PyTorchKR Papers

최신 논문

총 5,358편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2107.05233

UniSpeech at scale: An Empirical Study of Pre-training Method on Large-Scale Speech Recognition Dataset

Recently, there has been a vast interest in self-supervised learning (SSL) where the model is pre-trained on large scale unlabeled data and then fine-tuned on a small labeled dataset. The common wisdom is that SSL helps resource-limited tasks in which only a limited amount of labeled data is available. The benefit of SSL keeps diminishing when the labeled training data amount increases. To our best knowledge, at most a few thousand hours of labeled data was used in the study of SSL. In contrast, the industry usually uses tens of thousands of hours of labeled data to build high-accuracy speech recognition (ASR) systems for resource-rich languages. In this study, we take the challenge to investigate whether and how SSL can improve the ASR accuracy of a state-of-the-art production-scale Transformer-Transducer model, which was built with 65 thousand hours of anonymized labeled EN-US data.

arXiv:2107.04734공식 코드 · Papers with Code

Layer-wise Analysis of a Self-supervised Speech Representation Model

Recently proposed self-supervised learning approaches have been successful for pre-training speech representation models. The utility of these learned representations has been observed empirically, but not much has been studied about the type or extent of information encoded in the pre-trained representations themselves. Developing such insights can help understand the capabilities and limits of these models and enable the research community to more efficiently develop their usage for downstream applications. In this work, we begin to fill this gap by examining one recent and successful pre-trained model (wav2vec 2.0), via its intermediate representation vectors, using a suite of analysis tools. We use the metrics of canonical correlation, mutual information, and performance on simple downstream tasks with non-parametric probes, in order to (i) query for acoustic and linguistic information content, (ii) characterize the evolution of information across model layers, and (iii) understand how fine-tuning the model for automatic speech recognition (ASR) affects these observations. Our findings motivate modifying the fine-tuning protocol for ASR, which produces improved word error rates in a low-resource setting.

arXiv:2107.05624공식 코드 · Papers with Code

End-to-end Multi-modal Video Temporal Grounding

We address the problem of text-guided video temporal grounding, which aims to identify the time interval of a certain event based on a natural language description. Different from most existing methods that only consider RGB images as visual features, we propose a multi-modal framework to extract complementary information from videos. Specifically, we adopt RGB images for appearance, optical flow for motion, and depth maps for image structure. While RGB images provide abundant visual cues of certain events, the performance may be affected by background clutters. Therefore, we use optical flow to focus on large motion and depth maps to infer the scene configuration when the action is related to objects recognizable with their shapes. To integrate the three modalities more effectively and enable inter-modal learning, we design a dynamic fusion scheme with transformers to model the interactions between modalities. Furthermore, we apply intra-modal self-supervised learning to enhance feature representations across videos for each modality, which also facilitates multi-modal learning. We conduct extensive experiments on the Charades-STA and ActivityNet Captions datasets, and show that the proposed method performs favorably against state-of-the-art approaches.

arXiv:2107.04938

Deep Fiber Clustering: Anatomically Informed Unsupervised Deep Learning for Fast and Effective White Matter Parcellation

White matter fiber clustering (WMFC) enables parcellation of white matter tractography for applications such as disease classification and anatomical tract segmentation. However, the lack of ground truth and the ambiguity of fiber data (the points along a fiber can equivalently be represented in forward or reverse order) pose challenges to this task. We propose a novel WMFC framework based on unsupervised deep learning. We solve the unsupervised clustering problem as a self-supervised learning task. Specifically, we use a convolutional neural network to learn embeddings of input fibers, using pairwise fiber distances as pseudo annotations. This enables WMFC that is insensitive to fiber point ordering. In addition, anatomical coherence of fiber clusters is improved by incorporating brain anatomical segmentation data. The proposed framework enables outlier removal in a natural way by rejecting fibers with low cluster assignment probability. We train and evaluate our method using 200 datasets from the Human Connectome Project. Results demonstrate superior performance and efficiency of the proposed approach.

arXiv:2107.04886

Hierarchical Self-Supervised Learning for Medical Image Segmentation Based on Multi-Domain Data Aggregation

