Solving Inefficiency of Self-supervised Representation Learning

Self-supervised learning has attracted great interest due to its tremendous potentials in learning discriminative representations in an unsupervised manner. Along this direction, contrastive learning achieves current state-of-the-art performance. Despite the acknowledged successes, existing contrastive learning methods suffer from very low learning efficiency, e.g., taking about ten times more training epochs than supervised learning for comparable recognition accuracy. In this paper, we discover two contradictory phenomena in contrastive learning that we call under-clustering and over-clustering problems, which are major obstacles to learning efficiency. Under-clustering means that the model cannot efficiently learn to discover the dissimilarity between inter-class samples when the negative sample pairs for contrastive learning are insufficient to differentiate all the actual object categories. Over-clustering implies that the model cannot efficiently learn the feature representation from excessive negative sample pairs, which include many outliers and thus enforce the model to over-cluster samples of the same actual categories into different clusters. To simultaneously overcome these two problems, we propose a novel self-supervised learning framework using a median triplet loss. Precisely, we employ a triplet loss tending to maximize the relative distance between the positive pair and negative pairs to address the under-clustering problem; and we construct the negative pair by selecting the negative sample of a median similarity score from all negative samples to avoid the over-clustering problem, guaranteed by the Bernoulli Distribution model. We extensively evaluate our proposed framework in several large-scale benchmarks (e.g., ImageNet, SYSU-30k, and COCO). The results demonstrate the superior performance of our model over the latest state-of-the-art methods by a clear margin.

Joint Learning of Neural Transfer and Architecture Adaptation for Image Recognition

Current state-of-the-art visual recognition systems usually rely on the following pipeline: (a) pretraining a neural network on a large-scale dataset (e.g., ImageNet) and (b) finetuning the network weights on a smaller, task-specific dataset. Such a pipeline assumes the sole weight adaptation is able to transfer the network capability from one domain to another domain, based on a strong assumption that a fixed architecture is appropriate for all domains. However, each domain with a distinct recognition target may need different levels/paths of feature hierarchy, where some neurons may become redundant, and some others are re-activated to form new network structures. In this work, we prove that dynamically adapting network architectures tailored for each domain task along with weight finetuning benefits in both efficiency and effectiveness, compared to the existing image recognition pipeline that only tunes the weights regardless of the architecture. Our method can be easily generalized to an unsupervised paradigm by replacing supernet training with self-supervised learning in the source domain tasks and performing linear evaluation in the downstream tasks. This further improves the search efficiency of our method. Moreover, we also provide principled and empirical analysis to explain why our approach works by investigating the ineffectiveness of existing neural architecture search. We find that preserving the joint distribution of the network architecture and weights is of importance. This analysis not only benefits image recognition but also provides insights for crafting neural networks. Experiments on five representative image recognition tasks such as person re-identification, age estimation, gender recognition, image classification, and unsupervised domain adaptation demonstrate the effectiveness of our method.

Learning Class-Agnostic Pseudo Mask Generation for Box-Supervised Semantic Segmentation

Recently, several weakly supervised learning methods have been devoted to utilize bounding box supervision for training deep semantic segmentation models. Most existing methods usually leverage the generic proposal generators (\eg, dense CRF and MCG) to produce enhanced segmentation masks for further training segmentation models. These proposal generators, however, are generic and not specifically designed for box-supervised semantic segmentation, thereby leaving some leeway for improving segmentation performance. In this paper, we aim at seeking for a more accurate learning-based class-agnostic pseudo mask generator tailored to box-supervised semantic segmentation. To this end, we resort to a pixel-level annotated auxiliary dataset where the class labels are non-overlapped with those of the box-annotated dataset. For learning pseudo mask generator from the auxiliary dataset, we present a bi-level optimization formulation. In particular, the lower subproblem is used to learn box-supervised semantic segmentation, while the upper subproblem is used to learn an optimal class-agnostic pseudo mask generator. The learned pseudo segmentation mask generator can then be deployed to the box-annotated dataset for improving weakly supervised semantic segmentation. Experiments on PASCAL VOC 2012 dataset show that the learned pseudo mask generator is effective in boosting segmentation performance, and our method can further close the performance gap between box-supervised and fully-supervised models. Our code will be made publicly available at https://github.com/Vious/LPG_BBox_Segmentation .

