Neural Architecture Search (NAS) has been used recently to achieve improved performance in various tasks and most prominently in image classification. Yet, current search strategies rely on large labeled datasets, which limit their usage in the case where only a smaller fraction of the data is annotated. Self-supervised learning has shown great promise in training neural networks using unlabeled data. In this work, we propose a self-supervised neural architecture search (SSNAS) that allows finding novel network models without the need for labeled data. We show that such a search leads to comparable results to supervised training with a "fully labeled" NAS and that it can improve the performance of self-supervised learning. Moreover, we demonstrate the advantage of the proposed approach when the number of labels in the search is relatively small.
Deep neural networks are typically trained under a supervised learning framework where a model learns a single task using labeled data. Instead of relying solely on labeled data, practitioners can harness unlabeled or related data to improve model performance, which is often more accessible and ubiquitous. Self-supervised pre-training for transfer learning is becoming an increasingly popular technique to improve state-of-the-art results using unlabeled data. It involves first pre-training a model on a large amount of unlabeled data, then adapting the model to target tasks of interest. In this review, we survey self-supervised learning methods and their applications within the sequential transfer learning framework. We provide an overview of the taxonomy for self-supervised learning and transfer learning, and highlight some prominent methods for designing pre-training tasks across different domains. Finally, we discuss recent trends and suggest areas for future investigation.
We present a differentiable dynamics solver that is able to handle frictional contact for rigid and deformable objects within a unified framework. Through a principled mollification of normal and tangential contact forces, our method circumvents the main difficulties inherent to the non-smooth nature of frictional contact. We combine this new contact model with fully-implicit time integration to obtain a robust and efficient dynamics solver that is analytically differentiable. In conjunction with adjoint sensitivity analysis, our formulation enables gradient-based optimization with adaptive trade-offs between simulation accuracy and smoothness of objective function landscapes. We thoroughly analyse our approach on a set of simulation examples involving rigid bodies, visco-elastic materials, and coupled multi-body systems. We furthermore showcase applications of our differentiable simulator to parameter estimation for deformable objects, motion planning for robotic manipulation, trajectory optimization for compliant walking robots, as well as efficient self-supervised learning of control policies.
Convolutional Neural Networks (CNN)-based approaches have shown promising results in pansharpening of satellite images in recent years. However, they still exhibit limitations in producing high-quality pansharpening outputs. To that end, we propose a new self-supervised learning framework, where we treat pansharpening as a colorization problem, which brings an entirely novel perspective and solution to the problem compared to existing methods that base their solution solely on producing a super-resolution version of the multispectral image. Whereas CNN-based methods provide a reduced resolution panchromatic image as input to their model along with reduced resolution multispectral images, hence learn to increase their resolution together, we instead provide the grayscale transformed multispectral image as input, and train our model to learn the colorization of the grayscale input. We further address the fixed downscale ratio assumption during training, which does not generalize well to the full-resolution scenario. We introduce a noise injection into the training by randomly varying the downsampling ratios. Those two critical changes, along with the addition of adversarial training in the proposed PanColorization Generative Adversarial Networks (PanColorGAN) framework, help overcome the spatial detail loss and blur problems that are observed in CNN-based pansharpening. The proposed approach outperforms the previous CNN-based and traditional methods as demonstrated in our experiments.
We develop a model for representing visual texture in a low-dimensional feature space, along with a novel self-supervised learning objective that is used to train it on an unlabeled database of texture images. Inspired by the architecture of primate visual cortex, the model uses a first stage of oriented linear filters (corresponding to cortical area V1), consisting of both rectified units (simple cells) and pooled phase-invariant units (complex cells). These responses are processed by a second stage (analogous to cortical area V2) consisting of convolutional filters followed by half-wave rectification and pooling to generate V2 'complex cell' responses. The second stage filters are trained on a set of unlabeled homogeneous texture images, using a novel contrastive objective that maximizes the distance between the distribution of V2 responses to individual images and the distribution of responses across all images. When evaluated on texture classification, the trained model achieves substantially greater data-efficiency than a variety of deep hierarchical model architectures. Moreover, we show that the learned model exhibits stronger representational similarity to texture responses of neural populations recorded in primate V2 than pre-trained deep CNNs.
Self-supervised learning has proven to be invaluable in making best use of all of the available data in biomedical image segmentation. One particularly simple and effective mechanism to achieve self-supervision is inpainting, the task of predicting arbitrary missing areas based on the rest of an image. In this work, we focus on image inpainting as the self-supervised proxy task, and propose two novel structural changes to further enhance the performance of a deep neural network. We guide the process of generating images to inpaint by using supervoxel-based masking instead of random masking, and also by focusing on the area to be segmented in the primary task, which we term as the region-of-interest. We postulate that these additions force the network to learn semantics that are more attuned to the primary task, and test our hypotheses on two applications: brain tumour and white matter hyperintensities segmentation. We empirically show that our proposed approach consistently outperforms both supervised CNNs, without any self-supervision, and conventional inpainting-based self-supervision methods on both large and small training set sizes.
