Self-supervised learning for time-series data holds potential similar to that recently unleashed in Natural Language Processing and Computer Vision. While most existing works in this area focus on contrastive learning, we propose a conceptually simple yet powerful non-contrastive approach, based on the data2vec self-distillation framework. The core of our method is a student-teacher scheme that predicts the latent representation of an input time series from masked views of the same time series. This strategy avoids strong modality-specific assumptions and biases typically introduced by the design of contrastive sample pairs. We demonstrate the competitiveness of our approach for classification and forecasting as downstream tasks, comparing with state-of-the-art self-supervised learning methods on the UCR and UEA archives as well as the ETT and Electricity datasets.
This paper introduces a self-supervised learning framework designed for pre-training neural networks tailored to dense prediction tasks using event camera data. Our approach utilizes solely event data for training. Transferring achievements from dense RGB pre-training directly to event camera data yields subpar performance. This is attributed to the spatial sparsity inherent in an event image (converted from event data), where many pixels do not contain information. To mitigate this sparsity issue, we encode an event image into event patch features, automatically mine contextual similarity relationships among patches, group the patch features into distinctive contexts, and enforce context-to-context similarities to learn discriminative event features. For training our framework, we curate a synthetic event camera dataset featuring diverse scene and motion patterns. Transfer learning performance on downstream dense prediction tasks illustrates the superiority of our method over state-of-the-art approaches.
The process of annotating relevant data in the field of digital microscopy can be both time-consuming and especially expensive due to the required technical skills and human-expert knowledge. Consequently, large amounts of microscopic image data sets remain unlabeled, preventing their effective exploitation using deep-learning algorithms. In recent years it has been shown that a lot of relevant information can be drawn from unlabeled data. Self-supervised learning (SSL) is a promising solution based on learning intrinsic features under a pretext task that is similar to the main task without requiring labels. The trained result is transferred to the main task - image segmentation in our case. A ResNet50 U-Net was first trained to restore images of liver progenitor organoids from augmented images using the Structural Similarity Index Metric (SSIM), alone, and using SSIM combined with L1 loss. Both the encoder and decoder were trained in tandem. The weights were transferred to another U-Net model designed for segmentation with frozen encoder weights, using Binary Cross Entropy, Dice, and Intersection over Union (IoU) losses. For comparison, we used the same U-Net architecture to train two supervised models, one utilizing the ResNet50 encoder as well as a simple CNN. Results showed that self-supervised learning models using a 25\% pixel drop or image blurring augmentation performed better than the other augmentation techniques using the IoU loss. When trained on only 114 images for the main task, the self-supervised learning approach outperforms the supervised method achieving an F1-score of 0.85, with higher stability, in contrast to an F1=0.78 scored by the supervised method. Furthermore, when trained with larger data sets (1,000 images), self-supervised learning is still able to perform better, achieving an F1-score of 0.92, contrasting to a score of 0.85 for the supervised method.
In this paper, we present ShapeMatcher, a unified self-supervised learning framework for joint shape canonicalization, segmentation, retrieval and deformation. Given a partially-observed object in an arbitrary pose, we first canonicalize the object by extracting point-wise affine-invariant features, disentangling inherent structure of the object with its pose and size. These learned features are then leveraged to predict semantically consistent part segmentation and corresponding part centers. Next, our lightweight retrieval module aggregates the features within each part as its retrieval token and compare all the tokens with source shapes from a pre-established database to identify the most geometrically similar shape. Finally, we deform the retrieved shape in the deformation module to tightly fit the input object by harnessing part center guided neural cage deformation. The key insight of ShapeMaker is the simultaneous training of the four highly-associated processes: canonicalization, segmentation, retrieval, and deformation, leveraging cross-task consistency losses for mutual supervision. Extensive experiments on synthetic datasets PartNet, ComplementMe, and real-world dataset Scan2CAD demonstrate that ShapeMaker surpasses competitors by a large margin.
