Large-scale pre-trained models (PTMs) such as BERT and GPT have recently achieved great success and become a milestone in the field of artificial intelligence (AI). Owing to sophisticated pre-training objectives and huge model parameters, large-scale PTMs can effectively capture knowledge from massive labeled and unlabeled data. By storing knowledge into huge parameters and fine-tuning on specific tasks, the rich knowledge implicitly encoded in huge parameters can benefit a variety of downstream tasks, which has been extensively demonstrated via experimental verification and empirical analysis. It is now the consensus of the AI community to adopt PTMs as backbone for downstream tasks rather than learning models from scratch. In this paper, we take a deep look into the history of pre-training, especially its special relation with transfer learning and self-supervised learning, to reveal the crucial position of PTMs in the AI development spectrum. Further, we comprehensively review the latest breakthroughs of PTMs. These breakthroughs are driven by the surge of computational power and the increasing availability of data, towards four important directions: designing effective architectures, utilizing rich contexts, improving computational efficiency, and conducting interpretation and theoretical analysis. Finally, we discuss a series of open problems and research directions of PTMs, and hope our view can inspire and advance the future study of PTMs.
Multiple views of data, both naturally acquired (e.g., image and audio) and artificially produced (e.g., via adding different noise to data samples), have proven useful in enhancing representation learning. Natural views are often handled by multiview analysis tools, e.g., (deep) canonical correlation analysis [(D)CCA], while the artificial ones are frequently used in self-supervised learning (SSL) paradigms, e.g., BYOL and Barlow Twins. Both types of approaches often involve learning neural feature extractors such that the embeddings of data exhibit high cross-view correlations. Although intuitive, the effectiveness of correlation-based neural embedding is mostly empirically validated. This work aims to understand latent correlation maximization-based deep multiview learning from a latent component identification viewpoint. An intuitive generative model of multiview data is adopted, where the views are different nonlinear mixtures of shared and private components. Since the shared components are view/distortion-invariant, representing the data using such components is believed to reveal the identity of the samples effectively and robustly. Under this model, latent correlation maximization is shown to guarantee the extraction of the shared components across views (up to certain ambiguities). In addition, it is further shown that the private information in each view can be provably disentangled from the shared using proper regularization design. A finite sample analysis, which has been rare in nonlinear mixture identifiability study, is also presented. The theoretical results and newly designed regularization are tested on a series of tasks.
Effective molecular representation learning is of great importance to facilitate molecular property prediction, which is a fundamental task for the drug and material industry. Recent advances in graph neural networks (GNNs) have shown great promise in applying GNNs for molecular representation learning. Moreover, a few recent studies have also demonstrated successful applications of self-supervised learning methods to pre-train the GNNs to overcome the problem of insufficient labeled molecules. However, existing GNNs and pre-training strategies usually treat molecules as topological graph data without fully utilizing the molecular geometry information. Whereas, the three-dimensional (3D) spatial structure of a molecule, a.k.a molecular geometry, is one of the most critical factors for determining molecular physical, chemical, and biological properties. To this end, we propose a novel Geometry Enhanced Molecular representation learning method (GEM) for Chemical Representation Learning (ChemRL). At first, we design a geometry-based GNN architecture that simultaneously models atoms, bonds, and bond angles in a molecule. To be specific, we devised double graphs for a molecule: The first one encodes the atom-bond relations; The second one encodes bond-angle relations. Moreover, on top of the devised GNN architecture, we propose several novel geometry-level self-supervised learning strategies to learn spatial knowledge by utilizing the local and global molecular 3D structures. We compare ChemRL-GEM with various state-of-the-art (SOTA) baselines on different molecular benchmarks and exhibit that ChemRL-GEM can significantly outperform all baselines in both regression and classification tasks. For example, the experimental results show an overall improvement of 8.8% on average compared to SOTA baselines on the regression tasks, demonstrating the superiority of the proposed method.