Resective surgery may be curative for drug-resistant focal epilepsy, but only 40% to 70% of patients achieve seizure freedom after surgery. Retrospective quantitative analysis could elucidate patterns in resected structures and patient outcomes to improve resective surgery. However, the resection cavity must first be segmented on the postoperative MR image. Convolutional neural networks (CNNs) are the state-of-the-art image segmentation technique, but require large amounts of annotated data for training. Annotation of medical images is a time-consuming process requiring highly-trained raters, and often suffering from high inter-rater variability. Self-supervised learning can be used to generate training instances from unlabeled data. We developed an algorithm to simulate resections on preoperative MR images. We curated a new dataset, EPISURG, comprising 431 postoperative and 269 preoperative MR images from 431 patients who underwent resective surgery. In addition to EPISURG, we used three public datasets comprising 1813 preoperative MR images for training. We trained a 3D CNN on artificially resected images created on the fly during training, using images from 1) EPISURG, 2) public datasets and 3) both. To evaluate trained models, we calculate Dice score (DSC) between model segmentations and 200 manual annotations performed by three human raters. The model trained on data with manual annotations obtained a median (interquartile range) DSC of 65.3 (30.6). The DSC of our best-performing model, trained with no manual annotations, is 81.7 (14.2). For comparison, inter-rater agreement between human annotators was 84.0 (9.9). We demonstrate a training method for CNNs using simulated resection cavities that can accurately segment real resection cavities, without manual annotations.
Graph Convolutional Networks (GCNs) show promising results for semi-supervised learning tasks on graphs, thus become favorable comparing with other approaches. Despite the remarkable success of GCNs, it is difficult to train GCNs with insufficient supervision. When labeled data are limited, the performance of GCNs becomes unsatisfying for low-degree nodes. While some prior work analyze successes and failures of GCNs on the entire model level, profiling GCNs on individual node level is still underexplored. In this paper, we analyze GCNs in regard to the node degree distribution. From empirical observation to theoretical proof, we confirm that GCNs are biased towards nodes with larger degrees with higher accuracy on them, even if high-degree nodes are underrepresented in most graphs. We further develop a novel Self-Supervised-Learning Degree-Specific GCN (SL-DSGC) that mitigate the degree-related biases of GCNs from model and data aspects. Firstly, we propose a degree-specific GCN layer that captures both discrepancies and similarities of nodes with different degrees, which reduces the inner model-aspect biases of GCNs caused by sharing the same parameters with all nodes. Secondly, we design a self-supervised-learning algorithm that creates pseudo labels with uncertainty scores on unlabeled nodes with a Bayesian neural network. Pseudo labels increase the chance of connecting to labeled neighbors for low-degree nodes, thus reducing the biases of GCNs from the data perspective. Uncertainty scores are further exploited to weight pseudo labels dynamically in the stochastic gradient descent for SL-DSGC. Experiments on three benchmark datasets show SL-DSGC not only outperforms state-of-the-art self-training/self-supervised-learning GCN methods, but also improves GCN accuracy dramatically for low-degree nodes.
In this paper, we address the problem of image anomaly detection and segmentation. Anomaly detection involves making a binary decision as to whether an input image contains an anomaly, and anomaly segmentation aims to locate the anomaly on the pixel level. Support vector data description (SVDD) is a long-standing algorithm used for an anomaly detection, and we extend its deep learning variant to the patch-based method using self-supervised learning. This extension enables anomaly segmentation and improves detection performance. As a result, anomaly detection and segmentation performances measured in AUROC on MVTec AD dataset increased by 9.8% and 7.0%, respectively, compared to the previous state-of-the-art methods. Our results indicate the efficacy of the proposed method and its potential for industrial application. Detailed analysis of the proposed method offers insights regarding its behavior, and the code is available online.
This paper proposes a simple self-supervised approach for learning a representation for visual correspondence from raw video. We cast correspondence as prediction of links in a space-time graph constructed from video. In this graph, the nodes are patches sampled from each frame, and nodes adjacent in time can share a directed edge. We learn a representation in which pairwise similarity defines transition probability of a random walk, so that long-range correspondence is computed as a walk along the graph. We optimize the representation to place high probability along paths of similarity. Targets for learning are formed without supervision, by cycle-consistency: the objective is to maximize the likelihood of returning to the initial node when walking along a graph constructed from a palindrome of frames. Thus, a single path-level constraint implicitly supervises chains of intermediate comparisons. When used as a similarity metric without adaptation, the learned representation outperforms the self-supervised state-of-the-art on label propagation tasks involving objects, semantic parts, and pose. Moreover, we demonstrate that a technique we call edge dropout, as well as self-supervised adaptation at test-time, further improve transfer for object-centric correspondence.
