Designing robust loss functions is popular in learning with noisy labels while existing designs did not explicitly consider the overfitting property of deep neural networks (DNNs). As a result, applying these losses may still suffer from overfitting/memorizing noisy labels as training proceeds. In this paper, we first theoretically analyze the memorization effect and show that a lower-capacity model may perform better on noisy datasets. However, it is non-trivial to design a neural network with the best capacity given an arbitrary task. To circumvent this dilemma, instead of changing the model architecture, we decouple DNNs into an encoder followed by a linear classifier and propose to restrict the function space of a DNN by a representation regularizer. Particularly, we require the distance between two self-supervised features to be positively related to the distance between the corresponding two supervised model outputs. Our proposed framework is easily extendable and can incorporate many other robust loss functions to further improve performance. Extensive experiments and theoretical analyses support our claims. Code is available at github.com/UCSC-REAL/SelfSup_NoisyLabel.
Self-supervised learning is a promising unsupervised learning framework that has achieved success with large floating point networks. But such networks are not readily deployable to edge devices. To accelerate deployment of models with the benefit of unsupervised representation learning to such resource limited devices for various downstream tasks, we propose a self-supervised learning method for binary networks that uses a moving target network. In particular, we propose to jointly train a randomly initialized classifier, attached to a pretrained floating point feature extractor, with a binary network. Additionally, we propose a feature similarity loss, a dynamic loss balancing and modified multi-stage training to further improve the accuracy, and call our method BURN. Our empirical validations over five downstream tasks using seven datasets show that BURN outperforms self-supervised baselines for binary networks and sometimes outperforms supervised pretraining. Code is availabe at https://github.com/naver-ai/burn.
Self-supervised learning for depth estimation uses geometry in image sequences for supervision and shows promising results. Like many computer vision tasks, depth network performance is determined by the capability to learn accurate spatial and semantic representations from images. Therefore, it is natural to exploit semantic segmentation networks for depth estimation. In this work, based on a well-developed semantic segmentation network HRNet, we propose a novel depth estimation network DIFFNet, which can make use of semantic information in down and upsampling procedures. By applying feature fusion and an attention mechanism, our proposed method outperforms the state-of-the-art monocular depth estimation methods on the KITTI benchmark. Our method also demonstrates greater potential on higher resolution training data. We propose an additional extended evaluation strategy by establishing a test set of challenging cases, empirically derived from the standard benchmark.
Dimensionality reduction methods are unsupervised approaches which learn low-dimensional spaces where some properties of the initial space, typically the notion of "neighborhood", are preserved. Such methods usually require propagation on large k-NN graphs or complicated optimization solvers. On the other hand, self-supervised learning approaches, typically used to learn representations from scratch, rely on simple and more scalable frameworks for learning. In this paper, we propose TLDR, a dimensionality reduction method for generic input spaces that is porting the recent self-supervised learning framework of Zbontar et al. (2021) to the specific task of dimensionality reduction, over arbitrary representations. We propose to use nearest neighbors to build pairs from a training set and a redundancy reduction loss to learn an encoder that produces representations invariant across such pairs. TLDR is a method that is simple, easy to train, and of broad applicability; it consists of an offline nearest neighbor computation step that can be highly approximated, and a straightforward learning process. Aiming for scalability, we focus on improving linear dimensionality reduction, and show consistent gains on image and document retrieval tasks, e.g. gaining +4% mAP over PCA on ROxford for GeM- AP, improving the performance of DINO on ImageNet or retaining it with a 10x compression.
Self-supervised visual representation learning aims to learn useful representations without relying on human annotations. Joint embedding approach bases on maximizing the agreement between embedding vectors from different views of the same image. Various methods have been proposed to solve the collapsing problem where all embedding vectors collapse to a trivial constant solution. Among these methods, contrastive learning prevents collapse via negative sample pairs. It has been shown that non-contrastive methods suffer from a lesser collapse problem of a different nature: dimensional collapse, whereby the embedding vectors end up spanning a lower-dimensional subspace instead of the entire available embedding space. Here, we show that dimensional collapse also happens in contrastive learning. In this paper, we shed light on the dynamics at play in contrastive learning that leads to dimensional collapse. Inspired by our theory, we propose a novel contrastive learning method, called DirectCLR, which directly optimizes the representation space without relying on an explicit trainable projector. Experiments show that DirectCLR outperforms SimCLR with a trainable linear projector on ImageNet.
