We study three intriguing properties of contrastive learning. First, we generalize the standard contrastive loss to a broader family of losses, and we find that various instantiations of the generalized loss perform similarly under the presence of a multi-layer non-linear projection head. Second, we study if instance-based contrastive learning (with a global image representation) can learn well on images with multiple objects present. We find that meaningful hierarchical local features can be learned despite the fact that these objectives operate on global instance-level features. Finally, we study the phenomenon of feature suppression among competing features shared across augmented views, such as "color distribution" vs "object class". We construct datasets with explicit and controllable competing features, and show that, for contrastive learning, a few bits of easy-to-learn shared features can suppress, and even fully prevent, the learning of other sets of competing features. In scenarios where there are multiple objects in an image, the dominant object would suppress the learning of smaller objects. Existing contrastive learning methods critically rely on data augmentation to favor certain sets of features over others, and could suffer from learning saturation for scenarios where existing augmentations cannot fully address the feature suppression. This poses open challenges to existing contrastive learning techniques.
Technical University of Munich（慕尼黑工业大学）；Helmholtz Zentrum Munich（慕尼黑亥姆霍兹中心）；Imperial College London（伦敦帝国理工学院）
Image-text contrastive learning has proven effective for pretraining medical image models. When targeting localized downstream tasks like semantic segmentation or object detection, additional local contrastive losses that align image regions with sentences have shown promising results. We study how local contrastive losses are related to global (per-sample) contrastive losses and which effects they have on localized medical downstream tasks. Based on a theoretical comparison, we propose to remove some components of local losses and replace others by a novel distribution prior which enforces un…
NAVER LABS Europe（NAVER LABS 欧洲）；University of Amsterdam（阿姆斯特丹大学）
Contrastive losses have long been a key ingredient of deep metric learning and are now becoming more popular due to the success of self-supervised learning. Recent research has shown the benefit of decomposing such losses into two sub-losses which act in a complementary way when learning the representation network: a positive term and an entropy term. Although the overall loss is thus defined as a combination of two terms, the balance of these two terms is often hidden behind implementation details and is largely ignored and sub-optimal in practice. In this work, we approach the balance of con…
Siamese neural network is a very powerful architecture for both feature extraction and metric learning. It usually consists of several networks that share weights. The Siamese concept is topology-agnostic and can use any neural network as its backbone. The two most popular loss functions for training these networks are the triplet and contrastive loss functions. In this paper, we propose two novel loss functions, named Fisher Discriminant Triplet (FDT) and Fisher Discriminant Contrastive (FDC). The former uses anchor-neighbor-distant triplets while the latter utilizes pairs of anchor-neighbor …
A lot of recent advances in unsupervised feature learning are based on designing features which are invariant under semantic data augmentations. A common way to do this is contrastive learning, which uses positive and negative samples. Some recent works however have shown promising results for non-contrastive learning, which does not require negative samples. However, the non-contrastive losses have obvious "collapsed" minima, in which the encoders output a constant feature embedding, independent of the input. A folk conjecture is that so long as these collapsed solutions are avoided, the prod…
The University of Tokyo（东京大学）；RIKEN AIP（理化学研究所先进智能项目中心）
Contrastive representation learning encourages data representation to make semantically similar pairs closer than randomly drawn negative samples, which has been successful in various domains such as vision, language, and graphs. Recent theoretical studies have attempted to explain the benefit of the large negative sample size by upper-bounding the downstream classification loss with the contrastive loss. However, the previous surrogate bounds have two drawbacks: they are only legitimate for a limited range of negative sample sizes and prohibitively large even within that range. Due to these d…
MIT（麻省理工学院）；MSU（密歇根州立大学）；IBM Research AI（IBM研究院人工智能中心）；UCSD（加州大学圣地亚哥分校）
As a seminal tool in self-supervised representation learning, contrastive learning has gained unprecedented attention in recent years. In essence, contrastive learning aims to leverage pairs of positive and negative samples for representation learning, which relates to exploiting neighborhood information in a feature space. By investigating the connection between contrastive learning and neighborhood component analysis (NCA), we provide a novel stochastic nearest neighbor viewpoint of contrastive learning and subsequently propose a series of contrastive losses that outperform the existing ones…
