Self-supervised Learning Method Research Articles

Triple Generative Self-Supervised Learning Method for Molecular Property Prediction.

Molecular property prediction is an important task in drug discovery, and with help of self-supervised learning methods, the performance of molecular property prediction could be improved by utilizing large-scale unlabeled dataset. In this paper, we propose a triple generative self-supervised learning method for molecular property prediction, called TGSS. Three encoders including a bi-directional long short-term memory recurrent neural network (BiLSTM), a Transformer, and a graph attention network (GAT) are used in pre-training the model using molecular sequence and graph structure data to extract molecular features. The variational auto encoder (VAE) is used for reconstructing features from the three models. In the downstream task, in order to balance the information between different molecular features, a feature fusion module is added to assign different weights to each feature. In addition, to improve the interpretability of the model, atomic similarity heat maps were introduced to demonstrate the effectiveness and rationality of molecular feature extraction. We demonstrate the accuracy of the proposed method on chemical and biological benchmark datasets by comparative experiments.

Boundary-aware information maximization for self-supervised medical image segmentation

Self-supervised representation learning can boost the performance of a pre-trained network on downstream tasks for which labeled data is limited. A popular method based on this paradigm, known as contrastive learning, works by constructing sets of positive and negative pairs from the data, and then pulling closer the representations of positive pairs while pushing apart those of negative pairs. Although contrastive learning has been shown to improve performance in various classification tasks, its application to image segmentation has been more limited. This stems in part from the difficulty of defining positive and negative pairs for dense feature maps without having access to pixel-wise annotations.In this work, we propose a novel self-supervised pre-training method that overcomes the challenges of contrastive learning in image segmentation. Our method leverages Information Invariant Clustering (IIC) as an unsupervised task to learn a local representation of images in the decoder of a segmentation network, but addresses three important drawbacks of this approach: (i) the difficulty of optimizing the loss based on mutual information maximization; (ii) the lack of clustering consistency for different random transformations of the same image; (iii) the poor correspondence of clusters obtained by IIC with region boundaries in the image. Toward this goal, we first introduce a regularized mutual information maximization objective that encourages the learned clusters to be balanced and consistent across different image transformations. We also propose a boundary-aware loss based on cross-correlation, which helps the learned clusters to be more representative of important regions in the image. Compared to contrastive learning applied in dense features, our method does not require computing positive and negative pairs and also enhances interpretability through the visualization of learned clusters.Comprehensive experiments involving four different medical image segmentation tasks reveal the high effectiveness of our self-supervised representation learning method. Our results show the proposed method to outperform by a large margin several state-of-the-art self-supervised and semi-supervised approaches for segmentation, reaching a performance close to full supervision with only a few labeled examples.

Self-supervised dual-domain balanced dropblock-network for low-dose CT denoising

Objective. Self-supervised learning methods have been successfully applied for low-dose computed tomography (LDCT) denoising, with the advantage of not requiring labeled data. Conventional self-supervised methods operate only in the image domain, ignoring valuable priors in the sinogram domain. Recently proposed dual-domain methods address this limitation but encounter issues with blurring artifacts in the reconstructed image due to the inhomogeneous distribution of noise levels in low-dose sinograms. Approach. To tackle this challenge, this paper proposes SDBDNet, an end-to-end dual-domain self-supervised method for LDCT denoising. With the network designed based on the properties of inhomogeneous noise in low-dose sinograms and the principle of moderate sinogram-domain denoising, SDBDNet achieves effective denoising in dual domains without introducing blurring artifacts. Specifically, we split the sinogram into two subsets based on the positions of detector cells to generate paired training data with high similarity and independent noise. These sub-sinograms are then restored to their original size using 1D interpolation and learning-based correction. To achieve adaptive and moderate smoothing in the sinogram domain, we integrate Dropblock, a type of convolution layer with regularization, into SDBDNet, and set a weighted average between the denoised sinograms and their noisy counterparts, leading to a well-balanced dual-domain approach. Main results. Numerical experiments show that our method outperforms popular non-learning and self-supervised learning methods, demonstrating its effectiveness and superior performance. Significance. While introducing a novel high-performance dual-domain self-supervised LDCT denoising method, this paper also emphasizes and verifies the importance of appropriate sinogram-domain denoising in dual-domain methods, which might inspire future work.

