Since the meaning representations are detailed and accurate annotations which express fine-grained sequence-level semtantics, it is usually hard to train discriminative semantic parsers via Maximum Likelihood Estimation (MLE) in an autoregressive fashion. In this paper, we propose a semantic-aware contrastive learning algorithm, which can learn to distinguish fine-grained meaning representations and take the overall sequence-level semantic into consideration. Specifically, a multi-level online sampling algorithm is proposed to sample confusing and diverse instances. Three semantic-aware similarity functions are designed to accurately measure the distance between meaning representations as a whole. And a ranked contrastive loss is proposed to pull the representations of the semantic-identical instances together and push negative instances away. Experiments on two standard datasets show that our approach achieves significant improvements over MLE baselines and gets state-of-the-art performances by simply applying semantic-aware contrastive learning on a vanilla Seq2Seq model.
Contrastive learning is widely used for sentence representation learning. Despite this prevalence, most studies have focused exclusively on English and few concern domain adaptation for domain-specific downstream tasks, especially for low-resource languages like Japanese, which are characterized by insufficient target domain data and the lack of a proper training strategy. To overcome this, we propose a novel Japanese sentence representation framework, JCSE (derived from ``Contrastive learning of Sentence Embeddings for Japanese''), that creates training data by generating sentences and synthesizing them with sentences available in a target domain. Specifically, a pre-trained data generator is finetuned to a target domain using our collected corpus. It is then used to generate contradictory sentence pairs that are used in contrastive learning for adapting a Japanese language model to a specific task in the target domain. Another problem of Japanese sentence representation learning is the difficulty of evaluating existing embedding methods due to the lack of benchmark datasets. Thus, we establish a comprehensive Japanese Semantic Textual Similarity (STS) benchmark on which various embedding models are evaluated. Based on this benchmark result, multiple embedding methods are chosen and compared with JCSE on two domain-specific tasks, STS in a clinical domain and information retrieval in an educational domain. The results show that JCSE achieves significant performance improvement surpassing direct transfer and other training strategies. This empirically demonstrates JCSE's effectiveness and practicability for downstream tasks of a low-resource language.
Understanding mathematical questions effectively is a crucial task, which can benefit many applications, such as difficulty estimation. Researchers have drawn much attention to designing pre-training models for question representations due to the scarcity of human annotations (e.g., labeling difficulty). However, unlike general free-format texts (e.g., user comments), mathematical questions are generally designed with explicit purposes and mathematical logic, and usually consist of more complex content, such as formulas, and related mathematical knowledge (e.g., Function). Therefore, the problem of holistically representing mathematical questions remains underexplored. To this end, in this paper, we propose a novel contrastive pre-training approach for mathematical question representations, namely QuesCo, which attempts to bring questions with more similar purposes closer. Specifically, we first design two-level question augmentations, including content-level and structure-level, which generate literally diverse question pairs with similar purposes. Then, to fully exploit hierarchical information of knowledge concepts, we propose a knowledge hierarchy-aware rank strategy (KHAR), which ranks the similarities between questions in a fine-grained manner. Next, we adopt a ranking contrastive learning task to optimize our model based on the augmented and ranked questions. We conduct extensive experiments on two real-world mathematical datasets. The experimental results demonstrate the effectiveness of our model.
Reducing the quantity of annotations required for supervised training is vital when labels are scarce and costly. This reduction is particularly important for semantic segmentation tasks involving 3D datasets, which are often significantly smaller and more challenging to annotate than their image-based counterparts. Self-supervised pre-training on unlabelled data is one way to reduce the amount of manual annotations needed. Previous work has focused on pre-training with point clouds exclusively. While useful, this approach often requires two or more registered views. In the present work, we combine image and point cloud modalities by first learning self-supervised image features and then using these features to train a 3D model. By incorporating image data, which is often included in many 3D datasets, our pre-training method only requires a single scan of a scene and can be applied to cases where localization information is unavailable. We demonstrate that our pre-training approach, despite using single scans, achieves comparable performance to other multi-scan, point cloud-only methods.