Multilingual end-to-end(E2E) models have shown a great potential in the expansion of the language coverage in the realm of automatic speech recognition(ASR). In this paper, we aim to enhance the multilingual ASR performance in two ways, 1)studying the impact of feeding a one-hot vector identifying the language, 2)formulating the task with a meta-learning objective combined with self-supervised learning (SSL). We associate every language with a distinct task manifold and attempt to improve the performance by transferring knowledge across learning processes itself as compared to transferring through final model parameters. We employ this strategy on a dataset comprising of 6 languages for an in-domain ASR task, by minimizing an objective related to expected gradient path length. Experimental results reveal the best pre-training strategy resulting in 3.55% relative reduction in overall WER. A combination of LEAP and SSL yields 3.51% relative reduction in overall WER when using language ID.
Continually learning in the real world must overcome many challenges, among which noisy labels are a common and inevitable issue. In this work, we present a repla-ybased continual learning framework that simultaneously addresses both catastrophic forgetting and noisy labels for the first time. Our solution is based on two observations; (i) forgetting can be mitigated even with noisy labels via self-supervised learning, and (ii) the purity of the replay buffer is crucial. Building on this regard, we propose two key components of our method: (i) a self-supervised replay technique named Self-Replay which can circumvent erroneous training signals arising from noisy labeled data, and (ii) the Self-Centered filter that maintains a purified replay buffer via centrality-based stochastic graph ensembles. The empirical results on MNIST, CIFAR-10, CIFAR-100, and WebVision with real-world noise demonstrate that our framework can maintain a highly pure replay buffer amidst noisy streamed data while greatly outperforming the combinations of the state-of-the-art continual learning and noisy label learning methods. The source code is available at http://vision.snu.ac.kr/projects/SPR
Molecular graph representation learning is a fundamental problem in modern drug and material discovery. Molecular graphs are typically modeled by their 2D topological structures, but it has been recently discovered that 3D geometric information plays a more vital role in predicting molecular functionalities. However, the lack of 3D information in real-world scenarios has significantly impeded the learning of geometric graph representation. To cope with this challenge, we propose the Graph Multi-View Pre-training (GraphMVP) framework where self-supervised learning (SSL) is performed by leveraging the correspondence and consistency between 2D topological structures and 3D geometric views. GraphMVP effectively learns a 2D molecular graph encoder that is enhanced by richer and more discriminative 3D geometry. We further provide theoretical insights to justify the effectiveness of GraphMVP. Finally, comprehensive experiments show that GraphMVP can consistently outperform existing graph SSL methods.
Video monitoring of traffic is useful for traffic management and control, traffic counting, and traffic law enforcement. However, traffic monitoring during inclement weather such as rain is a challenging task because video quality is corrupted by streaks of falling rain on the video image, and this hinders reliable characterization not only of the road environment but also of road-user behavior during such adverse weather events. This study proposes a two-stage self-supervised learning method to remove rain streaks in traffic videos. The first and second stages address intra- and inter-frame noise, respectively. The results indicated that the model exhibits satisfactory performance in terms of the image visual quality and the Peak Signal-Noise Ratio value.
We present TWIST, a simple and theoretically explainable self-supervised representation learning method by classifying large-scale unlabeled datasets in an end-to-end way. We employ a siamese network terminated by a softmax operation to produce twin class distributions of two augmented images. Without supervision, we enforce the class distributions of different augmentations to be consistent. However, simply minimizing the divergence between augmentations will cause collapsed solutions, i.e., outputting the same class probability distribution for all images. In this case, no information about the input image is left. To solve this problem, we propose to maximize the mutual information between the input and the class predictions. Specifically, we minimize the entropy of the distribution for each sample to make the class prediction for each sample assertive and maximize the entropy of the mean distribution to make the predictions of different samples diverse. In this way, TWIST can naturally avoid the collapsed solutions without specific designs such as asymmetric network, stop-gradient operation, or momentum encoder. As a result, TWIST outperforms state-of-the-art methods on a wide range of tasks. Especially, TWIST performs surprisingly well on semi-supervised learning, achieving 61.2% top-1 accuracy with 1% ImageNet labels using a ResNet-50 as backbone, surpassing previous best results by an absolute improvement of 6.2%. Codes and pre-trained models are given on: https://github.com/bytedance/TWIST
Contrastive learning of auditory and visual perception has been extremely successful when investigated individually. However, there are still major questions on how we could integrate principles learned from both domains to attain effective audiovisual representations. In this paper, we present a contrastive framework to learn audiovisual representations from unlabeled videos. The type and strength of augmentations utilized during self-supervised pre-training play a crucial role for contrastive frameworks to work sufficiently. Hence, we extensively investigate composition of temporal augmentations suitable for learning audiovisual representations; we find lossy spatio-temporal transformations that do not corrupt the temporal coherency of videos are the most effective. Furthermore, we show that the effectiveness of these transformations scales with higher temporal resolution and stronger transformation intensity. Compared to self-supervised models pre-trained on only sampling-based temporal augmentation, self-supervised models pre-trained with our temporal augmentations lead to approximately 6.5% gain on linear classifier performance on AVE dataset. Lastly, we show that despite their simplicity, our proposed transformations work well across self-supervised learning frameworks (SimSiam, MoCoV3, etc), and benchmark audiovisual dataset (AVE).