A large labeled dataset is a key to the success of supervised deep learning, but for medical image segmentation, it is highly challenging to obtain sufficient annotated images for model training. In many scenarios, unannotated images are abundant and easy to acquire. Self-supervised learning (SSL) has shown great potentials in exploiting raw data information and representation learning. In this paper, we propose Hierarchical Self-Supervised Learning (HSSL), a new self-supervised framework that boosts medical image segmentation by making good use of unannotated data. Unlike the current literature on task-specific self-supervised pretraining followed by supervised fine-tuning, we utilize SSL to learn task-agnostic knowledge from heterogeneous data for various medical image segmentation tasks. Specifically, we first aggregate a dataset from several medical challenges, then pre-train the network in a self-supervised manner, and finally fine-tune on labeled data. We develop a new loss function by combining contrastive loss and classification loss and pretrain an encoder-decoder architecture for segmentation tasks. Our extensive experiments show that multi-domain joint pre-training benefits downstream segmentation tasks and outperforms single-domain pre-training significantly. Compared to learning from scratch, our new method yields better performance on various tasks (e.g., +0.69% to +18.60% in Dice scores with 5% of annotated data). With limited amounts of training data, our method can substantially bridge the performance gap w.r.t. denser annotations (e.g., 10% vs.~100% of annotated data).

arXiv:2107.04227

Dropout Regularization for Self-Supervised Learning of Transformer Encoder Speech Representation

Predicting the altered acoustic frames is an effective way of self-supervised learning for speech representation. However, it is challenging to prevent the pretrained model from overfitting. In this paper, we proposed to introduce two dropout regularization methods into the pretraining of transformer encoder: (1) attention dropout, (2) layer dropout. Both of the two dropout methods encourage the model to utilize global speech information, and avoid just copying local spectrum features when reconstructing the masked frames. We evaluated the proposed methods on phoneme classification and speaker recognition tasks. The experiments demonstrate that our dropout approaches achieve competitive results, and improve the performance of classification accuracy on downstream tasks.

arXiv:2107.04082

Improved Language Identification Through Cross-Lingual Self-Supervised Learning

Language identification greatly impacts the success of downstream tasks such as automatic speech recognition. Recently, self-supervised speech representations learned by wav2vec 2.0 have been shown to be very effective for a range of speech tasks. We extend previous self-supervised work on language identification by experimenting with pre-trained models which were learned on real-world unconstrained speech in multiple languages and not just on English. We show that models pre-trained on many languages perform better and enable language identification systems that require very little labeled data to perform well. Results on a 26 languages setup show that with only 10 minutes of labeled data per language, a cross-lingually pre-trained model can achieve over 89.2% accuracy.

arXiv:2107.03758

Investigate the Essence of Long-Tailed Recognition from a Unified Perspective

As the data scale grows, deep recognition models often suffer from long-tailed data distributions due to the heavy imbalanced sample number across categories. Indeed, real-world data usually exhibit some similarity relation among different categories (e.g., pigeons and sparrows), called category similarity in this work. It is doubly difficult when the imbalance occurs between such categories with similar appearances. However, existing solutions mainly focus on the sample number to re-balance data distribution. In this work, we systematically investigate the essence of the long-tailed problem from a unified perspective. Specifically, we demonstrate that long-tailed recognition suffers from both sample number and category similarity. Intuitively, using a toy example, we first show that sample number is not the unique influence factor for performance dropping of long-tailed recognition. Theoretically, we demonstrate that (1) category similarity, as an inevitable factor, would also influence the model learning under long-tailed distribution via similar samples, (2) using more discriminative representation methods (e.g., self-supervised learning) for similarity reduction, the classifier bias can be further alleviated with greatly improved performance. Extensive experiments on several long-tailed datasets verify the rationality of our theoretical analysis, and show that based on existing state-of-the-arts (SOTAs), the performance could be further improved by similarity reduction. Our investigations highlight the essence behind the long-tailed problem, and claim several feasible directions for future work.

