Noise robustness is essential for deploying automatic speech recognition (ASR) systems in real-world environments. One way to reduce the effect of noise interference is to employ a preprocessing module that conducts speech enhancement, and then feed the enhanced speech to an ASR backend. In this work, instead of suppressing background noise with a conventional cascaded pipeline, we employ a noise-robust representation learned by a refined self-supervised framework for noisy speech recognition. We propose to combine a reconstruction module with contrastive learning and perform multi-task continual pre-training on noisy data. The reconstruction module is used for auxiliary learning to improve the noise robustness of the learned representation and thus is not required during inference. Experiments demonstrate the effectiveness of our proposed method. Our model substantially reduces the word error rate (WER) for the synthesized noisy LibriSpeech test sets, and yields around 4.1/7.5% WER reduction on noisy clean/other test sets compared to data augmentation. For the real-world noisy speech from the CHiME-4 challenge (1-channel track), we have obtained the state of the art ASR performance without any denoising front-end. Moreover, we achieve comparable performance to the best supervised approach reported with only 16% of labeled data.
Radiology reports are unstructured and contain the imaging findings and corresponding diagnoses transcribed by radiologists which include clinical facts and negated and/or uncertain statements. Extracting pathologic findings and diagnoses from radiology reports is important for quality control, population health, and monitoring of disease progress. Existing works, primarily rely either on rule-based systems or transformer-based pre-trained model fine-tuning, but could not take the factual and uncertain information into consideration, and therefore generate false-positive outputs. In this work, we introduce three sedulous augmentation techniques which retain factual and critical information while generating augmentations for contrastive learning. We introduce RadBERT-CL, which fuses these information into BlueBert via a self-supervised contrastive loss. Our experiments on MIMIC-CXR show superior performance of RadBERT-CL on fine-tuning for multi-class, multi-label report classification. We illustrate that when few labeled data are available, RadBERT-CL outperforms conventional SOTA transformers (BERT/BlueBert) by significantly larger margins (6-11%). We also show that the representations learned by RadBERT-CL can capture critical medical information in the latent space.
Unsupervised domain adaptation which aims to adapt models trained on a labeled source domain to a completely unlabeled target domain has attracted much attention in recent years. While many domain adaptation techniques have been proposed for images, the problem of unsupervised domain adaptation in videos remains largely underexplored. In this paper, we introduce Contrast and Mix (CoMix), a new contrastive learning framework that aims to learn discriminative invariant feature representations for unsupervised video domain adaptation. First, unlike existing methods that rely on adversarial learning for feature alignment, we utilize temporal contrastive learning to bridge the domain gap by maximizing the similarity between encoded representations of an unlabeled video at two different speeds as well as minimizing the similarity between different videos played at different speeds. Second, we propose a novel extension to the temporal contrastive loss by using background mixing that allows additional positives per anchor, thus adapting contrastive learning to leverage action semantics shared across both domains. Moreover, we also integrate a supervised contrastive learning objective using target pseudo-labels to enhance discriminability of the latent space for video domain adaptation. Extensive experiments on several benchmark datasets demonstrate the superiority of our proposed approach over state-of-the-art methods. Project page: https://cvir.github.io/projects/comix
Visual object tracking performance has been dramatically improved in recent years, but some severe challenges remain open, like distractors and occlusions. We suspect the reason is that the feature representations of the tracking targets are only expressively learned but not fully discriminatively modeled. In this paper, we propose a novel Instance-Aware Tracker (IAT) to explicitly excavate the discriminability of feature representations, which improves the classical visual tracking pipeline with an instance-level classifier. First, we introduce a contrastive learning mechanism to formulate the classification task, ensuring that every training sample could be uniquely modeled and be highly distinguishable from plenty of other samples. Besides, we design an effective negative sample selection scheme to contain various intra and inter classes in the instance classification branch. Furthermore, we implement two variants of the proposed IAT, including a video-level one and an object-level one. They realize the concept of \textbf{instance} in different granularity as videos and target bounding boxes, respectively. The former enhances the ability to recognize the target from the background while the latter boosts the discriminative power for mitigating the target-distractor dilemma. Extensive experimental evaluations on 8 benchmark datasets show that both two versions of the proposed IAT achieve leading results against state-of-the-art methods while running at 30FPS. Code will be available when it is published.