Recent advancements in vision-language pre-training (e.g. CLIP) have shown that vision models can benefit from language supervision. While many models using language modality have achieved great success on 2D vision tasks, the joint representation learning of 3D point cloud with text remains under-explored due to the difficulty of 3D-Text data pair acquisition and the irregularity of 3D data structure. In this paper, we propose a novel Text4Point framework to construct language-guided 3D point cloud models. The key idea is utilizing 2D images as a bridge to connect the point cloud and the language modalities. The proposed Text4Point follows the pre-training and fine-tuning paradigm. During the pre-training stage, we establish the correspondence of images and point clouds based on the readily available RGB-D data and use contrastive learning to align the image and point cloud representations. Together with the well-aligned image and text features achieved by CLIP, the point cloud features are implicitly aligned with the text embeddings. Further, we propose a Text Querying Module to integrate language information into 3D representation learning by querying text embeddings with point cloud features. For fine-tuning, the model learns task-specific 3D representations under informative language guidance from the label set without 2D images. Extensive experiments demonstrate that our model shows consistent improvement on various downstream tasks, such as point cloud semantic segmentation, instance segmentation, and object detection. The code will be available here: https://github.com/LeapLabTHU/Text4Point
Video-Language Pre-training models have recently significantly improved various multi-modal downstream tasks. Previous dominant works mainly adopt contrastive learning to achieve global feature alignment across modalities. However, the local associations between videos and texts are not modeled, restricting the pre-training models' generality, especially for tasks requiring the temporal video boundary for certain query texts. This work introduces a novel text-video localization pre-text task to enable fine-grained temporal and semantic alignment such that the trained model can accurately perceive temporal boundaries in videos given the text description. Specifically, text-video localization consists of moment retrieval, which predicts start and end boundaries in videos given the text description, and text localization which matches the subset of texts with the video features. To produce temporal boundaries, frame features in several videos are manually merged into a long video sequence that interacts with a text sequence. With the localization task, our method connects the fine-grained frame representations with the word representations and implicitly distinguishes representations of different instances in the single modality. Notably, comprehensive experimental results show that our method significantly improves the state-of-the-art performance on various benchmarks, covering text-to-video retrieval, video question answering, video captioning, temporal action localization and temporal moment retrieval. The code will be released soon.
Self-supervised learning has attracted increasing attention as it learns data-driven representation from data without annotations. Vision transformer-based autoencoder (ViT-AE) by He et al. (2021) is a recent self-supervised learning technique that employs a patch-masking strategy to learn a meaningful latent space. In this paper, we focus on improving ViT-AE (nicknamed ViT-AE++) for a more effective representation of 2D and 3D medical images. We propose two new loss functions to enhance the representation during training. The first loss term aims to improve self-reconstruction by considering the structured dependencies and indirectly improving the representation. The second loss term leverages contrastive loss to optimize the representation from two randomly masked views directly. We extended ViT-AE++ to a 3D fashion for volumetric medical images as an independent contribution. We extensively evaluate ViT-AE++ on both natural images and medical images, demonstrating consistent improvement over vanilla ViT-AE and its superiority over other contrastive learning approaches. Codes are here: https://github.com/chinmay5/vit_ae_plus_plus.git.
Referring image segmentation aims to segment the image region of interest according to the given language expression, which is a typical multi-modal task. Existing methods either adopt the pixel classification-based or the learnable query-based framework for mask generation, both of which are insufficient to deal with various text-image pairs with a fix number of parametric prototypes. In this work, we propose an end-to-end framework built on transformer to perform Linguistic query-Guided mask generation, dubbed LGFormer. It views the linguistic features as query to generate a specialized prototype for arbitrary input image-text pair, thus generating more consistent segmentation results. Moreover, we design several cross-modal interaction modules (\eg, vision-language bidirectional attention module, VLBA) in both encoder and decoder to achieve better cross-modal alignment.