arXiv:2107.03019공식 코드 · Papers with Code

SelfCF: A Simple Framework for Self-supervised Collaborative Filtering

Collaborative filtering (CF) is widely used to learn informative latent representations of users and items from observed interactions. Existing CF-based methods commonly adopt negative sampling to discriminate different items. Training with negative sampling on large datasets is computationally expensive. Further, negative items should be carefully sampled under the defined distribution, in order to avoid selecting an observed positive item in the training dataset. Unavoidably, some negative items sampled from the training dataset could be positive in the test set. In this paper, we propose a self-supervised collaborative filtering framework (SelfCF), that is specially designed for recommender scenario with implicit feedback. The proposed SelfCF framework simplifies the Siamese networks and can be easily applied to existing deep-learning based CF models, which we refer to as backbone networks. The main idea of SelfCF is to augment the output embeddings generated by backbone networks, because it is infeasible to augment raw input of user/item ids. We propose and study three output perturbation techniques that can be applied to different types of backbone networks including both traditional CF models and graph-based models. The framework enables learning informative representations of users and items without negative samples, and is agnostic to the encapsulated backbones. We conduct comprehensive experiments on four datasets to show that our framework may achieve even better recommendation accuracy than the encapsulated supervised counterpart with a 2$\times$--4$\times$ faster training speed. We also show that SelfCF can boost up the accuracy by up to 17.79% on average, compared with a self-supervised framework BUIR.

arXiv:2107.02495

InfoNCE is variational inference in a recognition parameterised model

Here, we show that the InfoNCE objective is equivalent to the ELBO in a new class of probabilistic generative model, the recognition parameterised model (RPM). When we learn the optimal prior, the RPM ELBO becomes equal to the mutual information (MI; up to a constant), establishing a connection to pre-existing self-supervised learning methods such as InfoNCE. However, practical InfoNCE methods do not use the MI as an objective; the MI is invariant to arbitrary invertible transformations, so using an MI objective can lead to highly entangled representations (Tschannen et al., 2019). Instead, the actual InfoNCE objective is a simplified lower bound on the MI which is loose even in the infinite sample limit. Thus, an objective that works (i.e. the actual InfoNCE objective) appears to be motivated as a loose bound on an objective that does not work (i.e. the true MI which gives arbitrarily entangled representations). We give an alternative motivation for the actual InfoNCE objective. In particular, we show that in the infinite sample limit, and for a particular choice of prior, the actual InfoNCE objective is equal to the ELBO (up to a constant); and the ELBO is equal to the marginal likelihood with a deterministic recognition model. Thus, we argue that our VAE perspective gives a better motivation for InfoNCE than MI, as the actual InfoNCE objective is only loosely bounded by the MI, but is equal to the ELBO/marginal likelihood (up to a constant).

arXiv:2107.02639

Multi-Level Graph Contrastive Learning

Graph representation learning has attracted a surge of interest recently, whose target at learning discriminant embedding for each node in the graph. Most of these representation methods focus on supervised learning and heavily depend on label information. However, annotating graphs are expensive to obtain in the real world, especially in specialized domains (i.e. biology), as it needs the annotator to have the domain knowledge to label the graph. To approach this problem, self-supervised learning provides a feasible solution for graph representation learning. In this paper, we propose a Multi-Level Graph Contrastive Learning (MLGCL) framework for learning robust representation of graph data by contrasting space views of graphs. Specifically, we introduce a novel contrastive view - topological and feature space views. The original graph is first-order approximation structure and contains uncertainty or error, while the $k$NN graph generated by encoding features preserves high-order proximity. Thus $k$NN graph generated by encoding features not only provide a complementary view, but is more suitable to GNN encoder to extract discriminant representation. Furthermore, we develop a multi-level contrastive mode to preserve the local similarity and semantic similarity of graph-structured data simultaneously. Extensive experiments indicate MLGCL achieves promising results compared with the existing state-of-the-art graph representation learning methods on seven datasets.

arXiv:2107.01886

Self-Contrastive Learning with Hard Negative Sampling for Self-supervised Point Cloud Learning

Point clouds have attracted increasing attention. Significant progress has been made in methods for point cloud analysis, which often requires costly human annotation as supervision. To address this issue, we propose a novel self-contrastive learning for self-supervised point cloud representation learning, aiming to capture both local geometric patterns and nonlocal semantic primitives based on the nonlocal self-similarity of point clouds. The contributions are two-fold: on the one hand, instead of contrasting among different point clouds as commonly employed in contrastive learning, we exploit self-similar point cloud patches within a single point cloud as positive samples and otherwise negative ones to facilitate the task of contrastive learning. On the other hand, we actively learn hard negative samples that are close to positive samples for discriminative feature learning. Experimental results show that the proposed method achieves state-of-the-art performance on widely used benchmark datasets for self-supervised point cloud segmentation and transfer learning for classification.

arXiv:2107.01784공식 코드 · Papers with Code

Learning a Model for Inferring a Spatial Road Lane Network Graph using Self-Supervision

