Spoken language understanding (SLU) is an essential task for machines to understand human speech for better interactions. However, errors from the automatic speech recognizer (ASR) usually hurt the understanding performance. In reality, ASR systems may not be easy to adjust for the target scenarios. Therefore, this paper focuses on learning utterance representations that are robust to ASR errors using a contrastive objective, and further strengthens the generalization ability by combining supervised contrastive learning and self-distillation in model fine-tuning. Experiments on three benchmark datasets demonstrate the effectiveness of our proposed approach.
Graph contrastive learning has gained significant progress recently. However, existing works have rarely explored non-aligned node-node contrasting. In this paper, we propose a novel graph contrastive learning method named RoSA that focuses on utilizing non-aligned augmented views for node-level representation learning. First, we leverage the earth mover's distance to model the minimum effort to transform the distribution of one view to the other as our contrastive objective, which does not require alignment between views. Then we introduce adversarial training as an auxiliary method to increase sampling diversity and enhance the robustness of our model. Experimental results show that RoSA outperforms a series of graph contrastive learning frameworks on homophilous, non-homophilous and dynamic graphs, which validates the effectiveness of our work. To the best of our awareness, RoSA is the first work focuses on the non-aligned node-node graph contrastive learning problem. Our codes are available at: \href{https://github.com/ZhuYun97/RoSA}{\texttt{https://github.com/ZhuYun97/RoSA}}
Existing computer vision research in artwork struggles with artwork's fine-grained attributes recognition and lack of curated annotated datasets due to their costly creation. To the best of our knowledge, we are one of the first methods to use CLIP (Contrastive Language-Image Pre-Training) to train a neural network on a variety of artwork images and text descriptions pairs. CLIP is able to learn directly from free-form art descriptions, or, if available, curated fine-grained labels. Model's zero-shot capability allows predicting accurate natural language description for a given image, without directly optimizing for the task. Our approach aims to solve 2 challenges: instance retrieval and fine-grained artwork attribute recognition. We use the iMet Dataset, which we consider the largest annotated artwork dataset. In this benchmark we achieved competitive results using only self-supervision.
We present an approach to learn voice-face representations from the talking face videos, without any identity labels. Previous works employ cross-modal instance discrimination tasks to establish the correlation of voice and face. These methods neglect the semantic content of different videos, introducing false-negative pairs as training noise. Furthermore, the positive pairs are constructed based on the natural correlation between audio clips and visual frames. However, this correlation might be weak or inaccurate in a large amount of real-world data, which leads to deviating positives into the contrastive paradigm. To address these issues, we propose the cross-modal prototype contrastive learning (CMPC), which takes advantage of contrastive methods and resists adverse effects of false negatives and deviate positives. On one hand, CMPC could learn the intra-class invariance by constructing semantic-wise positives via unsupervised clustering in different modalities. On the other hand, by comparing the similarities of cross-modal instances from that of cross-modal prototypes, we dynamically recalibrate the unlearnable instances' contribution to overall loss. Experiments show that the proposed approach outperforms state-of-the-art unsupervised methods on various voice-face association evaluation protocols. Additionally, in the low-shot supervision setting, our method also has a significant improvement compared to previous instance-wise contrastive learning.
Image harmonization aims to achieve visual consistency in composite images by adapting a foreground to make it compatible with a background. However, existing methods always only use the real image as the positive sample to guide the training, and at most introduce the corresponding composite image as a single negative sample for an auxiliary constraint, which leads to limited distortion knowledge, and further causes a too large solution space, making the generated harmonized image distorted. Besides, none of them jointly constrain from the foreground self-style and foreground-background style consistency, which exacerbates this problem. Moreover, recent region-aware adaptive instance normalization achieves great success but only considers the global background feature distribution, making the aligned foreground feature distribution biased. To address these issues, we propose a self-consistent style contrastive learning scheme (SCS-Co). By dynamically generating multiple negative samples, our SCS-Co can learn more distortion knowledge and well regularize the generated harmonized image in the style representation space from two aspects of the foreground self-style and foreground-background style consistency, leading to a more photorealistic visual result. In addition, we propose a background-attentional adaptive instance normalization (BAIN) to achieve an attention-weighted background feature distribution according to the foreground-background feature similarity. Experiments demonstrate the superiority of our method over other state-of-the-art methods in both quantitative comparison and visual analysis.