In recent years, the field of 3D self-supervised learning has witnessed significant progress, resulting in the emergence of Multi-Modality Masked AutoEncoders (MAE) methods that leverage both 2D images and 3D point clouds for pre-training. However, a notable limitation of these approaches is that they do not fully utilize the multi-view attributes inherent in 3D point clouds, which is crucial for a deeper understanding of 3D structures. Building upon this insight, we introduce a novel approach employing a 3D to multi-view masked autoencoder to fully harness the multi-modal attributes of 3D point clouds. To be specific, our method uses the encoded tokens from 3D masked point clouds to generate original point clouds and multi-view depth images across various poses. This approach not only enriches the model's comprehension of geometric structures but also leverages the inherent multi-modal properties of point clouds. Our experiments illustrate the effectiveness of the proposed method for different tasks and under different settings. Remarkably, our method outperforms state-of-the-art counterparts by a large margin in a variety of downstream tasks, including 3D object classification, few-shot learning, part segmentation, and 3D object detection. Code will be available at: https://github.com/Zhimin-C/Multiview-MAE
Weakly supervised whole slide image classification is a key task in computational pathology, which involves predicting a slide-level label from a set of image patches constituting the slide. Constructing models to solve this task involves multiple design choices, often made without robust empirical or conclusive theoretical justification. To address this, we conduct a comprehensive benchmarking of feature extractors to answer three critical questions: 1) Is stain normalisation still a necessary preprocessing step? 2) Which feature extractors are best for downstream slide-level classification? 3) How does magnification affect downstream performance? Our study constitutes the most comprehensive evaluation of publicly available pathology feature extractors to date, involving more than 10,000 training runs across 14 feature extractors, 9 tasks, 5 datasets, 3 downstream architectures, 2 levels of magnification, and various preprocessing setups. Our findings challenge existing assumptions: 1) We observe empirically, and by analysing the latent space, that skipping stain normalisation and image augmentations does not degrade performance, while significantly reducing memory and computational demands. 2) We develop a novel evaluation metric to compare relative downstream performance, and show that the choice of feature extractor is the most consequential factor for downstream performance. 3) We find that lower-magnification slides are sufficient for accurate slide-level classification. Contrary to previous patch-level benchmarking studies, our approach emphasises clinical relevance by focusing on slide-level biomarker prediction tasks in a weakly supervised setting with external validation cohorts. Our findings stand to streamline digital pathology workflows by minimising preprocessing needs and informing the selection of feature extractors.
Content-based image retrieval (CBIR) with self-supervised learning (SSL) accelerates clinicians' interpretation of similar images without manual annotations. We develop a CBIR from the contrastive learning SimCLR and incorporate a generalized-mean (GeM) pooling followed by L2 normalization to classify lesion types and retrieve similar images before clinicians' analysis. Results have shown improved performance. We additionally build an open-source application for image analysis and retrieval. The application is easy to integrate, relieving manual efforts and suggesting the potential to support clinicians' everyday activities.
The ongoing challenges in time series anomaly detection (TSAD), notably the scarcity of anomaly labels and the variability in anomaly lengths and shapes, have led to the need for a more efficient solution. As limited anomaly labels hinder traditional supervised models in TSAD, various SOTA deep learning techniques, such as self-supervised learning, have been introduced to tackle this issue. However, they encounter difficulties handling variations in anomaly lengths and shapes, limiting their adaptability to diverse anomalies. Additionally, many benchmark datasets suffer from the problem of having explicit anomalies that even random functions can detect. This problem is exacerbated by ill-posed evaluation metrics, known as point adjustment (PA), which can result in inflated model performance. In this context, we propose a novel self-supervised learning based Tri-domain Anomaly Detector (TriAD), which addresses these challenges by modeling features across three data domains - temporal, frequency, and residual domains - without relying on anomaly labels. Unlike traditional contrastive learning methods, TriAD employs both inter-domain and intra-domain contrastive loss to learn common attributes among normal data and differentiate them from anomalies. Additionally, our approach can detect anomalies of varying lengths by integrating with a discord discovery algorithm. It is worth noting that this study is the first to reevaluate the deep learning potential in TSAD, utilizing both rigorously designed datasets (i.e., UCR Archive) and evaluation metrics (i.e., PA%K and affiliation). Through experimental results on the UCR dataset, TriAD achieves an impressive three-fold increase in PA%K based F1 scores over SOTA deep learning models, and 50% increase of accuracy as compared to SOTA discord discovery algorithms.