Interconnected road lanes are a central concept for navigating urban roads. Currently, most autonomous vehicles rely on preconstructed lane maps as designing an algorithmic model is difficult. However, the generation and maintenance of such maps is costly and hinders large-scale adoption of autonomous vehicle technology. This paper presents the first self-supervised learning method to train a model to infer a spatially grounded lane-level road network graph based on a dense segmented representation of the road scene generated from onboard sensors. A formal road lane network model is presented and proves that any structured road scene can be represented by a directed acyclic graph of at most depth three while retaining the notion of intersection regions, and that this is the most compressed representation. The formal model is implemented by a hybrid neural and search-based model, utilizing a novel barrier function loss formulation for robust learning from partial labels. Experiments are conducted for all common road intersection layouts. Results show that the model can generalize to new road layouts, unlike previous approaches, demonstrating its potential for real-world application as a practical learning-based lane-level map generator.

arXiv:2107.01776공식 코드 · Papers with Code

Continual Contrastive Learning for Image Classification

Recently, self-supervised representation learning gives further development in multimedia technology. Most existing self-supervised learning methods are applicable to packaged data. However, when it comes to streamed data, they are suffering from a catastrophic forgetting problem, which is not studied extensively. In this paper, we make the first attempt to tackle the catastrophic forgetting problem in the mainstream self-supervised methods, i.e., contrastive learning methods. Specifically, we first develop a rehearsal-based framework combined with a novel sampling strategy and a self-supervised knowledge distillation to transfer information over time efficiently. Then, we propose an extra sample queue to help the network separate the feature representations of old and new data in the embedding space. Experimental results show that compared with the naive self-supervised baseline, which learns tasks one by one without taking any technique, we improve the image classification accuracy by 1.60% on CIFAR-100, 2.86% on ImageNet-Sub, and 1.29% on ImageNet-Full under 10 incremental steps setting. Our code will be available at https://github.com/VDIGPKU/ContinualContrastiveLearning.

arXiv:2107.01691공식 코드 · Papers with Code

Bag of Instances Aggregation Boosts Self-supervised Distillation

Recent advances in self-supervised learning have experienced remarkable progress, especially for contrastive learning based methods, which regard each image as well as its augmentations as an individual class and try to distinguish them from all other images. However, due to the large quantity of exemplars, this kind of pretext task intrinsically suffers from slow convergence and is hard for optimization. This is especially true for small-scale models, in which we find the performance drops dramatically comparing with its supervised counterpart. In this paper, we propose a simple but effective distillation strategy for unsupervised learning. The highlight is that the relationship among similar samples counts and can be seamlessly transferred to the student to boost the performance. Our method, termed as BINGO, which is short for Bag of InstaNces aGgregatiOn, targets at transferring the relationship learned by the teacher to the student. Here bag of instances indicates a set of similar samples constructed by the teacher and are grouped within a bag, and the goal of distillation is to aggregate compact representations over the student with respect to instances in a bag. Notably, BINGO achieves new state-of-the-art performance on small-scale models, i.e., 65.5% and 68.9% top-1 accuracies with linear evaluation on ImageNet, using ResNet-18 and ResNet-34 as the backbones respectively, surpassing baselines (52.5% and 57.4% top-1 accuracies) by a significant margin. The code is available at https://github.com/haohang96/bingo.

arXiv:2107.00934

Hybrid Supervision Learning for Pathology Whole Slide Image Classification

Weak supervision learning on classification labels has demonstrated high performance in various tasks, while a few pixel-level fine annotations are also affordable. Naturally a question comes to us that whether the combination of pixel-level (e.g., segmentation) and image level (e.g., classification) annotation can introduce further improvement. However in computational pathology this is a difficult task for this reason: High resolution of whole slide images makes it difficult to do end-to-end classification model training, which is challenging to research of weak or hybrid supervision learning in the past. To handle this problem, we propose a hybrid supervision learning framework for this kind of high resolution images with sufficient image-level coarse annotations and a few pixel-level fine labels. This framework, when applied in training patch model, can carefully make use of coarse image-level labels to refine generated pixel-level pseudo labels. Complete strategy is proposed to suppress pixel-level false positives and false negatives. A large hybrid annotated dataset is used to evaluate the effectiveness of hybrid supervision learning. By extracting pixel-level pseudo labels in initially image-level labeled samples, we achieve 5.2% higher specificity than purely training on existing labels while retaining 100% sensitivity, in the task of image-level classification to be positive or negative.