Detecting abnormal cell behaviors from dry mass time series

The prediction of pathological changes on single cell behaviour is a challenging task for deep learning models. Indeed, in self-supervised learning methods, no prior labels are used for the training and all of the information for event predictions are extracted from the data themselves. We present here a novel self-supervised learning model for the detection of anomalies in a given cell population, StArDusTS. Cells are monitored over time, and analysed to extract time-series of dry mass values. We assessed its performances on different cell lines, showing a precision of 96% in the automatic detection of anomalies. Additionally, anomaly detection was also associated with cell measurement errors inherent to the acquisition or analysis pipelines, leading to an improvement of the upstream methods for feature extraction. Our results pave the way to novel architectures for the continuous monitoring of cell cultures in applied research or bioproduction applications, and for the prediction of pathological cellular changes.

Self-Supervised Framework Based on Subject-Wise Clustering for Human Subject Time Series Data

With the widespread adoption of IoT, wearable devices, and sensors, time series data from human subjects are significantly increasing in the healthcare domain. Due to the laborious nature of manual annotation in time series data and the requirement for human experts, self-supervised learning methods are attempted to alleviate the limited label situations. While existing self-supervised methods have been successful to achieve comparable performance to the fully supervised methods, there are still some limitations that need to be addressed, considering the nature of time series data from human subjects: In real-world clinical settings, data labels (e.g., sleep stages) are usually annotated by subject-level, and there is a substantial variation in patterns between subjects. Thus, a model should be designed to deal with not only the label scarcity but also subject-wise nature of data to ensure high performance in real-world scenarios. To mitigate these issues, we propose a novel self-supervised learning framework for human subject time series data: Subject-Aware Time Series Clustering (SA-TSC). In the unsupervised representation learning phase, SA-TSC adopts a subject-wise learning strategy rather than instance-wise learning which randomly samples data instances from different subjects within the batch during training. Specifically, we generate subject-graphs with our graph construction method based on Gumbel-Softmax and perform graph spectral clustering on each subject-graph. In addition, we utilize graph neural networks to capture dependencies between channels and design our own graph learning module motivated from self-supervised loss. Experimental results show the outstanding performance of our SA-TSC with the limited & subject-wise label setting, leading to its high applicability to the healthcare industry. The code is available at: https://github.com/DILAB-HYU/SA-TSC

Transformer-based self-supervised image super-resolution method for Rotating Synthetic Aperture system via multi-temporal fusion

Rotating Synthetic Aperture (RSA) technology is one of the distinctly advantageous Earth geostationary orbit optical remote sensing technologies. However, the continuous rotation of the RSA system’s rectangular primary mirror results in a discernible drop in resolution along the shorter side of the mirror. Additionally, the captured images exhibit periodic and time-varying characteristics. To improve the image quality to meet interpretation needs, we first delineate the imaging process of the rotating primary mirror and analyze the characteristics of image degradation based on the system’s imaging mechanism. Then, we propose a dual super-resolution (SR) framework based on Swin Transformer and introduce a self-supervised learning method for jointly training the unified SR network using wavelet fusion. The self-supervised learning method effectively utilizes the spatiotemporal correlation of the information contained in images captured at different rotation directions of the rectangular pupil. Moreover, the attention mechanism in Transformer can adopt a global perspective and utilize content-based interactions between image content and attention weights to model strong long-range dependencies in remote sensing images. This approach significantly enhances image quality along the pupil’s shorter side, consequently yielding superior results. Extensive digital and semi-physical imaging experiments, involving six aspect ratios of the primary mirror, demonstrate that our SR method surpasses state-of-the-art methods. The work in this paper can serve as a valuable reference for future space applications of the RSA technology.

MM-Point: Multi-View Information-Enhanced Multi-Modal Self-Supervised 3D Point Cloud Understanding