Recently, vision-language joint representation learning has proven to be highly effective in various scenarios. In this paper, we specifically adapt vision-language joint learning for scene text detection, a task that intrinsically involves cross-modal interaction between the two modalities: vision and language, since text is the written form of language. Concretely, we propose to learn contextualized, joint representations through vision-language pre-training, for the sake of enhancing the performance of scene text detectors. Towards this end, we devise a pre-training architecture with an image encoder, a text encoder and a cross-modal encoder, as well as three pretext tasks: image-text contrastive learning (ITC), masked language modeling (MLM) and word-in-image prediction (WIP). The pre-trained model is able to produce more informative representations with richer semantics, which could readily benefit existing scene text detectors (such as EAST and PSENet) in the down-stream text detection task. Extensive experiments on standard benchmarks demonstrate that the proposed paradigm can significantly improve the performance of various representative text detectors, outperforming previous pre-training approaches. The code and pre-trained models will be publicly released.
To train image-caption retrieval (ICR) methods, contrastive loss functions are a common choice for optimization functions. Unfortunately, contrastive ICR methods are vulnerable to predictive feature suppression. Predictive features are features that correctly indicate the similarity between a query and a candidate item. However, in the presence of multiple predictive features during training, encoder models tend to suppress redundant predictive features, since these features are not needed to learn to discriminate between positive and negative pairs. While some predictive features are redundant during training, these features might be relevant during evaluation. We introduce an approach to reduce predictive feature suppression for resource-constrained ICR methods: latent target decoding (LTD). We add an additional decoder to the contrastive ICR framework, to reconstruct the input caption in a latent space of a general-purpose sentence encoder, which prevents the image and caption encoder from suppressing predictive features. We implement the LTD objective as an optimization constraint, to ensure that the reconstruction loss is below a bound value while primarily optimizing for the contrastive loss. Importantly, LTD does not depend on additional training data or expensive (hard) negative mining strategies. Our experiments show that, unlike reconstructing the input caption in the input space, LTD reduces predictive feature suppression, measured by obtaining higher recall@k, r-precision, and nDCG scores than a contrastive ICR baseline. Moreover, we show that LTD should be implemented as an optimization constraint instead of a dual optimization objective. Finally, we show that LTD can be used with different contrastive learning losses and a wide variety of resource-constrained ICR methods.
Current contrastive learning frameworks focus on leveraging a single supervisory signal to learn representations, which limits the efficacy on unseen data and downstream tasks. In this paper, we present a hierarchical multi-label representation learning framework that can leverage all available labels and preserve the hierarchical relationship between classes. We introduce novel hierarchy preserving losses, which jointly apply a hierarchical penalty to the contrastive loss, and enforce the hierarchy constraint. The loss function is data driven and automatically adapts to arbitrary multi-label structures. Experiments on several datasets show that our relationship-preserving embedding performs well on a variety of tasks and outperform the baseline supervised and self-supervised approaches. Code is available at https://github.com/salesforce/hierarchicalContrastiveLearning.
Similarity metric is crucial for massive MIMO positioning utilizing channel state information~(CSI). In this letter, we propose a novel massive MIMO CSI similarity learning method via deep convolutional neural network~(DCNN) and contrastive learning. A contrastive loss function is designed considering multiple positive and negative CSI samples drawn from a training dataset. The DCNN encoder is trained using the loss so that positive samples are mapped to points close to the anchor's encoding, while encodings of negative samples are kept away from the anchor's in the representation space. Evaluation results of fingerprint-based positioning on a real-world CSI dataset show that the learned similarity metric improves positioning accuracy significantly compared with other known state-of-the-art methods.