We show that Contrastive Learning (CL) under a broad family of loss functions (including InfoNCE) has a unified formulation of coordinate-wise optimization on the network parameter $\boldsymbolθ$ and pairwise importance $α$, where the \emph{max player} $\boldsymbolθ$ learns representation for contrastiveness, and the \emph{min player} $α$ puts more weights on pairs of distinct samples that share similar representations. The resulting formulation, called $α$-CL, unifies not only various existing contrastive losses, which differ by how sample-pair importance $α$ is constructed, but also is able …
Contrastive learning allows us to flexibly define powerful losses by contrasting positive pairs from sets of negative samples. Recently, the principle has also been used to learn cross-modal embeddings for video and text, yet without exploiting its full potential. In particular, previous losses do not take the intra-modality similarities into account, which leads to inefficient embeddings, as the same content is mapped to multiple points in the embedding space. With CrossCLR, we present a contrastive loss that fixes this issue. Moreover, we define sets of highly related samples in terms of the…
The Chinese University of Hong Kong（香港中文大学）；Tencent AI Lab（腾讯AI实验室）
Exemplar-Guided Paraphrase Generation (EGPG) aims to generate a target sentence which conforms to the style of the given exemplar while encapsulating the content information of the source sentence. In this paper, we propose a new method with the goal of learning a better representation of the style andthe content. This method is mainly motivated by the recent success of contrastive learning which has demonstrated its power in unsupervised feature extraction tasks. The idea is to design two contrastive losses with respect to the content and the style by considering two problem characteristics d…
the University of Iowa（爱荷华大学）；Google Research（谷歌研究院）；Nanjing University（南京大学）
In this paper, we study contrastive learning from an optimization perspective, aiming to analyze and address a fundamental issue of existing contrastive learning methods that either rely on a large batch size or a large dictionary of feature vectors. We consider a global objective for contrastive learning, which contrasts each positive pair with all negative pairs for an anchor point. From the optimization perspective, we explain why existing methods such as SimCLR require a large batch size in order to achieve a satisfactory result. In order to remove such requirement, we propose a memory-eff…
In recent years, contrastive learning achieves impressive results on self-supervised visual representation learning, but there still lacks a rigorous understanding of its learning dynamics. In this paper, we show that if we cast a contrastive objective equivalently into the feature space, then its learning dynamics admits an interpretable form. Specifically, we show that its gradient descent corresponds to a specific message passing scheme on the corresponding augmentation graph. Based on this perspective, we theoretically characterize how contrastive learning gradually learns discriminative f…
Data61/CSIRO（澳大利亚联邦科学与工业研究组织 Data61）；Australian National University（澳大利亚国立大学）
Graph contrastive learning attracts/disperses node representations for similar/dissimilar node pairs under some notion of similarity. It may be combined with a low-dimensional embedding of nodes to preserve intrinsic and structural properties of a graph. In this paper, we extend the celebrated Laplacian Eigenmaps with contrastive learning, and call them COntrastive Laplacian EigenmapS (COLES). Starting from a GAN-inspired contrastive formulation, we show that the Jensen-Shannon divergence underlying many contrastive graph embedding models fails under disjoint positive and negative distribution…
We present federated momentum contrastive clustering (FedMCC), a learning framework that can not only extract discriminative representations over distributed local data but also perform data clustering. In FedMCC, a transformed data pair passes through both the online and target networks, resulting in four representations over which the losses are determined. The resulting high-quality representations generated by FedMCC can outperform several existing self-supervised learning methods for linear evaluation and semi-supervised learning tasks. FedMCC can easily be adapted to ordinary centralized…
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 …
Qing Yuan Research Institute, Shanghai Jiao Tong University（上海交通大学清源研究院）；Huawei Noah’s Ark Lab（华为诺亚方舟实验室）；School of Mathematical Sciences, Peking University（北京大学数学科学学院）
Recently, self-supervised learning has attracted great attention, since it only requires unlabeled data for model training. Contrastive learning is one popular method for self-supervised learning and has achieved promising empirical performance. However, the theoretical understanding of its generalization ability is still limited. To this end, we define a kind of $(σ,δ)$-measure to mathematically quantify the data augmentation, and then provide an upper bound of the downstream classification error rate based on the measure. It reveals that the generalization ability of contrastive self-supervi…
National University of Singapore（新加坡国立大学）；Youtu Lab Tencent（腾讯优图实验室）