Current speaker anonymization methods, especially with self-supervised learning (SSL) models, require massive computational resources when hiding speaker identity. This paper proposes an effective and parameter-efficient speaker anonymization method based on recent End-to-End model reprogramming technology. To improve the anonymization performance, we first extract speaker representation from large SSL models as the speaker identifies. To hide the speaker's identity, we reprogram the speaker representation by adapting the speaker to a pseudo domain. Extensive experiments are carried out on the VoicePrivacy Challenge (VPC) 2022 datasets to demonstrate the effectiveness of our proposed parameter-efficient learning anonymization methods. Additionally, while achieving comparable performance with the VPC 2022 strong baseline 1.b, our approach consumes less computational resources during anonymization.
Recently, researchers have shown an increasing interest in automatically predicting the subjective evaluation for speech synthesis systems. This prediction is a challenging task, especially on the out-of-domain test set. In this paper, we proposed a novel fusion model for MOS prediction that combines supervised and unsupervised approaches. In the supervised aspect, we developed an SSL-based predictor called LE-SSL-MOS. The LE-SSL-MOS utilizes pre-trained self-supervised learning models and further improves prediction accuracy by utilizing the opinion scores of each utterance in the listener enhancement branch. In the unsupervised aspect, two steps are contained: we fine-tuned the unit language model (ULM) using highly intelligible domain data to improve the correlation of an unsupervised metric - SpeechLMScore. Another is that we utilized ASR confidence as a new metric with the help of ensemble learning. To our knowledge, this is the first architecture that fuses supervised and unsupervised methods for MOS prediction. With these approaches, our experimental results on the VoiceMOS Challenge 2023 show that LE-SSL-MOS performs better than the baseline. Our fusion system achieved an absolute improvement of 13% over LE-SSL-MOS on the noisy and enhanced speech track. Our system ranked 1st and 2nd, respectively, in the French speech synthesis track and the challenge's noisy and enhanced speech track.
Panoptic segmentation is an important computer vision task which combines semantic and instance segmentation. It plays a crucial role in domains of medical image analysis, self-driving vehicles, and robotics by providing a comprehensive understanding of visual environments. Traditionally, deep learning panoptic segmentation models have relied on dense and accurately annotated training data, which is expensive and time consuming to obtain. Recent advancements in self-supervised learning approaches have shown great potential in leveraging synthetic and unlabelled data to generate pseudo-labels using self-training to improve the performance of instance and semantic segmentation models. The three available methods for self-supervised panoptic segmentation use proposal-based transformer architectures which are computationally expensive, complicated and engineered for specific tasks. The aim of this work is to develop a framework to perform embedding-based self-supervised panoptic segmentation using self-training in a synthetic-to-real domain adaptation problem setting.
Advancements in clinical treatment are increasingly constrained by the limitations of supervised learning techniques, which depend heavily on large volumes of annotated data. The annotation process is not only costly but also demands substantial time from clinical specialists. Addressing this issue, we introduce the S4MI (Self-Supervision and Semi-Supervision for Medical Imaging) pipeline, a novel approach that leverages advancements in self-supervised and semi-supervised learning. These techniques engage in auxiliary tasks that do not require labeling, thus simplifying the scaling of machine supervision compared to fully-supervised methods. Our study benchmarks these techniques on three distinct medical imaging datasets to evaluate their effectiveness in classification and segmentation tasks. Notably, we observed that self supervised learning significantly surpassed the performance of supervised methods in the classification of all evaluated datasets. Remarkably, the semi-supervised approach demonstrated superior outcomes in segmentation, outperforming fully-supervised methods while using 50% fewer labels across all datasets. In line with our commitment to contributing to the scientific community, we have made the S4MI code openly accessible, allowing for broader application and further development of these methods.
Contrastive, self-supervised learning (SSL) is used to train a model that predicts cancer type from miRNA, mRNA or RPPA expression data. This model, a pretrained FT-Transformer, is shown to outperform XGBoost and CatBoost, standard benchmarks for tabular data, when labelled samples are scarce but the number of unlabelled samples is high. This is despite the fact that the datasets we use have $\mathcal{O}(10^{1})$ classes and $\mathcal{O}(10^{2})-\mathcal{O}(10^{4})$ features. After demonstrating the efficacy of our chosen method of self-supervised pretraining, we investigate SSL for multi-modal models. A late-fusion model is proposed, where each omics is passed through its own sub-network, the outputs of which are averaged and passed to the pretraining or downstream objective function. Multi-modal pretraining is shown to improve predictions from a single omics, and we argue that this is useful for datasets with many unlabelled multi-modal samples, but few labelled unimodal samples. Additionally, we show that pretraining each omics-specific module individually is highly effective. This enables the application of the proposed model in a variety of contexts where a large amount of unlabelled data is available from each omics, but only a few labelled samples.