Recent general-purpose audio representations show state-of-the-art performance on various audio tasks. These representations are pre-trained by self-supervised learning methods that create training signals from the input. For example, typical audio contrastive learning uses temporal relationships among input sounds to create training signals, whereas some methods use a difference among input views created by data augmentations. However, these training signals do not provide information derived from the intact input sound, which we think is suboptimal for learning representation that describes the input as it is. In this paper, we seek to learn audio representations from the input itself as supervision using a pretext task of auto-encoding of masked spectrogram patches, Masked Spectrogram Modeling (MSM, a variant of Masked Image Modeling applied to audio spectrogram). To implement MSM, we use Masked Autoencoders (MAE), an image self-supervised learning method. MAE learns to efficiently encode the small number of visible patches into latent representations to carry essential information for reconstructing a large number of masked patches. While training, MAE minimizes the reconstruction error, which uses the input as training signal, consequently achieving our goal. We conducted experiments on our MSM using MAE (MSM-MAE) models under the evaluation benchmark of the HEAR 2021 NeurIPS Challenge. Our MSM-MAE models outperformed the HEAR 2021 Challenge results on seven out of 15 tasks (e.g., accuracies of 73.4% on CREMA-D and 85.8% on LibriCount), while showing top performance on other tasks where specialized models perform better. We also investigate how the design choices of MSM-MAE impact the performance and conduct qualitative analysis of visualization outcomes to gain an understanding of learned representations. We make our code available online.
Most modern recommender systems predict users preferences with two components: user and item embedding learning, followed by the user-item interaction modeling. By utilizing the auxiliary review information accompanied with user ratings, many of the existing review-based recommendation models enriched user/item embedding learning ability with historical reviews or better modeled user-item interactions with the help of available user-item target reviews. Though significant progress has been made, we argue that current solutions for review-based recommendation suffer from two drawbacks. First, as review-based recommendation can be naturally formed as a user-item bipartite graph with edge features from corresponding user-item reviews, how to better exploit this unique graph structure for recommendation? Second, while most current models suffer from limited user behaviors, can we exploit the unique self-supervised signals in the review-aware graph to guide two recommendation components better? To this end, in this paper, we propose a novel Review-aware Graph Contrastive Learning (RGCL) framework for review-based recommendation. Specifically, we first construct a review-aware user-item graph with feature-enhanced edges from reviews, where each edge feature is composed of both the user-item rating and the corresponding review semantics. This graph with feature-enhanced edges can help attentively learn each neighbor node weight for user and item representation learning. After that, we design two additional contrastive learning tasks (i.e., Node Discrimination and Edge Discrimination) to provide self-supervised signals for the two components in recommendation process. Finally, extensive experiments over five benchmark datasets demonstrate the superiority of our proposed RGCL compared to the state-of-the-art baselines.
Sounds reach one microphone in a stereo pair sooner than the other, resulting in an interaural time delay that conveys their directions. Estimating a sound's time delay requires finding correspondences between the signals recorded by each microphone. We propose to learn these correspondences through self-supervision, drawing on recent techniques from visual tracking. We adapt the contrastive random walk of Jabri et al. to learn a cycle-consistent representation from unlabeled stereo sounds, resulting in a model that performs on par with supervised methods on "in the wild" internet recordings. We also propose a multimodal contrastive learning model that solves a visually-guided localization task: estimating the time delay for a particular person in a multi-speaker mixture, given a visual representation of their face. Project site: https://ificl.github.io/stereocrw/
Collaborative Filtering (CF) has emerged as fundamental paradigms for parameterizing users and items into latent representation space, with their correlative patterns from interaction data. Among various CF techniques, the development of GNN-based recommender systems, e.g., PinSage and LightGCN, has offered the state-of-the-art performance. However, two key challenges have not been well explored in existing solutions: i) The over-smoothing effect with deeper graph-based CF architecture, may cause the indistinguishable user representations and degradation of recommendation results. ii) The supervision signals (i.e., user-item interactions) are usually scarce and skewed distributed in reality, which limits the representation power of CF paradigms. To tackle these challenges, we propose a new self-supervised recommendation framework Hypergraph Contrastive Collaborative Filtering (HCCF) to jointly capture local and global collaborative relations with a hypergraph-enhanced cross-view contrastive learning architecture. In particular, the designed hypergraph structure learning enhances the discrimination ability of GNN-based CF paradigm, so as to comprehensively capture the complex high-order dependencies among users. Additionally, our HCCF model effectively integrates the hypergraph structure encoding with self-supervised learning to reinforce the representation quality of recommender systems, based on the hypergraph-enhanced self-discrimination. Extensive experiments on three benchmark datasets demonstrate the superiority of our model over various state-of-the-art recommendation methods, and the robustness against sparse user interaction data. Our model implementation codes are available at https://github.com/akaxlh/HCCF.