This paper proposes a probabilistic contrastive loss function for self-supervised learning. The well-known contrastive loss is deterministic and involves a temperature hyperparameter that scales the inner product between two normed feature embeddings. By reinterpreting the temperature hyperparameter as a quantity related to the radius of the hypersphere, we derive a new loss function that involves a confidence measure which quantifies uncertainty in a mathematically grounding manner. Some intriguing properties of the proposed loss function are empirically demonstrated, which agree with human-l…
Contrastive self-supervised learning has been successfully used in many domains, such as images, texts, graphs, etc., to learn features without requiring label information. In this paper, we propose a new local contrastive feature learning (LoCL) framework, and our theme is to learn local patterns/features from tabular data. In order to create a niche for local learning, we use feature correlations to create a maximum-spanning tree, and break the tree into feature subsets, with strongly correlated features being assigned next to each other. Convolutional learning of the features is used to lea…
University of Salzburg（萨尔茨堡大学）；UNC Chapel Hill（北卡罗来纳大学教堂山分校）
Minimizing cross-entropy over the softmax scores of a linear map composed with a high-capacity encoder is arguably the most popular choice for training neural networks on supervised learning tasks. However, recent works show that one can directly optimize the encoder instead, to obtain equally (or even more) discriminative representations via a supervised variant of a contrastive objective. In this work, we address the question whether there are fundamental differences in the sought-for representation geometry in the output space of the encoder at minimal loss. Specifically, we prove, under mi…
Hard-Negative Sampling for Contrastive Learning: Optimal Representation Geometry and Neural- vs Dimensional-Collapse
Department of Electrical Engineering, Tufts University（Tufts大学电气工程系）；Department of Engineering, Engineering Technology, East Tennessee State University（东田纳西州立大学工程系）；Department of Electrical and Computer Engineering, Boston University（波士顿大学电气与计算机工程系）
Google Research（谷歌研究院）；Boston University（波士顿大学）；Snap Inc.（森宝公司）；MIT（麻省理工学院）
Contrastive learning applied to self-supervised representation learning has seen a resurgence in recent years, leading to state of the art performance in the unsupervised training of deep image models. Modern batch contrastive approaches subsume or significantly outperform traditional contrastive losses such as triplet, max-margin and the N-pairs loss. In this work, we extend the self-supervised batch contrastive approach to the fully-supervised setting, allowing us to effectively leverage label information. Clusters of points belonging to the same class are pulled together in embedding space,…
Contrastive learning is a popular form of self-supervised learning that encourages augmentations (views) of the same input to have more similar representations compared to augmentations of different inputs. Recent attempts to theoretically explain the success of contrastive learning on downstream classification tasks prove guarantees depending on properties of {\em augmentations} and the value of {\em contrastive loss} of representations. We demonstrate that such analyses, that ignore {\em inductive biases} of the function class and training algorithm, cannot adequately explain the success of …
MIT CSAIL（麻省理工学院计算机科学与人工智能实验室）；Microsoft Research（微软研究院）
Contrastive learning relies on an assumption that positive pairs contain related views, e.g., patches of an image or co-occurring multimodal signals of a video, that share certain underlying information about an instance. But what if this assumption is violated? The literature suggests that contrastive learning produces suboptimal representations in the presence of noisy views, e.g., false positive pairs with no apparent shared information. In this work, we propose a new contrastive loss function that is robust against noisy views. We provide rigorous theoretical justifications by showing conn…
Jiangsu University（江苏大学）；Jiangsu Engineering Research Center of Big Data Ubiquitous Perception and Intelligent Agriculture Applications（江苏省大数据泛在感知与智慧农业应用工程研究中心）；MIIT Key Laboratory of Pattern Analysis and Machine Intelligence（工信部模式分析与机器智能重点实验室）；Nanjing University of Aeronautics and Astronautics（南京航空航天大学）
Contrastive learning (CL) has shown impressive advances in image representation learning in whichever supervised multi-class classification or unsupervised learning. However, these CL methods fail to be directly adapted to multi-label image classification due to the difficulty in defining the positive and negative instances to contrast a given anchor image in multi-label scenario, let the label missing one alone, implying that borrowing a commonly-used way from contrastive multi-class learning to define them will incur a lot of false negative instances unfavorable for learning. In this paper, …
University of Tübingen（蒂宾根大学）；IMPRS for Intelligent Systems（智能系统马克斯·普朗克国际研究学院）；EPFL（洛桑联邦理工学院）
Contrastive learning has recently seen tremendous success in self-supervised learning. So far, however, it is largely unclear why the learned representations generalize so effectively to a large variety of downstream tasks. We here prove that feedforward models trained with objectives belonging to the commonly used InfoNCE family learn to implicitly invert the underlying generative model of the observed data. While the proofs make certain statistical assumptions about the generative model, we observe empirically that our findings hold even if these assumptions are severely violated. Our theory…