Table-to-text generation involves generating appropriate textual descriptions given structured tabular data. It has attracted increasing attention in recent years thanks to the popularity of neural network models and the availability of large-scale datasets. A common feature across existing methods is their treatment of the input as a string, i.e., by employing linearization techniques that do not always preserve information in the table, are verbose, and lack space efficiency. We propose to rethink data-to-text generation as a visual recognition task, removing the need for rendering the input in a string format. We present PixT3, a multimodal table-to-text model that overcomes the challenges of linearization and input size limitations encountered by existing models. PixT3 is trained with a new self-supervised learning objective to reinforce table structure awareness and is applicable to open-ended and controlled generation settings. Experiments on the ToTTo and Logic2Text benchmarks show that PixT3 is competitive and, in some settings, superior to generators that operate solely on text.
In recent years, self-supervised contrastive learning has emerged as a distinguished paradigm in the artificial intelligence landscape. It facilitates unsupervised feature learning through contrastive delineations at the instance level. However, crafting an effective self-supervised paradigm remains a pivotal challenge within this field. This paper delves into two crucial factors impacting self-supervised contrastive learning-bach size and pretext tasks, and from a data processing standpoint, proposes an adaptive technique of batch fusion. The proposed method, via dimensionality reduction and reconstruction of batch data, enables formerly isolated individual data to partake in intra-batch communication through the Embedding Layer. Moreover, it adaptively amplifies the self-supervised feature encoding capability as the training progresses. We conducted a linear classification test of this method based on the classic contrastive learning framework on ImageNet-1k. The empirical findings illustrate that our approach achieves state-of-the-art performance under equitable comparisons. Benefiting from its "plug-and-play" characteristics, we further explored other contrastive learning methods. On the ImageNet-100, compared to the original performance, the top1 has seen a maximum increase of 1.25%. We suggest that the proposed method may contribute to the advancement of data-driven self-supervised learning research, bringing a fresh perspective to this community.
Airborne hyperspectral images can be used to map the land cover in large urban areas, thanks to their very high spatial and spectral resolutions on a wide spectral domain. While the spectral dimension of hyperspectral images is highly informative of the chemical composition of the land surface, the use of state-of-the-art machine learning algorithms to map the land cover has been dramatically limited by the availability of training data. To cope with the scarcity of annotations, semi-supervised and self-supervised techniques have lately raised a lot of interest in the community. Yet, the publicly available hyperspectral data sets commonly used to benchmark machine learning models are not totally suited to evaluate their generalization performances due to one or several of the following properties: a limited geographical coverage (which does not reflect the spectral diversity in metropolitan areas), a small number of land cover classes and a lack of appropriate standard train / test splits for semi-supervised and self-supervised learning. Therefore, we release in this paper the Toulouse Hyperspectral Data Set that stands out from other data sets in the above-mentioned respects in order to meet key issues in spectral representation learning and classification over large-scale hyperspectral images with very few labeled pixels. Besides, we discuss and experiment self-supervised techniques for spectral representation learning, including the Masked Autoencoder, and establish a baseline for pixel-wise classification achieving 85% overall accuracy and 77% F1 score. The Toulouse Hyperspectral Data Set and our code are publicly available at https://www.toulouse-hyperspectral-data-set.com and https://www.github.com/Romain3Ch216/tlse-experiments, respectively.
Human perception and understanding is a major domain of computer vision which, like many other vision subdomains recently, stands to gain from the use of large models pre-trained on large datasets. We hypothesize that the most common pre-training strategy of relying on general purpose, object-centric image datasets such as ImageNet, is limited by an important domain shift. On the other hand, collecting domain-specific ground truth such as 2D or 3D labels does not scale well. Therefore, we propose a pre-training approach based on self-supervised learning that works on human-centric data using only images. Our method uses pairs of images of humans: the first is partially masked and the model is trained to reconstruct the masked parts given the visible ones and a second image. It relies on both stereoscopic (cross-view) pairs, and temporal (cross-pose) pairs taken from videos, in order to learn priors about 3D as well as human motion. We pre-train a model for body-centric tasks and one for hand-centric tasks. With a generic transformer architecture, these models outperform existing self-supervised pre-training methods on a wide set of human-centric downstream tasks, and obtain state-of-the-art performance for instance when fine-tuning for model-based and model-free human mesh recovery.