Self-supervised learning can significantly improve the performance of downstream tasks, however, the dimensions of learned representations normally lack explicit physical meanings. In this work, we propose a novel self-supervised approach to solve inverse problems by employing the corresponding physical forward process so that the learned representations can have explicit physical meanings. The proposed approach works in an analysis-by-synthesis manner to learn an inference network by iteratively sampling and training. At the sampling step, given observed data, the inference network is used to approximate the intractable posterior, from which we sample input parameters and feed them to a physical process to generate data in the observational space; At the training step, the same network is optimized with the sampled paired data. We prove the feasibility of the proposed method by tackling the acoustic-to-articulatory inversion problem to infer articulatory information from speech. Given an articulatory synthesizer, an inference model can be trained completely from scratch with random initialization. Our experiments demonstrate that the proposed method can converge steadily and the network learns to control the articulatory synthesizer to speak like a human. We also demonstrate that trained models can generalize well to unseen speakers or even new languages, and performance can be further improved through self-adaptation.
A molecular and cellular understanding of how SARS-CoV-2 variably infects and causes severe COVID-19 remains a bottleneck in developing interventions to end the pandemic. We sought to use deep learning to study the biology of SARS-CoV-2 infection and COVID-19 severity by identifying transcriptomic patterns and cell types associated with SARS-CoV-2 infection and COVID-19 severity. To do this, we developed a new approach to generating self-supervised edge features. We propose a model that builds on Graph Attention Networks (GAT), creates edge features using self-supervised learning, and ingests these edge features via a Set Transformer. This model achieves significant improvements in predicting the disease state of individual cells, given their transcriptome. We apply our model to single-cell RNA sequencing datasets of SARS-CoV-2 infected lung organoids and bronchoalveolar lavage fluid samples of patients with COVID-19, achieving state-of-the-art performance on both datasets with our model. We then borrow from the field of explainable AI (XAI) to identify the features (genes) and cell types that discriminate bystander vs. infected cells across time and moderate vs. severe COVID-19 disease. To the best of our knowledge, this represents the first application of deep learning to identifying the molecular and cellular determinants of SARS-CoV-2 infection and COVID-19 severity using single-cell omics data.
Most approaches in few-shot learning rely on costly annotated data related to the goal task domain during (pre-)training. Recently, unsupervised meta-learning methods have exchanged the annotation requirement for a reduction in few-shot classification performance. Simultaneously, in settings with realistic domain shift, common transfer learning has been shown to outperform supervised meta-learning. Building on these insights and on advances in self-supervised learning, we propose a transfer learning approach which constructs a metric embedding that clusters unlabeled prototypical samples and their augmentations closely together. This pre-trained embedding is a starting point for few-shot classification by summarizing class clusters and fine-tuning. We demonstrate that our self-supervised prototypical transfer learning approach ProtoTransfer outperforms state-of-the-art unsupervised meta-learning methods on few-shot tasks from the mini-ImageNet dataset. In few-shot experiments with domain shift, our approach even has comparable performance to supervised methods, but requires orders of magnitude fewer labels.
We address the problem of camera pose estimation in visual localization. Current regression-based methods for pose estimation are trained and evaluated scene-wise. They depend on the coordinate frame of the training dataset and show a low generalization across scenes and datasets. We identify the dataset shift an important barrier to generalization and consider transfer learning as an alternative way towards a better reuse of pose estimation models. We revise domain adaptation techniques for classification and extend them to camera pose estimation, which is a multi-regression task. We develop a deep adaptation network for learning scene-invariant image representations and use adversarial learning to generate such representations for model transfer. We enrich the network with self-supervised learning and use the adaptability theory to validate the existence of scene-invariant representation of images in two given scenes. We evaluate our network on two public datasets, Cambridge Landmarks and 7Scene, demonstrate its superiority over several baselines and compare to the state of the art methods.
Deep clustering against self-supervised learning is a very important and promising direction for unsupervised visual representation learning since it requires little domain knowledge to design pretext tasks. However, the key component, embedding clustering, limits its extension to the extremely large-scale dataset due to its prerequisite to save the global latent embedding of the entire dataset. In this work, we aim to make this framework more simple and elegant without performance decline. We propose an unsupervised image classification framework without using embedding clustering, which is very similar to standard supervised training manner. For detailed interpretation, we further analyze its relation with deep clustering and contrastive learning. Extensive experiments on ImageNet dataset have been conducted to prove the effectiveness of our method. Furthermore, the experiments on transfer learning benchmarks have verified its generalization to other downstream tasks, including multi-label image classification, object detection, semantic segmentation and few-shot image classification.