arXiv:2107.00594공식 코드 · Papers with Code

Pretext Tasks selection for multitask self-supervised speech representation learning

Through solving pretext tasks, self-supervised learning leverages unlabeled data to extract useful latent representations replacing traditional input features in the downstream task. In audio/speech signal processing, a wide range of features where engineered through decades of research efforts. As it turns out, learning to predict such features (a.k.a pseudo-labels) has proven to be a particularly relevant pretext task, leading to useful self-supervised representations which prove to be effective for downstream tasks. However, methods and common practices for combining such pretext tasks for better performance on the downstream task have not been explored and understood properly. In fact, the process relies almost exclusively on a computationally heavy experimental procedure, which becomes intractable with the increase of the number of pretext tasks. This paper introduces a method to select a group of pretext tasks among a set of candidates. The method we propose estimates calibrated weights for the partial losses corresponding to the considered pretext tasks during the self-supervised training process. The experiments conducted on automatic speech recognition, speaker and emotion recognition validate our approach, as the groups selected and weighted with our method perform better than classic baselines, thus facilitating the selection and combination of relevant pseudo-labels for self-supervised representation learning.

arXiv:2106.16060

Leveraging Hidden Structure in Self-Supervised Learning

This work considers the problem of learning structured representations from raw images using self-supervised learning. We propose a principled framework based on a mutual information objective, which integrates self-supervised and structure learning. Furthermore, we devise a post-hoc procedure to interpret the meaning of the learnt representations. Preliminary experiments on CIFAR-10 show that the proposed framework achieves higher generalization performance in downstream classification tasks and provides more interpretable representations compared to the ones learnt through traditional self-supervised learning.

arXiv:2106.15395

Two-Stage Self-Supervised Cycle-Consistency Network for Reconstruction of Thin-Slice MR Images

The thick-slice magnetic resonance (MR) images are often structurally blurred in coronal and sagittal views, which causes harm to diagnosis and image post-processing. Deep learning (DL) has shown great potential to re-construct the high-resolution (HR) thin-slice MR images from those low-resolution (LR) cases, which we refer to as the slice interpolation task in this work. However, since it is generally difficult to sample abundant paired LR-HR MR images, the classical fully supervised DL-based models cannot be effectively trained to get robust performance. To this end, we propose a novel Two-stage Self-supervised Cycle-consistency Network (TSCNet) for MR slice interpolation, in which a two-stage self-supervised learning (SSL) strategy is developed for unsupervised DL network training. The paired LR-HR images are synthesized along the sagittal and coronal directions of input LR images for network pretraining in the first-stage SSL, and then a cyclic in-terpolation procedure based on triplet axial slices is designed in the second-stage SSL for further refinement. More training samples with rich contexts along all directions are exploited as guidance to guarantee the improved in-terpolation performance. Moreover, a new cycle-consistency constraint is proposed to supervise this cyclic procedure, which encourages the network to reconstruct more realistic HR images. The experimental results on a real MRI dataset indicate that TSCNet achieves superior performance over the conventional and other SSL-based algorithms, and obtains competitive quali-tative and quantitative results compared with the fully supervised algorithm.

arXiv:2106.15252공식 코드 · Papers with Code

AutoNovel: Automatically Discovering and Learning Novel Visual Categories

We tackle the problem of discovering novel classes in an image collection given labelled examples of other classes. We present a new approach called AutoNovel to address this problem by combining three ideas: (1) we suggest that the common approach of bootstrapping an image representation using the labelled data only introduces an unwanted bias, and that this can be avoided by using self-supervised learning to train the representation from scratch on the union of labelled and unlabelled data; (2) we use ranking statistics to transfer the model's knowledge of the labelled classes to the problem of clustering the unlabelled images; and, (3) we train the data representation by optimizing a joint objective function on the labelled and unlabelled subsets of the data, improving both the supervised classification of the labelled data, and the clustering of the unlabelled data. Moreover, we propose a method to estimate the number of classes for the case where the number of new categories is not known a priori. We evaluate AutoNovel on standard classification benchmarks and substantially outperform current methods for novel category discovery. In addition, we also show that AutoNovel can be used for fully unsupervised image clustering, achieving promising results.