SARS-CoV-2 has presented direct and indirect challenges to the scientific community. One of the most prominent indirect challenges advents from the mandatory use of face masks in a large number of countries. Face recognition methods struggle to perform identity verification with similar accuracy on masked and unmasked individuals. It has been shown that the performance of these methods drops considerably in the presence of face masks, especially if the reference image is unmasked. We propose FocusFace, a multi-task architecture that uses contrastive learning to be able to accurately perform masked face recognition. The proposed architecture is designed to be trained from scratch or to work on top of state-of-the-art face recognition methods without sacrificing the capabilities of a existing models in conventional face recognition tasks. We also explore different approaches to design the contrastive learning module. Results are presented in terms of masked-masked (M-M) and unmasked-masked (U-M) face verification performance. For both settings, the results are on par with published methods, but for M-M specifically, the proposed method was able to outperform all the solutions that it was compared to. We further show that when using our method on top of already existing methods the training computational costs decrease significantly while retaining similar performances. The implementation and the trained models are available at GitHub.
Computer-aided diagnosis plays a salient role in more accessible and accurate cardiopulmonary diseases classification and localization on chest radiography. Millions of people get affected and die due to these diseases without an accurate and timely diagnosis. Recently proposed contrastive learning heavily relies on data augmentation, especially positive data augmentation. However, generating clinically-accurate data augmentations for medical images is extremely difficult because the common data augmentation methods in computer vision, such as sharp, blur, and crop operations, can severely alter the clinical settings of medical images. In this paper, we proposed a novel and simple data augmentation method based on patient metadata and supervised knowledge to create clinically accurate positive and negative augmentations for chest X-rays. We introduce an end-to-end framework, SCALP, which extends the self-supervised contrastive approach to a supervised setting. Specifically, SCALP pulls together chest X-rays from the same patient (positive keys) and pushes apart chest X-rays from different patients (negative keys). In addition, it uses ResNet-50 along with the triplet-attention mechanism to identify cardiopulmonary diseases, and Grad-CAM++ to highlight the abnormal regions. Our extensive experiments demonstrate that SCALP outperforms existing baselines with significant margins in both classification and localization tasks. Specifically, the average classification AUCs improve from 82.8% (SOTA using DenseNet-121) to 83.9% (SCALP using ResNet-50), while the localization results improve on average by 3.7% over different IoU thresholds.
Graph representation learning is crucial for many real-world applications (e.g. social relation analysis). A fundamental problem for graph representation learning is how to effectively learn representations without human labeling, which is usually costly and time-consuming. Graph contrastive learning (GCL) addresses this problem by pulling the positive node pairs (or similar nodes) closer while pushing the negative node pairs (or dissimilar nodes) apart in the representation space. Despite the success of the existing GCL methods, they primarily sample node pairs based on the node-level proximity yet the community structures have rarely been taken into consideration. As a result, two nodes from the same community might be sampled as a negative pair. We argue that the community information should be considered to identify node pairs in the same communities, where the nodes insides are semantically similar. To address this issue, we propose a novel Graph Communal Contrastive Learning (gCooL) framework to jointly learn the community partition and learn node representations in an end-to-end fashion. Specifically, the proposed gCooL consists of two components: a Dense Community Aggregation (DeCA) algorithm for community detection and a Reweighted Self-supervised Cross-contrastive (ReSC) training scheme to utilize the community information. Additionally, the real-world graphs are complex and often consist of multiple views. In this paper, we demonstrate that the proposed gCooL can also be naturally adapted to multiplex graphs. Finally, we comprehensively evaluate the proposed gCooL on a variety of real-world graphs. The experimental results show that the gCooL outperforms the state-of-the-art methods.
A good visual representation is an inference map from observations (images) to features (vectors) that faithfully reflects the hidden modularized generative factors (semantics). In this paper, we formulate the notion of "good" representation from a group-theoretic view using Higgins' definition of disentangled representation, and show that existing Self-Supervised Learning (SSL) only disentangles simple augmentation features such as rotation and colorization, thus unable to modularize the remaining semantics. To break the limitation, we propose an iterative SSL algorithm: Iterative Partition-based Invariant Risk Minimization (IP-IRM), which successfully grounds the abstract semantics and the group acting on them into concrete contrastive learning. At each iteration, IP-IRM first partitions the training samples into two subsets that correspond to an entangled group element. Then, it minimizes a subset-invariant contrastive loss, where the invariance guarantees to disentangle the group element. We prove that IP-IRM converges to a fully disentangled representation and show its effectiveness on various benchmarks. Codes are available at https://github.com/Wangt-CN/IP-IRM.