GTAE: Graph-Transformer based Auto-Encoders for Linguistic-Constrained Text Style Transfer

Non-parallel text style transfer has attracted increasing research interests in recent years. Despite successes in transferring the style based on the encoder-decoder framework, current approaches still lack the ability to preserve the content and even logic of original sentences, mainly due to the large unconstrained model space or too simplified assumptions on latent embedding space. Since language itself is an intelligent product of humans with certain grammars and has a limited rule-based model space by its nature, relieving this problem requires reconciling the model capacity of deep neural networks with the intrinsic model constraints from human linguistic rules. To this end, we propose a method called Graph Transformer based Auto Encoder (GTAE), which models a sentence as a linguistic graph and performs feature extraction and style transfer at the graph level, to maximally retain the content and the linguistic structure of original sentences. Quantitative experiment results on three non-parallel text style transfer tasks show that our model outperforms state-of-the-art methods in content preservation, while achieving comparable performance on transfer accuracy and sentence naturalness.

Temporal Contrastive Graph for Self-supervised Video Representation Learning

Attempt to fully explore the fine-grained temporal structure and global-local chronological characteristics for self-supervised video representation learning, this work takes a closer look at exploiting the temporal structure of videos and further proposes a novel self-supervised method named Temporal Contrastive Graph (TCG). In contrast to the existing methods that randomly shuffle the video frames or video snippets within a video, our proposed TCG roots in a hybrid graph contrastive learning strategy to regard the inter-snippet and intra-snippet temporal relationships as self-supervision signals for temporal representation learning. To increase the temporal diversity of features more comprehensively and precisely, our proposed TCG integrates the prior knowledge about the frame and snippet orders into temporal contrastive graph structures, i.e., the intra-/inter- snippet temporal contrastive graph modules. By randomly removing edges and masking node features of the intra-snippet graphs or inter-snippet graphs, our TCG can generate different correlated graph views. Then, specific contrastive losses are designed to maximize the agreement between node embeddings in different views. To learn the global context representation and recalibrate the channel-wise features adaptively, we introduce an adaptive video snippet order prediction module, which leverages the relational knowledge among video snippets to predict the actual snippet orders.Experimental results demonstrate the superiority of our TCG over the state-of-the-art methods on large-scale action recognition and video retrieval benchmarks.

Graphonomy: Universal Image Parsing via Graph Reasoning and Transfer

Prior highly-tuned image parsing models are usually studied in a certain domain with a specific set of semantic labels and can hardly be adapted into other scenarios (e.g., sharing discrepant label granularity) without extensive re-training. Learning a single universal parsing model by unifying label annotations from different domains or at various levels of granularity is a crucial but rarely addressed topic. This poses many fundamental learning challenges, e.g., discovering underlying semantic structures among different label granularity or mining label correlation across relevant tasks. To address these challenges, we propose a graph reasoning and transfer learning framework, named "Graphonomy", which incorporates human knowledge and label taxonomy into the intermediate graph representation learning beyond local convolutions. In particular, Graphonomy learns the global and structured semantic coherency in multiple domains via semantic-aware graph reasoning and transfer, enforcing the mutual benefits of the parsing across domains (e.g., different datasets or co-related tasks). The Graphonomy includes two iterated modules: Intra-Graph Reasoning and Inter-Graph Transfer modules. The former extracts the semantic graph in each domain to improve the feature representation learning by propagating information with the graph; the latter exploits the dependencies among the graphs from different domains for bidirectional knowledge transfer. We apply Graphonomy to two relevant but different image understanding research topics: human parsing and panoptic segmentation, and show Graphonomy can handle both of them well via a standard pipeline against current state-of-the-art approaches. Moreover, some extra benefit of our framework is demonstrated, e.g., generating the human parsing at various levels of granularity by unifying annotations across different datasets.