Digital pathological analysis is run as the main examination used for cancer diagnosis. Recently, deep learning-driven feature extraction from pathology images is able to detect genetic variations and tumor environment, but few studies focus on differential gene expression in tumor cells. In this paper, we propose a self-supervised contrastive learning framework, HistCode, to infer differential gene expressions from whole slide images (WSIs). We leveraged contrastive learning on large-scale unannotated WSIs to derive slide-level histopathological feature in latent space, and then transfer it to tumor diagnosis and prediction of differentially expressed cancer driver genes. Our extensive experiments showed that our method outperformed other state-of-the-art models in tumor diagnosis tasks, and also effectively predicted differential gene expressions. Interestingly, we found the higher fold-changed genes can be more precisely predicted. To intuitively illustrate the ability to extract informative features from pathological images, we spatially visualized the WSIs colored by the attentive scores of image tiles. We found that the tumor and necrosis areas were highly consistent with the annotations of experienced pathologists. Moreover, the spatial heatmap generated by lymphocyte-specific gene expression patterns was also consistent with the manually labeled WSI.
Learning semantically meaningful sentence embeddings is an open problem in natural language processing. In this work, we propose a sentence embedding learning approach that exploits both visual and textual information via a multimodal contrastive objective. Through experiments on a variety of semantic textual similarity tasks, we demonstrate that our approach consistently improves the performance across various datasets and pre-trained encoders. In particular, combining a small amount of multimodal data with a large text-only corpus, we improve the state-of-the-art average Spearman's correlation by 1.7%. By analyzing the properties of the textual embedding space, we show that our model excels in aligning semantically similar sentences, providing an explanation for its improved performance.
Dense retrieval has shown promising results in many information retrieval (IR) related tasks, whose foundation is high-quality text representation learning for effective search. Some recent studies have shown that autoencoder-based language models are able to boost the dense retrieval performance using a weak decoder. However, we argue that 1) it is not discriminative to decode all the input texts and, 2) even a weak decoder has the bypass effect on the encoder. Therefore, in this work, we introduce a novel contrastive span prediction task to pre-train the encoder alone, but still retain the bottleneck ability of the autoencoder. % Therefore, in this work, we propose to drop out the decoder and introduce a novel contrastive span prediction task to pre-train the encoder alone. The key idea is to force the encoder to generate the text representation close to its own random spans while far away from others using a group-wise contrastive loss. In this way, we can 1) learn discriminative text representations efficiently with the group-wise contrastive learning over spans and, 2) avoid the bypass effect of the decoder thoroughly. Comprehensive experiments over publicly available retrieval benchmark datasets show that our approach can outperform existing pre-training methods for dense retrieval significantly.