Anomaly detection (AD), separating anomalies from normal data, has many applications across domains, from security to healthcare. While most previous works were shown to be effective for cases with fully or partially labeled data, that setting is in practice less common due to labeling being particularly tedious for this task. In this paper, we focus on fully unsupervised AD, in which the entire training dataset, containing both normal and anomalous samples, is unlabeled. To tackle this problem effectively, we propose to improve the robustness of one-class classification trained on self-supervised representations using a data refinement process. Our proposed data refinement approach is based on an ensemble of one-class classifiers (OCCs), each of which is trained on a disjoint subset of training data. Representations learned by self-supervised learning on the refined data are iteratively updated as the data refinement improves. We demonstrate our method on various unsupervised AD tasks with image and tabular data. With a 10% anomaly ratio on CIFAR-10 image data / 2.5% anomaly ratio on Thyroid tabular data, the proposed method outperforms the state-of-the-art one-class classifier by 6.3 AUC and 12.5 average precision / 22.9 F1-score.
Self-supervised speech representation learning (speech SSL) has demonstrated the benefit of scale in learning rich representations for Automatic Speech Recognition (ASR) with limited paired data, such as wav2vec 2.0. We investigate the existence of sparse subnetworks in pre-trained speech SSL models that achieve even better low-resource ASR results. However, directly applying widely adopted pruning methods such as the Lottery Ticket Hypothesis (LTH) is suboptimal in the computational cost needed. Moreover, we show that the discovered subnetworks yield minimal performance gain compared to the original dense network. We present Prune-Adjust-Re-Prune (PARP), which discovers and finetunes subnetworks for much better performance, while only requiring a single downstream ASR finetuning run. PARP is inspired by our surprising observation that subnetworks pruned for pre-training tasks need merely a slight adjustment to achieve a sizeable performance boost in downstream ASR tasks. Extensive experiments on low-resource ASR verify (1) sparse subnetworks exist in mono-lingual/multi-lingual pre-trained speech SSL, and (2) the computational advantage and performance gain of PARP over baseline pruning methods. In particular, on the 10min Librispeech split without LM decoding, PARP discovers subnetworks from wav2vec 2.0 with an absolute 10.9%/12.6% WER decrease compared to the full model. We further demonstrate the effectiveness of PARP via: cross-lingual pruning without any phone recognition degradation, the discovery of a multi-lingual subnetwork for 10 spoken languages in 1 finetuning run, and its applicability to pre-trained BERT/XLNet for natural language tasks.
Recent advances in representation learning have demonstrated an ability to represent information from different modalities such as video, text, and audio in a single high-level embedding vector. In this work we present a self-supervised learning framework that is able to learn a representation that captures finer levels of granularity across different modalities such as concepts or events represented by visual objects or spoken words. Our framework relies on a discretized embedding space created via vector quantization that is shared across different modalities. Beyond the shared embedding space, we propose a Cross-Modal Code Matching objective that forces the representations from different views (modalities) to have a similar distribution over the discrete embedding space such that cross-modal objects/actions localization can be performed without direct supervision. In our experiments we show that the proposed discretized multi-modal fine-grained representation (e.g., pixel/word/frame) can complement high-level summary representations (e.g., video/sentence/waveform) for improved performance on cross-modal retrieval tasks. We also observe that the discretized representation uses individual clusters to represent the same semantic concept across modalities.
Contrastive self-supervised learning has outperformed supervised pretraining on many downstream tasks like segmentation and object detection. However, current methods are still primarily applied to curated datasets like ImageNet. In this paper, we first study how biases in the dataset affect existing methods. Our results show that current contrastive approaches work surprisingly well across: (i) object- versus scene-centric, (ii) uniform versus long-tailed and (iii) general versus domain-specific datasets. Second, given the generality of the approach, we try to realize further gains with minor modifications. We show that learning additional invariances -- through the use of multi-scale cropping, stronger augmentations and nearest neighbors -- improves the representations. Finally, we observe that MoCo learns spatially structured representations when trained with a multi-crop strategy. The representations can be used for semantic segment retrieval and video instance segmentation without finetuning. Moreover, the results are on par with specialized models. We hope this work will serve as a useful study for other researchers. The code and models are available at https://github.com/wvangansbeke/Revisiting-Contrastive-SSL.