For infrastructure inspections, damage representation does not constantly match the predefined classes of damage grade, resulting in detailed clusters of unseen damages or more complex clusters from overlapped space between two grades. The damage representation has fundamentally complex features; consequently, not all the damage classes can be perfectly predefined. The proposed MN-pair contrastive learning method helps to explore an embedding damage representation beyond the predefined classes by including more detailed clusters. It maximizes both the similarity of M-1 positive images close to an anchor and dissimilarity of N-1 negative images using both weighting loss functions. It learns faster than the N-pair algorithm using one positive image. We proposed a pipeline to obtain the damage representation and used a density-based clustering on a 2-D reduction space to automate finer cluster discrimination. We also visualized the explanation of the damage feature using Grad-CAM for MN-pair damage metric learning. We demonstrated our method in three experimental studies: steel product defect, concrete crack, and the effectiveness of our method and discuss future works.
Video grounding aims to locate a moment of interest matching the given query sentence from an untrimmed video. Previous works ignore the {sparsity dilemma} in video annotations, which fails to provide the context information between potential events and query sentences in the dataset. In this paper, we contend that exploiting easily available captions which describe general actions, i.e., auxiliary captions defined in our paper, will significantly boost the performance. To this end, we propose an Auxiliary Caption Network (ACNet) for video grounding. Specifically, we first introduce dense video captioning to generate dense captions and then obtain auxiliary captions by Non-Auxiliary Caption Suppression (NACS). To capture the potential information in auxiliary captions, we propose Caption Guided Attention (CGA) project the semantic relations between auxiliary captions and query sentences into temporal space and fuse them into visual representations. Considering the gap between auxiliary captions and ground truth, we propose Asymmetric Cross-modal Contrastive Learning (ACCL) for constructing more negative pairs to maximize cross-modal mutual information. Extensive experiments on three public datasets (i.e., ActivityNet Captions, TACoS and ActivityNet-CG) demonstrate that our method significantly outperforms state-of-the-art methods.
Federated learning is widely used to perform decentralized training of a global model on multiple devices while preserving the data privacy of each device. However, it suffers from heterogeneous local data on each training device which increases the difficulty to reach the same level of accuracy as the centralized training. Supervised Contrastive Learning which outperform cross-entropy tries to minimizes the difference between feature space of points belongs to the same class and pushes away points from different classes. We propose Supervised Contrastive Federated Learning in which devices can share the learned class-wise feature spaces with each other and add the supervised-contrastive learning loss as a regularization term to foster the feature space learning. The loss tries to minimize the cosine similarity distance between the feature map and the averaged feature map from another device in the same class and maximizes the distance between the feature map and that in a different class. This new regularization term when added on top of the moon regularization term is found to outperform the other state-of-the-art regularization terms in solving the heterogeneous data distribution problem.
An effective framework for learning 3D representations for perception tasks is distilling rich self-supervised image features via contrastive learning. However, image-to point representation learning for autonomous driving datasets faces two main challenges: 1) the abundance of self-similarity, which results in the contrastive losses pushing away semantically similar point and image regions and thus disturbing the local semantic structure of the learned representations, and 2) severe class imbalance as pretraining gets dominated by over-represented classes. We propose to alleviate the self-similarity problem through a novel semantically tolerant image-to-point contrastive loss that takes into consideration the semantic distance between positive and negative image regions to minimize contrasting semantically similar point and image regions. Additionally, we address class imbalance by designing a class-agnostic balanced loss that approximates the degree of class imbalance through an aggregate sample-to-samples semantic similarity measure. We demonstrate that our semantically-tolerant contrastive loss with class balancing improves state-of-the art 2D-to-3D representation learning in all evaluation settings on 3D semantic segmentation. Our method consistently outperforms state-of-the-art 2D-to-3D representation learning frameworks across a wide range of 2D self-supervised pretrained models.