As an effective data augmentation method, Mixup synthesizes an extra amount of samples through linear interpolations. Despite its theoretical dependency on data properties, Mixup reportedly performs well as a regularizer and calibrator contributing reliable robustness and generalization to deep model training. In this paper, inspired by Universum Learning which uses out-of-class samples to assist the target tasks, we investigate Mixup from a largely under-explored perspective - the potential to generate in-domain samples that belong to none of the target classes, that is, universum. We find that in the framework of supervised contrastive learning, Mixup-induced universum can serve as surprisingly high-quality hard negatives, greatly relieving the need for large batch sizes in contrastive learning. With these findings, we propose Universum-inspired supervised Contrastive learning (UniCon), which incorporates Mixup strategy to generate Mixup-induced universum as universum negatives and pushes them apart from anchor samples of the target classes. We extend our method to the unsupervised setting, proposing Unsupervised Universum-inspired contrastive model (Un-Uni). Our approach not only improves Mixup with hard labels, but also innovates a novel measure to generate universum data. With a linear classifier on the learned representations, UniCon shows state-of-the-art performance on various datasets. Specially, UniCon achieves 81.7% top-1 accuracy on CIFAR-100, surpassing the state of art by a significant margin of 5.2% with a much smaller batch size, typically, 256 in UniCon vs. 1024 in SupCon using ResNet-50. Un-Uni also outperforms SOTA methods on CIFAR-100. The code of this paper is released on https://github.com/hannaiiyanggit/UniCon.
Deep learning has recently achieved significant progress in trajectory forecasting. However, the scarcity of trajectory data inhibits the data-hungry deep-learning models from learning good representations. While mature representation learning methods exist in computer vision and natural language processing, these pre-training methods require large-scale data. It is hard to replicate these approaches in trajectory forecasting due to the lack of adequate trajectory data (e.g., 34K samples in the nuScenes dataset). To work around the scarcity of trajectory data, we resort to another data modality closely related to trajectories-HD-maps, which is abundantly provided in existing datasets. In this paper, we propose PreTraM, a self-supervised pre-training scheme via connecting trajectories and maps for trajectory forecasting. Specifically, PreTraM consists of two parts: 1) Trajectory-Map Contrastive Learning, where we project trajectories and maps to a shared embedding space with cross-modal contrastive learning, and 2) Map Contrastive Learning, where we enhance map representation with contrastive learning on large quantities of HD-maps. On top of popular baselines such as AgentFormer and Trajectron++, PreTraM boosts their performance by 5.5% and 6.9% relatively in FDE-10 on the challenging nuScenes dataset. We show that PreTraM improves data efficiency and scales well with model size.
Test-time adaptation is a special setting of unsupervised domain adaptation where a trained model on the source domain has to adapt to the target domain without accessing source data. We propose a novel way to leverage self-supervised contrastive learning to facilitate target feature learning, along with an online pseudo labeling scheme with refinement that significantly denoises pseudo labels. The contrastive learning task is applied jointly with pseudo labeling, contrasting positive and negative pairs constructed similarly as MoCo but with source-initialized encoder, and excluding same-class negative pairs indicated by pseudo labels. Meanwhile, we produce pseudo labels online and refine them via soft voting among their nearest neighbors in the target feature space, enabled by maintaining a memory queue. Our method, AdaContrast, achieves state-of-the-art performance on major benchmarks while having several desirable properties compared to existing works, including memory efficiency, insensitivity to hyper-parameters, and better model calibration. Project page: sites.google.com/view/adacontrast.
We propose DiffCSE, an unsupervised contrastive learning framework for learning sentence embeddings. DiffCSE learns sentence embeddings that are sensitive to the difference between the original sentence and an edited sentence, where the edited sentence is obtained by stochastically masking out the original sentence and then sampling from a masked language model. We show that DiffSCE is an instance of equivariant contrastive learning (Dangovski et al., 2021), which generalizes contrastive learning and learns representations that are insensitive to certain types of augmentations and sensitive to other "harmful" types of augmentations. Our experiments show that DiffCSE achieves state-of-the-art results among unsupervised sentence representation learning methods, outperforming unsupervised SimCSE by 2.3 absolute points on semantic textual similarity tasks.