With the growth of the academic engines, the mining and analysis acquisition of massive researcher data, such as collaborator recommendation and researcher retrieval, has become indispensable. It can improve the quality of services and intelligence of academic engines. Most of the existing studies for researcher data mining focus on a single task for a particular application scenario and learning a task-specific model, which is usually unable to transfer to out-of-scope tasks. The pre-training technology provides a generalized and sharing model to capture valuable information from enormous unlabeled data. The model can accomplish multiple downstream tasks via a few fine-tuning steps. In this paper, we propose a multi-task self-supervised learning-based researcher data pre-training model named RPT. Specifically, we divide the researchers' data into semantic document sets and community graph. We design the hierarchical Transformer and the local community encoder to capture information from the two categories of data, respectively. Then, we propose three self-supervised learning objectives to train the whole model. Finally, we also propose two transfer modes of RPT for fine-tuning in different scenarios. We conduct extensive experiments to evaluate RPT, results on three downstream tasks verify the effectiveness of pre-training for researcher data mining.
Representation learning from unlabeled data has been of major interest in artificial intelligence research. While self-supervised speech representation learning has been popular in the speech research community, very few works have comprehensively analyzed audio representation learning for non-speech audio tasks. In this paper, we propose a self-supervised audio representation learning method and apply it to a variety of downstream non-speech audio tasks. We combine the well-known wav2vec 2.0 framework, which has shown success in self-supervised learning for speech tasks, with parameter-efficient conformer architectures. Our self-supervised pre-training can reduce the need for labeled data by two-thirds. On the AudioSet benchmark, we achieve a mean average precision (mAP) score of 0.415, which is a new state-of-the-art on this dataset through audio-only self-supervised learning. Our fine-tuned conformers also surpass or match the performance of previous systems pre-trained in a supervised way on several downstream tasks. We further discuss the important design considerations for both pre-training and fine-tuning.
Contrastive learning (CL) is one of the most successful paradigms for self-supervised learning (SSL). In a principled way, it considers two augmented "views" of the same image as positive to be pulled closer, and all other images as negative to be pushed further apart. However, behind the impressive success of CL-based techniques, their formulation often relies on heavy-computation settings, including large sample batches, extensive training epochs, etc. We are thus motivated to tackle these issues and establish a simple, efficient, yet competitive baseline of contrastive learning. Specifically, we identify, from theoretical and empirical studies, a noticeable negative-positive-coupling (NPC) effect in the widely used InfoNCE loss, leading to unsuitable learning efficiency concerning the batch size. By removing the NPC effect, we propose decoupled contrastive learning (DCL) loss, which removes the positive term from the denominator and significantly improves the learning efficiency. DCL achieves competitive performance with less sensitivity to sub-optimal hyperparameters, requiring neither large batches in SimCLR, momentum encoding in MoCo, or large epochs. We demonstrate with various benchmarks while manifesting robustness as much less sensitive to suboptimal hyperparameters. Notably, SimCLR with DCL achieves 68.2% ImageNet-1K top-1 accuracy using batch size 256 within 200 epochs pre-training, outperforming its SimCLR baseline by 6.4%. Further, DCL can be combined with the SOTA contrastive learning method, NNCLR, to achieve 72.3% ImageNet-1K top-1 accuracy with 512 batch size in 400 epochs, which represents a new SOTA in contrastive learning. We believe DCL provides a valuable baseline for future contrastive SSL studies.
Deep learning has shown remarkable progress in a wide range of problems. However, efficient training of such models requires large-scale datasets, and getting annotations for such datasets can be challenging and costly. In this work, we explore the use of user-generated freely available labels from web videos for video understanding. We create a benchmark dataset consisting of around 2 million videos with associated user-generated annotations and other meta information. We utilize the collected dataset for action classification and demonstrate its usefulness with existing small-scale annotated datasets, UCF101 and HMDB51. We study different loss functions and two pretraining strategies, simple and self-supervised learning. We also show how a network pretrained on the proposed dataset can help against video corruption and label noise in downstream datasets. We present this as a benchmark dataset in noisy learning for video understanding. The dataset, code, and trained models will be publicly available for future research.