We show for the first time that learning powerful representations from speech audio alone followed by fine-tuning on transcribed speech can outperform the best semi-supervised methods while being conceptually simpler. wav2vec 2.0 masks the speech input in the latent space and solves a contrastive task defined over a quantization of the latent representations which are jointly learned. Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/other test sets. When lowering the amount of labeled data to one hour, wav2vec 2.0 outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8/8.2 WER. This demonstrates the feasibility of speech recognition with limited amounts of labeled data.
In the absence of large labelled datasets, self-supervised learning techniques can boost performance by learning useful representations from unlabelled data, which is often more readily available. However, there is often a domain shift between the unlabelled collection and the downstream target problem data. We show that by learning Bayesian instance weights for the unlabelled data, we can improve the downstream classification accuracy by prioritising the most useful instances. Additionally, we show that the training time can be reduced by discarding unnecessary datapoints. Our method, BetaDataWeighter is evaluated using the popular self-supervised rotation prediction task on STL-10 and Visual Decathlon. We compare to related instance weighting schemes, both hand-designed heuristics and meta-learning, as well as conventional self-supervised learning. BetaDataWeighter achieves both the highest average accuracy and rank across datasets, and on STL-10 it prunes up to 78% of unlabelled images without significant loss in accuracy, corresponding to over 50% reduction in training time.
The recent growth in the number of satellite images fosters the development of effective deep-learning techniques for Remote Sensing (RS). However, their full potential is untapped due to the lack of large annotated datasets. Such a problem is usually countered by fine-tuning a feature extractor that is previously trained on the ImageNet dataset. Unfortunately, the domain of natural images differs from the RS one, which hinders the final performance. In this work, we propose to learn meaningful representations from satellite imagery, leveraging its high-dimensionality spectral bands to reconstruct the visible colors. We conduct experiments on land cover classification (BigEarthNet) and West Nile Virus detection, showing that colorization is a solid pretext task for training a feature extractor. Furthermore, we qualitatively observe that guesses based on natural images and colorization rely on different parts of the input. This paves the way to an ensemble model that eventually outperforms both the above-mentioned techniques.
The success of deep learning notoriously requires larger amounts of costly annotated data. This has led to the development of self-supervised learning (SSL) that aims to alleviate this limitation by creating domain specific pretext tasks on unlabeled data. Simultaneously, there are increasing interests in generalizing deep learning to the graph domain in the form of graph neural networks (GNNs). GNNs can naturally utilize unlabeled nodes through the simple neighborhood aggregation that is unable to thoroughly make use of unlabeled nodes. Thus, we seek to harness SSL for GNNs to fully exploit the unlabeled data. Different from data instances in the image and text domains, nodes in graphs present unique structure information and they are inherently linked indicating not independent and identically distributed (or i.i.d.). Such complexity is a double-edged sword for SSL on graphs. On the one hand, it determines that it is challenging to adopt solutions from the image and text domains to graphs and dedicated efforts are desired. On the other hand, it provides rich information that enables us to build SSL from a variety of perspectives. Thus, in this paper, we first deepen our understandings on when, why, and which strategies of SSL work with GNNs by empirically studying numerous basic SSL pretext tasks on graphs. Inspired by deep insights from the empirical studies, we propose a new direction SelfTask to build advanced pretext tasks that are able to achieve state-of-the-art performance on various real-world datasets. The specific experimental settings to reproduce our results can be found in \url{https://github.com/ChandlerBang/SelfTask-GNN}.
A key requirement for the success of supervised deep learning is a large labeled dataset - a condition that is difficult to meet in medical image analysis. Self-supervised learning (SSL) can help in this regard by providing a strategy to pre-train a neural network with unlabeled data, followed by fine-tuning for a downstream task with limited annotations. Contrastive learning, a particular variant of SSL, is a powerful technique for learning image-level representations. In this work, we propose strategies for extending the contrastive learning framework for segmentation of volumetric medical images in the semi-supervised setting with limited annotations, by leveraging domain-specific and problem-specific cues. Specifically, we propose (1) novel contrasting strategies that leverage structural similarity across volumetric medical images (domain-specific cue) and (2) a local version of the contrastive loss to learn distinctive representations of local regions that are useful for per-pixel segmentation (problem-specific cue). We carry out an extensive evaluation on three Magnetic Resonance Imaging (MRI) datasets. In the limited annotation setting, the proposed method yields substantial improvements compared to other self-supervision and semi-supervised learning techniques. When combined with a simple data augmentation technique, the proposed method reaches within 8% of benchmark performance using only two labeled MRI volumes for training, corresponding to only 4% (for ACDC) of the training data used to train the benchmark. The code is made public at https://github.com/krishnabits001/domain_specific_cl.