Unsupervised representation learning has recently received lots of interest due to its powerful generalizability through effectively leveraging large-scale unlabeled data. There are two prevalent approaches for this, contrastive learning and generative pre-training, where the former learns representations from instance-wise discrimination tasks and the latter learns them from estimating the likelihood. These seemingly orthogonal approaches have their own strengths and weaknesses. Contrastive learning tends to extract semantic information and discards details irrelevant for classifying objects,…
Xiamen University（厦门大学）；The Hong Kong University of Science and Technology（香港科技大学）；Peking University（北京大学）；The Hong Kong Polytechnic University（香港理工大学）；Manteia Technologies Co.,Ltd（Manteia科技有限公司）
Automatic delineation of organ-at-risk (OAR) and gross-tumor-volume (GTV) is of great significance for radiotherapy planning. However, it is a challenging task to learn powerful representations for accurate delineation under limited pixel (voxel)-wise annotations. Contrastive learning at pixel-level can alleviate the dependency on annotations by learning dense representations from unlabeled data. Recent studies in this direction design various contrastive losses on the feature maps, to yield discriminative features for each pixel in the map. However, pixels in the same map inevitably share sem…
We present modality gap, an intriguing geometric phenomenon of the representation space of multi-modal models. Specifically, we show that different data modalities (e.g. images and text) are embedded at arm's length in their shared representation in multi-modal models such as CLIP. Our systematic analysis demonstrates that this gap is caused by a combination of model initialization and contrastive learning optimization. In model initialization, we show empirically and theoretically that the representation of a common deep neural network is restricted to a narrow cone. As a consequence, in a mu…
Graduate School of Artificial Intelligence（人工智能研究生院）；POSTECH（浦项科技大学）
Conditional image generation is the task of generating diverse images using class label information. Although many conditional Generative Adversarial Networks (GAN) have shown realistic results, such methods consider pairwise relations between the embedding of an image and the embedding of the corresponding label (data-to-class relations) as the conditioning losses. In this paper, we propose ContraGAN that considers relations between multiple image embeddings in the same batch (data-to-data relations) as well as the data-to-class relations by using a conditional contrastive loss. The discrimin…
In reinforcement learning (RL), it is easier to solve a task if given a good representation. While deep RL should automatically acquire such good representations, prior work often finds that learning representations in an end-to-end fashion is unstable and instead equip RL algorithms with additional representation learning parts (e.g., auxiliary losses, data augmentation). How can we design RL algorithms that directly acquire good representations? In this paper, instead of adding representation learning parts to an existing RL algorithm, we show (contrastive) representation learning methods ca…
Notable progress has been made in numerous fields of machine learning based on neural network-driven mutual information (MI) bounds. However, utilizing the conventional MI-based losses is often challenging due to their practical and mathematical limitations. In this work, we first identify the symptoms behind their instability: (1) the neural network not converging even after the loss seemed to converge, and (2) saturating neural network outputs causing the loss to diverge. We mitigate both issues by adding a novel regularization term to the existing losses. We theoretically and experimentally…
The paper proposes a new algorithm called SymBa that aims to achieve more biologically plausible learning than Back-Propagation (BP). The algorithm is based on the Forward-Forward (FF) algorithm, which is a BP-free method for training neural networks. SymBa improves the FF algorithm's convergence behavior by addressing the problem of asymmetric gradients caused by conflicting converging directions for positive and negative samples. The algorithm balances positive and negative losses to enhance performance and convergence speed. Furthermore, it modifies the FF algorithm by adding Intrinsic Clas…
GRACE: Generative Representation Learning via Contrastive Policy Optimization
University of Illinois Urbana–Champaign（伊利诺伊大学厄巴纳-香槟分校）；Australian National University（澳大利亚国立大学）；Hong Kong University of Science and Technology（香港科学与技术大学）；University of Wisconsin–Madison（威斯康星大学麦迪逊分校）；University of Washington（华盛顿大学）
The Chinese University of Hong Kong（香港中文大学）；Australian National University（澳大利亚国立大学）；Data61/CSIRO（Data61/澳大利亚联邦科学与工业研究组织）
Although augmentations (e.g., perturbation of graph edges, image crops) boost the efficiency of Contrastive Learning (CL), feature level augmentation is another plausible, complementary yet not well researched strategy. Thus, we present a novel spectral feature argumentation for contrastive learning on graphs (and images). To this end, for each data view, we estimate a low-rank approximation per feature map and subtract that approximation from the map to obtain its complement. This is achieved by the proposed herein incomplete power iteration, a non-standard power iteration regime which enjoys…