Self-supervised learning of graph neural networks (GNN) is in great need because of the widespread label scarcity issue in real-world graph/network data. Graph contrastive learning (GCL), by training GNNs to maximize the correspondence between the representations of the same graph in its different augmented forms, may yield robust and transferable GNNs even without using labels. However, GNNs trained by traditional GCL often risk capturing redundant graph features and thus may be brittle and provide sub-par performance in downstream tasks. Here, we propose a novel principle, termed adversarial-GCL (AD-GCL), which enables GNNs to avoid capturing redundant information during the training by optimizing adversarial graph augmentation strategies used in GCL. We pair AD-GCL with theoretical explanations and design a practical instantiation based on trainable edge-dropping graph augmentation. We experimentally validate AD-GCL by comparing with the state-of-the-art GCL methods and achieve performance gains of up-to $14\%$ in unsupervised, $6\%$ in transfer, and $3\%$ in semi-supervised learning settings overall with 18 different benchmark datasets for the tasks of molecule property regression and classification, and social network classification.
Transformer has been widely used for self-supervised pre-training in Natural Language Processing (NLP) and achieved great success. However, it has not been fully explored in visual self-supervised learning. Meanwhile, previous methods only consider the high-level feature and learning representation from a global perspective, which may fail to transfer to the downstream dense prediction tasks focusing on local features. In this paper, we present a novel Masked Self-supervised Transformer approach named MST, which can explicitly capture the local context of an image while preserving the global semantic information. Specifically, inspired by the Masked Language Modeling (MLM) in NLP, we propose a masked token strategy based on the multi-head self-attention map, which dynamically masks some tokens of local patches without damaging the crucial structure for self-supervised learning. More importantly, the masked tokens together with the remaining tokens are further recovered by a global image decoder, which preserves the spatial information of the image and is more friendly to the downstream dense prediction tasks. The experiments on multiple datasets demonstrate the effectiveness and generality of the proposed method. For instance, MST achieves Top-1 accuracy of 76.9% with DeiT-S only using 300-epoch pre-training by linear evaluation, which outperforms supervised methods with the same epoch by 0.4% and its comparable variant DINO by 1.0\%. For dense prediction tasks, MST also achieves 42.7% mAP on MS COCO object detection and 74.04% mIoU on Cityscapes segmentation only with 100-epoch pre-training.
Unsupervised domain adaptation (UDA) aims to transfer knowledge learned from a fully-labeled source domain to a different unlabeled target domain. Most existing UDA methods learn domain-invariant feature representations by minimizing feature distances across domains. In this work, we build upon contrastive self-supervised learning to align features so as to reduce the domain discrepancy between training and testing sets. Exploring the same set of categories shared by both domains, we introduce a simple yet effective framework CDCL, for domain alignment. In particular, given an anchor image from one domain, we minimize its distances to cross-domain samples from the same class relative to those from different categories. Since target labels are unavailable, we use a clustering-based approach with carefully initialized centers to produce pseudo labels. In addition, we demonstrate that CDCL is a general framework and can be adapted to the data-free setting, where the source data are unavailable during training, with minimal modification. We conduct experiments on two widely used domain adaptation benchmarks, i.e., Office-31 and VisDA-2017, for image classification tasks, and demonstrate that CDCL achieves state-of-the-art performance on both datasets.
Graph self-supervised learning has gained increasing attention due to its capacity to learn expressive node representations. Many pretext tasks, or loss functions have been designed from distinct perspectives. However, we observe that different pretext tasks affect downstream tasks differently cross datasets, which suggests that searching pretext tasks is crucial for graph self-supervised learning. Different from existing works focusing on designing single pretext tasks, this work aims to investigate how to automatically leverage multiple pretext tasks effectively. Nevertheless, evaluating representations derived from multiple pretext tasks without direct access to ground truth labels makes this problem challenging. To address this obstacle, we make use of a key principle of many real-world graphs, i.e., homophily, or the principle that "like attracts like," as the guidance to effectively search various self-supervised pretext tasks. We provide theoretical understanding and empirical evidence to justify the flexibility of homophily in this search task. Then we propose the AutoSSL framework which can automatically search over combinations of various self-supervised tasks. By evaluating the framework on 7 real-world datasets, our experimental results show that AutoSSL can significantly boost the performance on downstream tasks including node clustering and node classification compared with training under individual tasks. Code is released at https://github.com/ChandlerBang/AutoSSL.