Unifying Relational Sentence Generation and Retrieval for Medical Image Report Composition

Beyond generating long and topic-coherent paragraphs in traditional captioning tasks, the medical image report composition task poses more task-oriented challenges by requiring both the highly-accurate medical term diagnosis and multiple heterogeneous forms of information including impression and findings. Current methods often generate the most common sentences due to dataset bias for individual case, regardless of whether the sentences properly capture key entities and relationships. Such limitations severely hinder their applicability and generalization capability in medical report composition where the most critical sentences lie in the descriptions of abnormal diseases that are relatively rare. Moreover, some medical terms appearing in one report are often entangled with each other and co-occurred, e.g. symptoms associated with a specific disease. To enforce the semantic consistency of medical terms to be incorporated into the final reports and encourage the sentence generation for rare abnormal descriptions, we propose a novel framework that unifies template retrieval and sentence generation to handle both common and rare abnormality while ensuring the semantic-coherency among the detected medical terms. Specifically, our approach exploits hybrid-knowledge co-reasoning: i) explicit relationships among all abnormal medical terms to induce the visual attention learning and topic representation encoding for better topic-oriented symptoms descriptions; ii) adaptive generation mode that changes between the template retrieval and sentence generation according to a contextual topic encoder. Experimental results on two medical report benchmarks demonstrate the superiority of the proposed framework in terms of both human and metrics evaluation.

REM-Net: Recursive Erasure Memory Network for Commonsense Evidence Refinement

When answering a question, people often draw upon their rich world knowledge in addition to the particular context. While recent works retrieve supporting facts/evidence from commonsense knowledge bases to supply additional information to each question, there is still ample opportunity to advance it on the quality of the evidence. It is crucial since the quality of the evidence is the key to answering commonsense questions, and even determines the upper bound on the QA systems performance. In this paper, we propose a recursive erasure memory network (REM-Net) to cope with the quality improvement of evidence. To address this, REM-Net is equipped with a module to refine the evidence by recursively erasing the low-quality evidence that does not explain the question answering. Besides, instead of retrieving evidence from existing knowledge bases, REM-Net leverages a pre-trained generative model to generate candidate evidence customized for the question. We conduct experiments on two commonsense question answering datasets, WIQA and CosmosQA. The results demonstrate the performance of REM-Net and show that the refined evidence is explainable.

AU-Expression Knowledge Constrained Representation Learning for Facial Expression Recognition

Recognizing human emotion/expressions automatically is quite an expected ability for intelligent robotics, as it can promote better communication and cooperation with humans. Current deep-learning-based algorithms may achieve impressive performance in some lab-controlled environments, but they always fail to recognize the expressions accurately for the uncontrolled in-the-wild situation. Fortunately, facial action units (AU) describe subtle facial behaviors, and they can help distinguish uncertain and ambiguous expressions. In this work, we explore the correlations among the action units and facial expressions, and devise an AU-Expression Knowledge Constrained Representation Learning (AUE-CRL) framework to learn the AU representations without AU annotations and adaptively use representations to facilitate facial expression recognition. Specifically, it leverages AU-expression correlations to guide the learning of the AU classifiers, and thus it can obtain AU representations without incurring any AU annotations. Then, it introduces a knowledge-guided attention mechanism that mines useful AU representations under the constraint of AU-expression correlations. In this way, the framework can capture local discriminative and complementary features to enhance facial representation for facial expression recognition. We conduct experiments on the challenging uncontrolled datasets to demonstrate the superiority of the proposed framework over current state-of-the-art methods.

Adversarial Meta Sampling for Multilingual Low-Resource Speech Recognition

Low-resource automatic speech recognition (ASR) is challenging, as the low-resource target language data cannot well train an ASR model. To solve this issue, meta-learning formulates ASR for each source language into many small ASR tasks and meta-learns a model initialization on all tasks from different source languages to access fast adaptation on unseen target languages. However, for different source languages, the quantity and difficulty vary greatly because of their different data scales and diverse phonological systems, which leads to task-quantity and task-difficulty imbalance issues and thus a failure of multilingual meta-learning ASR (MML-ASR). In this work, we solve this problem by developing a novel adversarial meta sampling (AMS) approach to improve MML-ASR. When sampling tasks in MML-ASR, AMS adaptively determines the task sampling probability for each source language. Specifically, for each source language, if the query loss is large, it means that its tasks are not well sampled to train ASR model in terms of its quantity and difficulty and thus should be sampled more frequently for extra learning. Inspired by this fact, we feed the historical task query loss of all source language domain into a network to learn a task sampling policy for adversarially increasing the current query loss of MML-ASR. Thus, the learnt task sampling policy can master the learning situation of each language and thus predicts good task sampling probability for each language for more effective learning. Finally, experiment results on two multilingual datasets show significant performance improvement when applying our AMS on MML-ASR, and also demonstrate the applicability of AMS to other low-resource speech tasks and transfer learning ASR approaches. Our codes are available at: https://github.com/iamxiaoyubei/AMS.