Unsupervised representation learning (URL), which learns compact embeddings of high-dimensional data without supervision, has made remarkable progress recently. However, the development of URLs for different requirements is independent, which limits the generalization of the algorithms, especially prohibitive as the number of tasks grows. For example, dimension reduction methods, t-SNE, and UMAP optimize pair-wise data relationships by preserving the global geometric structure, while self-supervised learning, SimCLR, and BYOL focus on mining the local statistics of instances under specific augmentations. To address this dilemma, we summarize and propose a unified similarity-based URL framework, GenURL, which can smoothly adapt to various URL tasks. In this paper, we regard URL tasks as different implicit constraints on the data geometric structure that help to seek optimal low-dimensional representations that boil down to data structural modeling (DSM) and low-dimensional transformation (LDT). Specifically, DMS provides a structure-based submodule to describe the global structures, and LDT learns compact low-dimensional embeddings with given pretext tasks. Moreover, an objective function, General Kullback-Leibler divergence (GKL), is proposed to connect DMS and LDT naturally. Comprehensive experiments demonstrate that GenURL achieves consistent state-of-the-art performance in self-supervised visual learning, unsupervised knowledge distillation (KD), graph embeddings (GE), and dimension reduction.
During natural disasters, aircraft and satellites are used to survey the impacted regions. Usually human experts are needed to manually label the degrees of the building damage so that proper humanitarian assistance and disaster response (HADR) can be achieved, which is labor-intensive and time-consuming. Expecting human labeling of major disasters over a wide area gravely slows down the HADR efforts. It is thus of crucial interest to take advantage of the cutting-edge Artificial Intelligence and Machine Learning techniques to speed up the natural infrastructure damage assessment process to achieve effective HADR. Accordingly, the paper demonstrates a systematic effort to achieve efficient building damage classification. First, two novel generative adversarial nets (GANs) are designed to augment data used to train the deep-learning-based classifier. Second, a contrastive learning based method using novel data structures is developed to achieve great performance. Third, by using information fusion, the classifier is effectively trained with very few training data samples for transfer learning. All the classifiers are small enough to be loaded in a smart phone or simple laptop for first responders. Based on the available overhead imagery dataset, results demonstrate data and computational efficiency with 10% of the collected data combined with a GAN reducing the time of computation from roughly half a day to about 1 hour with roughly similar classification performances.
Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or domain shift. In this paper, we show that by generating carefully designed negative samples, contrastive learning can learn more robust representations with less dependence on such features. Contrastive learning utilizes positive pairs that preserve semantic information while perturbing superficial features in the training images. Similarly, we propose to generate negative samples in a reversed way, where only the superfluous instead of the semantic features are preserved. We develop two methods, texture-based and patch-based augmentations, to generate negative samples. These samples achieve better generalization, especially under out-of-domain settings. We also analyze our method and the generated texture-based samples, showing that texture features are indispensable in classifying particular ImageNet classes and especially finer classes. We also show that model bias favors texture and shape features differently under different test settings. Our code, trained models, and ImageNet-Texture dataset can be found at https://github.com/SongweiGe/Contrastive-Learning-with-Non-Semantic-Negatives.
Recent contrastive methods show significant improvement in self-supervised learning in several domains. In particular, contrastive methods are most effective where data augmentation can be easily constructed e.g. in computer vision. However, they are less successful in domains without established data transformations such as time series data. In this paper, we propose a novel self-supervised learning framework that combines contrastive learning with neural processes. It relies on recent advances in neural processes to perform time series forecasting. This allows to generate augmented versions of data by employing a set of various sampling functions and, hence, avoid manually designed augmentations. We extend conventional neural processes and propose a new contrastive loss to learn times series representations in a self-supervised setup. Therefore, unlike previous self-supervised methods, our augmentation pipeline is task-agnostic, enabling our method to perform well across various applications. In particular, a ResNet with a linear classifier trained using our approach is able to outperform state-of-the-art techniques across industrial, medical and audio datasets improving accuracy over 10% in ECG periodic data. We further demonstrate that our self-supervised representations are more efficient in the latent space, improving multiple clustering indexes and that fine-tuning our method on 10% of labels achieves results competitive to fully-supervised learning.