Image-text contrastive learning models such as CLIP have demonstrated strong task transfer ability. The high generality and usability of these visual models is achieved via a web-scale data collection process to ensure broad concept coverage, followed by expensive pre-training to feed all the knowledge into model weights. Alternatively, we propose REACT, REtrieval-Augmented CusTomization, a framework to acquire the relevant web knowledge to build customized visual models for target domains. We retrieve the most relevant image-text pairs (~3% of CLIP pre-training data) from the web-scale database as external knowledge, and propose to customize the model by only training new modualized blocks while freezing all the original weights. The effectiveness of REACT is demonstrated via extensive experiments on classification, retrieval, detection and segmentation tasks, including zero, few, and full-shot settings. Particularly, on the zero-shot classification task, compared with CLIP, it achieves up to 5.4% improvement on ImageNet and 3.7% on the ELEVATER benchmark (20 datasets).
As a fundamental and challenging task in bridging language and vision domains, Image-Text Retrieval (ITR) aims at searching for the target instances that are semantically relevant to the given query from the other modality, and its key challenge is to measure the semantic similarity across different modalities. Although significant progress has been achieved, existing approaches typically suffer from two major limitations: (1) It hurts the accuracy of the representation by directly exploiting the bottom-up attention based region-level features where each region is equally treated. (2) It limits the scale of negative sample pairs by employing the mini-batch based end-to-end training mechanism. To address these limitations, we propose a Unified Semantic Enhancement Momentum Contrastive Learning (USER) method for ITR. Specifically, we delicately design two simple but effective Global representation based Semantic Enhancement (GSE) modules. One learns the global representation via the self-attention algorithm, noted as Self-Guided Enhancement (SGE) module. The other module benefits from the pre-trained CLIP module, which provides a novel scheme to exploit and transfer the knowledge from an off-the-shelf model, noted as CLIP-Guided Enhancement (CGE) module. Moreover, we incorporate the training mechanism of MoCo into ITR, in which two dynamic queues are employed to enrich and enlarge the scale of negative sample pairs. Meanwhile, a Unified Training Objective (UTO) is developed to learn from mini-batch based and dynamic queue based samples. Extensive experiments on the benchmark MSCOCO and Flickr30K datasets demonstrate the superiority of both retrieval accuracy and inference efficiency. Our source code will be released at https://github.com/zhangy0822/USER.
Medical image segmentation methods are generally designed as fully-supervised to guarantee model performance, which require a significant amount of expert annotated samples that are high-cost and laborious. Semi-supervised image segmentation can alleviate the problem by utilizing a large number of unlabeled images along with limited labeled images. However, learning a robust representation from numerous unlabeled images remains challenging due to potential noise in pseudo labels and insufficient class separability in feature space, which undermines the performance of current semi-supervised segmentation approaches. To address the issues above, we propose a novel semi-supervised segmentation method named as Rectified Contrastive Pseudo Supervision (RCPS), which combines a rectified pseudo supervision and voxel-level contrastive learning to improve the effectiveness of semi-supervised segmentation. Particularly, we design a novel rectification strategy for the pseudo supervision method based on uncertainty estimation and consistency regularization to reduce the noise influence in pseudo labels. Furthermore, we introduce a bidirectional voxel contrastive loss to the network to ensure intra-class consistency and inter-class contrast in feature space, which increases class separability in the segmentation. The proposed RCPS segmentation method has been validated on two public datasets and an in-house clinical dataset. Experimental results reveal that the proposed method yields better segmentation performance compared with the state-of-the-art methods in semi-supervised medical image segmentation. The source code is available at https://github.com/hsiangyuzhao/RCPS.