Estimating the motion of the camera together with the 3D structure of the scene from a monocular vision system is a complex task that often relies on the so-called scene rigidity assumption. When observing a dynamic environment, this assumption is violated which leads to an ambiguity between the ego-motion of the camera and the motion of the objects. To solve this problem, we present a self-supervised learning framework for 3D object motion field estimation from monocular videos. Our contributions are two-fold. First, we propose a two-stage projection pipeline to explicitly disentangle the camera ego-motion and the object motions with dynamics attention module, called DAM. Specifically, we design an integrated motion model that estimates the motion of the camera and object in the first and second warping stages, respectively, controlled by the attention module through a shared motion encoder. Second, we propose an object motion field estimation through contrastive sample consensus, called CSAC, taking advantage of weak semantic prior (bounding box from an object detector) and geometric constraints (each object respects the rigid body motion model). Experiments on KITTI, Cityscapes, and Waymo Open Dataset demonstrate the relevance of our approach and show that our method outperforms state-of-the-art algorithms for the tasks of self-supervised monocular depth estimation, object motion segmentation, monocular scene flow estimation, and visual odometry.
Self-supervised learning (SSL) is a long-standing goal for speech processing, since it utilizes large-scale unlabeled data and avoids extensive human labeling. Recent years witness great successes in applying self-supervised learning in speech recognition, while limited exploration was attempted in applying SSL for modeling speaker characteristics. In this paper, we aim to improve the existing SSL framework for speaker representation learning. Two methods are introduced for enhancing the unsupervised speaker information extraction. First, we apply the multi-task learning to the current SSL framework, where we integrate the utterance-wise contrastive loss with the SSL objective function. Second, for better speaker discrimination, we propose an utterance mixing strategy for data augmentation, where additional overlapped utterances are created unsupervisely and incorporate during training. We integrate the proposed methods into the HuBERT framework. Experiment results on SUPERB benchmark show that the proposed system achieves state-of-the-art performance in universal representation learning, especially for speaker identification oriented tasks. An ablation study is performed verifying the efficacy of each proposed method. Finally, we scale up training dataset to 94 thousand hours public audio data and achieve further performance improvement in all SUPERB tasks.
Self-supervised learning holds promise to revolutionize molecule property prediction - a central task to drug discovery and many more industries - by enabling data efficient learning from scarce experimental data. Despite significant progress, non-pretrained methods can be still competitive in certain settings. We reason that architecture might be a key bottleneck. In particular, enriching the backbone architecture with domain-specific inductive biases has been key for the success of self-supervised learning in other domains. In this spirit, we methodologically explore the design space of the self-attention mechanism tailored to molecular data. We identify a novel variant of self-attention adapted to processing molecules, inspired by the relative self-attention layer, which involves fusing embedded graph and distance relationships between atoms. Our main contribution is Relative Molecule Attention Transformer (R-MAT): a novel Transformer-based model based on the developed self-attention layer that achieves state-of-the-art or very competitive results across a~wide range of molecule property prediction tasks.
Wav2vec 2.0 is an end-to-end framework of self-supervised learning for speech representation that is successful in automatic speech recognition (ASR), but most of the work on the topic has been developed with a single language: English. Therefore, it is unclear whether the self-supervised framework is effective in recognizing other languages with different writing systems, such as Korean which uses the Hangul having a unique writing system. In this paper, we present K-Wav2Vec 2.0, which is a modified version of Wav2vec 2.0 designed for Korean automatic speech recognition by exploring and optimizing various factors of the original Wav2vec 2.0. In fine-tuning, we propose a multi-task hierarchical architecture to reflect the Korean writing structure. Moreover, a joint decoder is applied to alleviate the problem of words existing outside of the vocabulary. In pre-training, we attempted the cross-lingual transfer of the pre-trained model by further pre-training the English Wav2vec 2.0 on a Korean dataset, considering limited resources. Our experimental results demonstrate that the proposed method yields the best performance on both Korean ASR datasets: Ksponspeech (a large-scale Korean speech corpus) and Clovacall (a call-based dialog corpus). Further pre-training is also effective in language adaptation, leading to large improvements without additional data.
The goal of self-supervised learning (SSL) for automatic speech recognition (ASR) is to learn good speech representations from a large amount of unlabeled speech for the downstream ASR task. However, most SSL frameworks do not consider noise robustness which is crucial for real-world applications. In this paper we propose wav2vec-Switch, a method to encode noise robustness into contextualized representations of speech via contrastive learning. Specifically, we feed original-noisy speech pairs simultaneously into the wav2vec 2.0 network. In addition to the existing contrastive learning task, we switch the quantized representations of the original and noisy speech as additional prediction targets of each other. By doing this, it enforces the network to have consistent predictions for the original and noisy speech, thus allows to learn contextualized representation with noise robustness. Our experiments on synthesized and real noisy data show the effectiveness of our method: it achieves 2.9--4.9% relative word error rate (WER) reduction on the synthesized noisy LibriSpeech data without deterioration on the original data, and 5.7% on CHiME-4 real 1-channel noisy data compared to a data augmentation baseline even with a strong language model for decoding. Our results on CHiME-4 can match or even surpass those with well-designed speech enhancement components.