Recently, motivated by real applications, a major research direction in graph neural networks (GNNs) is to explore deeper structures. For instance, the graph connectivity is not always consistent with the label distribution (e.g., the closest neighbors of some nodes are not from the same category). In this case, GNNs need to stack more layers, in order to find the same categorical neighbors in a longer path for capturing the class-discriminative information. However, two major problems hinder the deeper GNNs to obtain satisfactory performance, i.e., vanishing gradient and over-smoothing. On one hand, stacking layers makes the neural network hard to train as the gradients of the first few layers vanish. Moreover, when simply addressing vanishing gradient in GNNs, we discover the shading neighbors effect (i.e., stacking layers inappropriately distorts the non-IID information of graphs and degrade the performance of GNNs). On the other hand, deeper GNNs aggregate much more information from common neighbors such that individual node representations share more overlapping features, which makes the final output representations not discriminative (i.e., overly smoothed). In this paper, for the first time, we address both problems to enable deeper GNNs, and propose Deeper-GXX, which consists of the Weight-Decaying Graph Residual Connection module (WDG-ResNet) and Topology-Guided Graph Contrastive Loss (TGCL). Extensive experiments on real-world data sets demonstrate that Deeper-GXX outperforms state-of-the-art deeper baselines.
We consider the problem of obtaining image quality representations in a self-supervised manner. We use prediction of distortion type and degree as an auxiliary task to learn features from an unlabeled image dataset containing a mixture of synthetic and realistic distortions. We then train a deep Convolutional Neural Network (CNN) using a contrastive pairwise objective to solve the auxiliary problem. We refer to the proposed training framework and resulting deep IQA model as the CONTRastive Image QUality Evaluator (CONTRIQUE). During evaluation, the CNN weights are frozen and a linear regressor maps the learned representations to quality scores in a No-Reference (NR) setting. We show through extensive experiments that CONTRIQUE achieves competitive performance when compared to state-of-the-art NR image quality models, even without any additional fine-tuning of the CNN backbone. The learned representations are highly robust and generalize well across images afflicted by either synthetic or authentic distortions. Our results suggest that powerful quality representations with perceptual relevance can be obtained without requiring large labeled subjective image quality datasets. The implementations used in this paper are available at \url{https://github.com/pavancm/CONTRIQUE}.
Intent classification (IC) and slot filling (SF) are critical building blocks in task-oriented dialogue systems. These two tasks are closely-related and can flourish each other. Since only a few utterances can be utilized for identifying fast-emerging new intents and slots, data scarcity issue often occurs when implementing IC and SF. However, few IC/SF models perform well when the number of training samples per class is quite small. In this paper, we propose a novel explicit-joint and supervised-contrastive learning framework for few-shot intent classification and slot filling. Its highlights are as follows. (i) The model extracts intent and slot representations via bidirectional interactions, and extends prototypical network to achieve explicit-joint learning, which guarantees that IC and SF tasks can mutually reinforce each other. (ii) The model integrates with supervised contrastive learning, which ensures that samples from same class are pulled together and samples from different classes are pushed apart. In addition, the model follows a not common but practical way to construct the episode, which gets rid of the traditional setting with fixed way and shot, and allows for unbalanced datasets. Extensive experiments on three public datasets show that our model can achieve promising performance.
Existed pre-trained models have achieved state-of-the-art performance on various text classification tasks. These models have proven to be useful in learning universal language representations. However, the semantic discrepancy between similar texts cannot be effectively distinguished by advanced pre-trained models, which have a great influence on the performance of hard-to-distinguish classes. To address this problem, we propose a novel Contrastive Learning with Label Distance (CLLD) in this work. Inspired by recent advances in contrastive learning, we specifically design a classification method with label distance for learning contrastive classes. CLLD ensures the flexibility within the subtle differences that lead to different label assignments, and generates the distinct representations for each class having similarity simultaneously. Extensive experiments on public benchmarks and internal datasets demonstrate that our method improves the performance of pre-trained models on classification tasks. Importantly, our experiments suggest that the learned label distance relieve the adversarial nature of interclasses.