Self-supervised learning, which benefits from automatically constructing labels through pre-designed pretext task, has recently been applied for strengthen supervised learning. Since previous self-supervised pretext tasks are based on input, they may incur huge additional training overhead. In this paper we find that features in CNNs can be also used for self-supervision. Thus we creatively design the \emph{feature-based pretext task} which requires only a small amount of additional training overhead. In our task we discard different particular regions of features, and then train the model to distinguish these different features. In order to fully apply our feature-based pretext task in supervised learning, we also propose a novel learning framework containing multi-classifiers for further improvement. Original labels will be expanded to joint labels via self-supervision of feature transformations. With more semantic information provided by our self-supervised tasks, this approach can train CNNs more effectively. Extensive experiments on various supervised learning tasks demonstrate the accuracy improvement and wide applicability of our method.
Traditional self-supervised learning requires CNNs using external pretext tasks (i.e., image- or video-based tasks) to encode high-level semantic visual representations. In this paper, we show that feature transformations within CNNs can also be regarded as supervisory signals to construct the self-supervised task, called \emph{internal pretext task}. And such a task can be applied for the enhancement of supervised learning. Specifically, we first transform the internal feature maps by discarding different channels, and then define an additional internal pretext task to identify the discarded channels. CNNs are trained to predict the joint labels generated by the combination of self-supervised labels and original labels. By doing so, we let CNNs know which channels are missing while classifying in the hope to mine richer feature information. Extensive experiments show that our approach is effective on various models and datasets. And it's worth noting that we only incur negligible computational overhead. Furthermore, our approach can also be compatible with other methods to get better results.
Electronic Health Records (EHR) have been heavily used in modern healthcare systems for recording patients' admission information to hospitals. Many data-driven approaches employ temporal features in EHR for predicting specific diseases, readmission times, or diagnoses of patients. However, most existing predictive models cannot fully utilize EHR data, due to an inherent lack of labels in supervised training for some temporal events. Moreover, it is hard for existing works to simultaneously provide generic and personalized interpretability. To address these challenges, we first propose a hyperbolic embedding method with information flow to pre-train medical code representations in a hierarchical structure. We incorporate these pre-trained representations into a graph neural network to detect disease complications, and design a multi-level attention method to compute the contributions of particular diseases and admissions, thus enhancing personalized interpretability. We present a new hierarchy-enhanced historical prediction proxy task in our self-supervised learning framework to fully utilize EHR data and exploit medical domain knowledge. We conduct a comprehensive set of experiments and case studies on widely used publicly available EHR datasets to verify the effectiveness of our model. The results demonstrate our model's strengths in both predictive tasks and interpretable abilities.
Self-supervised representation learning has shown remarkable success in a number of domains. A common practice is to perform data augmentation via hand-crafted transformations intended to leave the semantics of the data invariant. We seek to understand the empirical success of this approach from a theoretical perspective. We formulate the augmentation process as a latent variable model by postulating a partition of the latent representation into a content component, which is assumed invariant to augmentation, and a style component, which is allowed to change. Unlike prior work on disentanglement and independent component analysis, we allow for both nontrivial statistical and causal dependencies in the latent space. We study the identifiability of the latent representation based on pairs of views of the observations and prove sufficient conditions that allow us to identify the invariant content partition up to an invertible mapping in both generative and discriminative settings. We find numerical simulations with dependent latent variables are consistent with our theory. Lastly, we introduce Causal3DIdent, a dataset of high-dimensional, visually complex images with rich causal dependencies, which we use to study the effect of data augmentations performed in practice.
Recent works in self-supervised learning have advanced the state-of-the-art by relying on the contrastive learning paradigm, which learns representations by pushing positive pairs, or similar examples from the same class, closer together while keeping negative pairs far apart. Despite the empirical successes, theoretical foundations are limited -- prior analyses assume conditional independence of the positive pairs given the same class label, but recent empirical applications use heavily correlated positive pairs (i.e., data augmentations of the same image). Our work analyzes contrastive learning without assuming conditional independence of positive pairs using a novel concept of the augmentation graph on data. Edges in this graph connect augmentations of the same data, and ground-truth classes naturally form connected sub-graphs. We propose a loss that performs spectral decomposition on the population augmentation graph and can be succinctly written as a contrastive learning objective on neural net representations. Minimizing this objective leads to features with provable accuracy guarantees under linear probe evaluation. By standard generalization bounds, these accuracy guarantees also hold when minimizing the training contrastive loss. Empirically, the features learned by our objective can match or outperform several strong baselines on benchmark vision datasets. In all, this work provides the first provable analysis for contrastive learning where guarantees for linear probe evaluation can apply to realistic empirical settings.