A well-designed recommender system can accurately capture the attributes of users and items, reflecting the unique preferences of individuals. Traditional recommendation techniques usually focus on modeling the singular type of behaviors between users and items. However, in many practical recommendation scenarios (e.g., social media, e-commerce), there exist multi-typed interactive behaviors in user-item relationships, such as click, tag-as-favorite, and purchase in online shopping platforms. Thus, how to make full use of multi-behavior information for recommendation is of great importance to the existing system, which presents challenges in two aspects that need to be explored: (1) Utilizing users' personalized preferences to capture multi-behavioral dependencies; (2) Dealing with the insufficient recommendation caused by sparse supervision signal for target behavior. In this work, we propose a Knowledge Enhancement Multi-Behavior Contrastive Learning Recommendation (KMCLR) framework, including two Contrastive Learning tasks and three functional modules to tackle the above challenges, respectively. In particular, we design the multi-behavior learning module to extract users' personalized behavior information for user-embedding enhancement, and utilize knowledge graph in the knowledge enhancement module to derive more robust knowledge-aware representations for items. In addition, in the optimization stage, we model the coarse-grained commonalities and the fine-grained differences between multi-behavior of users to further improve the recommendation effect. Extensive experiments and ablation tests on the three real-world datasets indicate our KMCLR outperforms various state-of-the-art recommendation methods and verify the effectiveness of our method.
The data collected from a vehicle's Controller Area Network (CAN) can quickly exceed human analysis or annotation capabilities when considering fleets of vehicles, which stresses the importance of unsupervised machine learning methods. This work presents a simultaneous clustering and segmentation approach for vehicle CAN-data that identifies common driving events in an unsupervised manner. The approach builds on self-supervised learning (SSL) for multivariate time series to distinguish different driving events in the learned latent space. We evaluate our approach with a dataset of real Tesla Model 3 vehicle CAN-data and a two-hour driving session that we annotated with different driving events. With our approach, we evaluate the applicability of recent time series-related contrastive and generative SSL techniques to learn representations that distinguish driving events. Compared to state-of-the-art (SOTA) generative SSL methods for driving event discovery, we find that contrastive learning approaches reach similar performance.
The large size and fast growth of data repositories, such as data lakes, has spurred the need for data discovery to help analysts find related data. The problem has become challenging as (i) a user typically does not know what datasets exist in an enormous data repository; and (ii) there is usually a lack of a unified data model to capture the interrelationships between heterogeneous datasets from disparate sources. In this work, we address one important class of discovery needs: finding union-able tables. The task is to find tables in a data lake that can be unioned with a given query table. The challenge is to recognize union-able columns even if they are represented differently. In this paper, we propose a data-driven learning approach: specifically, an unsupervised representation learning and embedding retrieval task. Our key idea is to exploit self-supervised contrastive learning to learn an embedding model that takes into account the indexing/search data structure and produces embeddings close by for columns with semantically similar values while pushing apart columns with semantically dissimilar values. We then find union-able tables based on similarities between their constituent columns in embedding space. On a real-world data lake, we demonstrate that our best-performing model achieves significant improvements in precision ($16\% \uparrow$), recall ($17\% \uparrow $), and query response time (7x faster) compared to the state-of-the-art.
We develop a self-supervised method for density-based anomaly detection using contrastive learning, and test it using event-level anomaly data from CMS ADC2021. The AnomalyCLR technique is data-driven and uses augmentations of the background data to mimic non-Standard-Model events in a model-agnostic way. It uses a permutation-invariant Transformer Encoder architecture to map the objects measured in a collider event to the representation space, where the data augmentations define a representation space which is sensitive to potential anomalous features. An AutoEncoder trained on background representations then computes anomaly scores for a variety of signals in the representation space. With AnomalyCLR we find significant improvements on performance metrics for all signals when compared to the raw data baseline.
Video summarization aims at choosing parts of a video that narrate a story as close as possible to the original one. Most of the existing video summarization approaches focus on hand-crafted labels. As the number of videos grows exponentially, there emerges an increasing need for methods that can learn meaningful summarizations without labeled annotations. In this paper, we aim to maximally exploit unsupervised video summarization while concentrating the supervision to a few, personalized labels as an add-on. To do so, we formulate the key requirements for the informative video summarization. Then, we propose contrastive learning as the answer to both questions. To further boost Contrastive video Summarization (CSUM), we propose to contrast top-k features instead of a mean video feature as employed by the existing method, which we implement with a differentiable top-k feature selector. Our experiments on several benchmarks demonstrate, that our approach allows for meaningful and diverse summaries when no labeled data is provided.