A general-purpose robot should be able to master a wide range of tasks and quickly learn a novel one by leveraging past experiences. One-shot imitation learning (OSIL) approaches this goal by training an agent with (pairs of) expert demonstrations, such that at test time, it can directly execute a new task from just one demonstration. However, so far this framework has been limited to training on many variations of one task, and testing on other unseen but similar variations of the same task. In this work, we push for a higher level of generalization ability by investigating a more ambitious multi-task setup. We introduce a diverse suite of vision-based robot manipulation tasks, consisting of 7 tasks, a total of 61 variations, and a continuum of instances within each variation. For consistency and comparison purposes, we first train and evaluate single-task agents (as done in prior few-shot imitation work). We then study the multi-task setting, where multi-task training is followed by (i) one-shot imitation on variations within the training tasks, (ii) one-shot imitation on new tasks, and (iii) fine-tuning on new tasks. Prior state-of-the-art, while performing well within some single tasks, struggles in these harder multi-task settings. To address these limitations, we propose MOSAIC (Multi-task One-Shot Imitation with self-Attention and Contrastive learning), which integrates a self-attention model architecture and a temporal contrastive module to enable better task disambiguation and more robust representation learning. Our experiments show that MOSAIC outperforms prior state of the art in learning efficiency, final performance, and learns a multi-task policy with promising generalization ability via fine-tuning on novel tasks.
Despite being able to capture a range of features of the data, high accuracy models trained with supervision tend to make similar predictions. This seemingly implies that high-performing models share similar biases regardless of training methodology, which would limit ensembling benefits and render low-accuracy models as having little practical use. Against this backdrop, recent work has developed quite different training techniques, such as large-scale contrastive learning, yielding competitively high accuracy on generalization and robustness benchmarks. This motivates us to revisit the assumption that models necessarily learn similar functions. We conduct a large-scale empirical study of models across hyper-parameters, architectures, frameworks, and datasets. We find that model pairs that diverge more in training methodology display categorically different generalization behavior, producing increasingly uncorrelated errors. We show these models specialize in subdomains of the data, leading to higher ensemble performance: with just 2 models (each with ImageNet accuracy ~76.5%), we can create ensembles with 83.4% (+7% boost). Surprisingly, we find that even significantly low-accuracy models can be used to improve high-accuracy models. Finally, we show diverging training methodology yield representations that capture overlapping (but not supersetting) feature sets which, when combined, lead to increased downstream performance.
Recent empirical studies show that adversarial topic models (ATM) can successfully capture semantic patterns of the document by differentiating a document with another dissimilar sample. However, utilizing that discriminative-generative architecture has two important drawbacks: (1) the architecture does not relate similar documents, which has the same document-word distribution of salient words; (2) it restricts the ability to integrate external information, such as sentiments of the document, which has been shown to benefit the training of neural topic model. To address those issues, we revisit the adversarial topic architecture in the viewpoint of mathematical analysis, propose a novel approach to re-formulate discriminative goal as an optimization problem, and design a novel sampling method which facilitates the integration of external variables. The reformulation encourages the model to incorporate the relations among similar samples and enforces the constraint on the similarity among dissimilar ones; while the sampling method, which is based on the internal input and reconstructed output, helps inform the model of salient words contributing to the main topic. Experimental results show that our framework outperforms other state-of-the-art neural topic models in three common benchmark datasets that belong to various domains, vocabulary sizes, and document lengths in terms of topic coherence.
User Modeling plays an essential role in industry. In this field, task-agnostic approaches, which generate general-purpose representation applicable to diverse downstream user cognition tasks, is a promising direction being more valuable and economical than task-specific representation learning. With the rapid development of Internet service platforms, user behaviors have been accumulated continuously. However, existing general-purpose user representation researches have little ability for full-life cycle modeling on extremely long behavior sequences since user registration. In this study, we propose a novel framework called full- Life cycle User Representation Model (LURM) to tackle this challenge. Specifically, LURM consists of two cascaded sub-models: (I) Bag-of-Interests (BoI) encodes user behaviors in any time period into a sparse vector with super-high dimension (e.g., 10^5); (II) Self-supervised Multi-anchor Encoder Network (SMEN) maps sequences of BoI features to multiple low-dimensional user representations. Specially, SMEN achieves almost lossless dimensionality reduction, benefiting from a novel multi-anchor module which can learn different aspects of user interests. Experiments on several benchmark datasets show that our approach outperforms state-of-the-art general-purpose representation methods.