 74, 5, 81].”
Weight Standardization (2019, background): “Deep neural networks achieve state-of-the-arts in many vision tasks [4, 14, 20, 25, 31, 32, 39, 46].”
A soft nearest-neighbor framework for continual semi-supervised learning (2022, neutral): “Other approaches demonstrated the benefit of co-training [37] and distillation [55].”
Adversarial Examples Improve Image Recognition (2019, background): “VAT [26] and deep co-training [30] attempt to utilize adversarial examples in semi-supervised settings, but they require enormous extra unlabeled images.”
Audio-Visual Instance Discrimination with Cross-Modal Agreement (2020, background): “Classical approaches can be broadly categorized in co-training procedures [6, 7, 38, 45, 63, 81] or semi-supervised methods [49, 66] which seek to maximize the mutual agreement of two views of unlabeled data, multiple kernel learning procedures [5, 35, 39] which naturally model different views by di”
Guided Collaborative Training for Pixel-wise Semi-Supervised Learning (2020, methodology): “The consistentbased methods [23,35,41] utilize data augmentations to alter the inputs.”
Label Propagation for Deep Semi-Supervised Learning (2019, background): “[29], while complementarity of the two networks is enforced via adversarial examples.”
Semi-Supervised Semantic Segmentation With Cross-Consistency Training (2020, methodology): “The current dominant SSL methods in deep learning are consistency training [43, 29, 50, 36], pseudo labeling [30], entropy minimization [17] and bootstrapping [42].”
Revisiting Weak-to-Strong Consistency in Semi-Supervised Semantic Segmentation (2022, background): “Similar to the core spirit of co-training [7, 52] and mutual learning [76], CPS [12] introduces dual independent models to supervise each other.”
Pseudo-Labeling and Confirmation Bias in Deep Semi-Supervised Learning (2019, methodology): “Co-training [21] uses two (or more) networks trained simultaneously to agree on their predictions (consistency regularization) and disagree on their errors.”
A Survey on Deep Semi-Supervised Learning (2021, methodology): “To address this problem, [217] forces to add the View Difference Constraint to the previous Co-training model, and formulated as: where X (cid:48) denotes the adversarial examples of X , thus X (cid:48) ∩ X = ∅ .”
An Overview of Deep Semi-Supervised Learning (2020, methodology): “[119] used adversarial perturbations to produce new views for deep co-training for image classification, where the models are encouraged to have the same predictions on Dl but make different errors when they are exposed to adversarial attacks.”
Unbiased Teacher for Semi-Supervised Object Detection (2021, background): “…along the adversarial direction (Miyato et al., 2018; Yu et al., 2019), utilize multiple networks to generate various views of the same input data (Qiao et al., 2018), mix input data to generate augmented training data and labels (Zhang et al., 2018; Yun et al., 2019; Guo et al., 2019; Hendrycks…”
Annotation-efficient deep learning for automatic medical image segmentation (2020, methodology): “Co-training is one of the most prevalent methods for SSL 23,24 that works by training two classifiers for two complementary views using the labeled data, generating pseudo-labels for unlabeled data by enforcing agreement between the classifier predictions, and combining the labeled and…”
Learning with Limited Annotations: A Survey on Deep Semi-Supervised Learning for Medical Image Segmentation (2022, background): “[80], [81] generate adversarial examples as another view.”
Neural Rejuvenation: Improving Deep Network Training by Enhancing Computational Resource Utilization (2018, background): “They are not included in Table 2 because their image size is 128x128 and the learning rate follows the cosine learning rate schedule starting from 0.1 [48].”
Efficient Semi-supervised Gross Target Volume of Nasopharyngeal Carcinoma Segmentation via Uncertainty Rectified Pyramid Consistency (2020, background): “For example, the co-training-based methods need to train several models at the same time [20], and the uncertainty estimation-based frameworks need multiple forward passes [27].”