A smartphone-based user mobility tracking could be effective in finding his/her location, while the unpredictable error therein due to low specification of built-in inertial measurement units (IMUs) rejects its standalone usage but demands the integration to another positioning technique like WiFi positioning. This paper aims to propose a novel integration technique using a graph neural network called Mobility-INduced Graph LEarning (MINGLE), which is designed based on two types of graphs made by capturing different user mobility features. Specifically, considering sequential measurement points (MPs) as nodes, a user's regular mobility pattern allows us to connect neighbor MPs as edges, called time-driven mobility graph (TMG). Second, a user's relatively straight transition at a constant pace when moving from one position to another can be captured by connecting the nodes on each path, called a direction-driven mobility graph (DMG). Then, we can design graph convolution network (GCN)-based cross-graph learning, where two different GCN models for TMG and DMG are jointly trained by feeding different input features created by WiFi RTTs yet sharing their weights. Besides, the loss function includes a mobility regularization term such that the differences between adjacent location estimates should be less variant due to the user's stable moving pace. Noting that the regularization term does not require ground-truth location, MINGLE can be designed under semi- and self-supervised learning frameworks. The proposed MINGLE's effectiveness is extensively verified through field experiments, showing a better positioning accuracy than benchmarks, say root mean square errors (RMSEs) being 1.398 (m) and 1.073 (m) for self- and semi-supervised learning cases, respectively.
In pathology, the rarity of certain diseases and the complexity in annotating pathological images significantly hinder the creation of extensive, high-quality datasets. This limitation impedes the progress of deep learning-assisted diagnostic systems in pathology. Consequently, it becomes imperative to devise a technology that can discern new disease categories from a minimal number of annotated examples. Such a technology would substantially advance deep learning models for rare diseases. Addressing this need, we introduce the Dual-channel Prototype Network (DCPN), rooted in the few-shot learning paradigm, to tackle the challenge of classifying pathological images with limited samples. DCPN augments the Pyramid Vision Transformer (PVT) framework for few-shot classification via self-supervised learning and integrates it with convolutional neural networks. This combination forms a dual-channel architecture that extracts multi-scale, highly precise pathological features. The approach enhances the versatility of prototype representations and elevates the efficacy of prototype networks in few-shot pathological image classification tasks. We evaluated DCPN using three publicly available pathological datasets, configuring small-sample classification tasks that mirror varying degrees of clinical scenario domain shifts. Our experimental findings robustly affirm DCPN's superiority in few-shot pathological image classification, particularly in tasks within the same domain, where it achieves the benchmarks of supervised learning.
Providing accurate and reliable predictions about the future of an epidemic is an important problem for enabling informed public health decisions. Recent works have shown that leveraging data-driven solutions that utilize advances in deep learning methods to learn from past data of an epidemic often outperform traditional mechanistic models. However, in many cases, the past data is sparse and may not sufficiently capture the underlying dynamics. While there exists a large amount of data from past epidemics, leveraging prior knowledge from time-series data of other diseases is a non-trivial challenge. Motivated by the success of pre-trained models in language and vision tasks, we tackle the problem of pre-training epidemic time-series models to learn from multiple datasets from different diseases and epidemics. We introduce Pre-trained Epidemic Time-Series Models (PEMS) that learn from diverse time-series datasets of a variety of diseases by formulating pre-training as a set of self-supervised learning (SSL) tasks. We tackle various important challenges specific to pre-training for epidemic time-series such as dealing with heterogeneous dynamics and efficiently capturing useful patterns from multiple epidemic datasets by carefully designing the SSL tasks to learn important priors about the epidemic dynamics that can be leveraged for fine-tuning to multiple downstream tasks. The resultant PEM outperforms previous state-of-the-art methods in various downstream time-series tasks across datasets of varying seasonal patterns, geography, and mechanism of contagion including the novel Covid-19 pandemic unseen in pre-trained data with better efficiency using smaller fraction of datasets.