As deep networks begin to be deployed as autonomous agents, the issue of how they can communicate with each other becomes important. Here, we train two deep nets from scratch to perform realistic referent identification through unsupervised emergent communication. We show that the largely interpretable emergent protocol allows the nets to successfully communicate even about object types they did not see at training time. The visual representations induced as a by-product of our training regime, moreover, show comparable quality, when re-used as generic visual features, to a recent self-supervised learning model. Our results provide concrete evidence of the viability of (interpretable) emergent deep net communication in a more realistic scenario than previously considered, as well as establishing an intriguing link between this field and self-supervised visual learning.
We present DistillFlow, a knowledge distillation approach to learning optical flow. DistillFlow trains multiple teacher models and a student model, where challenging transformations are applied to the input of the student model to generate hallucinated occlusions as well as less confident predictions. Then, a self-supervised learning framework is constructed: confident predictions from teacher models are served as annotations to guide the student model to learn optical flow for those less confident predictions. The self-supervised learning framework enables us to effectively learn optical flow from unlabeled data, not only for non-occluded pixels, but also for occluded pixels. DistillFlow achieves state-of-the-art unsupervised learning performance on both KITTI and Sintel datasets. Our self-supervised pre-trained model also provides an excellent initialization for supervised fine-tuning, suggesting an alternate training paradigm in contrast to current supervised learning methods that highly rely on pre-training on synthetic data. At the time of writing, our fine-tuned models ranked 1st among all monocular methods on the KITTI 2015 benchmark, and outperform all published methods on the Sintel Final benchmark. More importantly, we demonstrate the generalization capability of DistillFlow in three aspects: framework generalization, correspondence generalization and cross-dataset generalization.
Entity Matching (EM) aims at recognizing entity records that denote the same real-world object. Neural EM models learn vector representation of entity descriptions and match entities end-to-end. Though robust, these methods require many resources for training, and lack of interpretability. In this paper, we propose a novel EM framework that consists of Heterogeneous Information Fusion (HIF) and Key Attribute Tree (KAT) Induction to decouple feature representation from matching decision. Using self-supervised learning and mask mechanism in pre-trained language modeling, HIF learns the embeddings of noisy attribute values by inter-attribute attention with unlabeled data. Using a set of comparison features and a limited amount of annotated data, KAT Induction learns an efficient decision tree that can be interpreted by generating entity matching rules whose structure is advocated by domain experts. Experiments on 6 public datasets and 3 industrial datasets show that our method is highly efficient and outperforms SOTA EM models in most cases. Our codes and datasets can be obtained from https://github.com/THU-KEG/HIF-KAT.
Unsupervised Domain Adaptation (UDA) aims to generalize the knowledge learned from a well-labeled source domain to an unlabeled target domain. Recently, adversarial domain adaptation with two distinct classifiers (bi-classifier) has been introduced into UDA which is effective to align distributions between different domains. Previous bi-classifier adversarial learning methods only focus on the similarity between the outputs of two distinct classifiers. However, the similarity of the outputs cannot guarantee the accuracy of target samples, i.e., target samples may match to wrong categories even if the discrepancy between two classifiers is small. To challenge this issue, in this paper, we propose a cross-domain gradient discrepancy minimization (CGDM) method which explicitly minimizes the discrepancy of gradients generated by source samples and target samples. Specifically, the gradient gives a cue for the semantic information of target samples so it can be used as a good supervision to improve the accuracy of target samples. In order to compute the gradient signal of target samples, we further obtain target pseudo labels through a clustering-based self-supervised learning. Extensive experiments on three widely used UDA datasets show that our method surpasses many previous state-of-the-arts. Codes are available at https://github.com/lijin118/CGDM.