In perception, multiple sensory information is integrated to map visual information from 2D views onto 3D objects, which is beneficial for understanding in 3D environments. But in terms of a single 2D view rendered from different angles, only limited partial information can be provided. The richness and value of Multi-view 2D information can provide superior self-supervised signals for 3D objects. In this paper, we propose a novel self-supervised point cloud representation learning method, MM-Point, which is driven by intra-modal and inter-modal similarity objectives. The core of MM-Point lies in the Multi-modal interaction and transmission between 3D objects and multiple 2D views at the same time. In order to more effectively simultaneously perform the consistent cross-modal objective of 2D multi-view information based on contrastive learning, we further propose Multi-MLP and Multi-level Augmentation strategies. Through carefully designed transformation strategies, we further learn Multi-level invariance in 2D Multi-views. MM-Point demonstrates state-of-the-art (SOTA) performance in various downstream tasks. For instance, it achieves a peak accuracy of 92.4% on the synthetic dataset ModelNet40, and a top accuracy of 87.8% on the real-world dataset ScanObjectNN, comparable to fully supervised methods. Additionally, we demonstrate its effectiveness in tasks such as few-shot classification, 3D part segmentation and 3D semantic segmentation.

SwitchTab: Switched Autoencoders Are Effective Tabular Learners

Self-supervised representation learning methods have achieved significant success in computer vision and natural language processing (NLP), where data samples exhibit explicit spatial or semantic dependencies. However, applying these methods to tabular data is challenging due to the less pronounced dependencies among data samples. In this paper, we address this limitation by introducing SwitchTab, a novel self-supervised method specifically designed to capture latent dependencies in tabular data. SwitchTab leverages an asymmetric encoder-decoder framework to decouple mutual and salient features among data pairs, resulting in more representative embeddings. These embeddings, in turn, contribute to better decision boundaries and lead to improved results in downstream tasks. To validate the effectiveness of SwitchTab, we conduct extensive experiments across various domains involving tabular data. The results showcase superior performance in end-to-end prediction tasks with fine-tuning. Moreover, we demonstrate that pre-trained salient embeddings can be utilized as plug-and-play features to enhance the performance of various traditional classification methods (e.g., Logistic Regression, XGBoost, etc.). Lastly, we highlight the capability of SwitchTab to create explainable representations through visualization of decoupled mutual and salient features in the latent space.

A review of self-supervised, generative, and few-shot deep learning methods for data-limited magnetic resonance imaging segmentation.

Magnetic resonance imaging (MRI) is a ubiquitous medical imaging technology with applications in disease diagnostics, intervention, and treatment planning. Accurate MRI segmentation is critical for diagnosing abnormalities, monitoring diseases, and deciding on a course of treatment. With the advent of advanced deep learning frameworks, fully automated and accurate MRI segmentation is advancing. Traditional supervised deep learning techniques have advanced tremendously, reaching clinical-level accuracy in the field of segmentation. However, these algorithms still require a large amount of annotated data, which is oftentimes unavailable or impractical. One way to circumvent this issue is to utilize algorithms that exploit a limited amount of labeled data. This paper aims to review such state-of-the-art algorithms that use a limited number of annotated samples. We explain the fundamental principles of self-supervised learning, generative models, few-shot learning, and semi-supervised learning and summarize their applications in cardiac, abdomen, and brain MRI segmentation. Throughout this review, we highlight algorithms that can be employed based on the quantity of annotated data available. We also present a comprehensive list of notable publicly available MRI segmentation datasets. To conclude, we discuss possible future directions of the field-including emerging algorithms, such as contrastive language-image pretraining, and potential combinations across the methods discussed-that can further increase the efficacy of image segmentation with limited labels.

Self-Supervised Representation Learning with Meta Comprehensive Regularization

Self-Supervised Learning (SSL) methods harness the concept of semantic invariance by utilizing data augmentation strategies to produce similar representations for different deformations of the same input. Essentially, the model captures the shared information among multiple augmented views of samples, while disregarding the non-shared information that may be beneficial for downstream tasks. To address this issue, we introduce a module called CompMod with Meta Comprehensive Regularization (MCR), embedded into existing self-supervised frameworks, to make the learned representations more comprehensive. Specifically, we update our proposed model through a bi-level optimization mechanism, enabling it to capture comprehensive features. Additionally, guided by the constrained extraction of features using maximum entropy coding, the self-supervised learning model learns more comprehensive features on top of learning consistent features. In addition, we provide theoretical support for our proposed method from information theory and causal counterfactual perspective. Experimental results show that our method achieves significant improvement in classification, object detection and semantic segmentation tasks on multiple benchmark datasets.

Self-Supervised Disentangled Representation Learning for Robust Target Speech Extraction

Speech signals are inherently complex as they encompass both global acoustic characteristics and local semantic information. However, in the task of target speech extraction, certain elements of global and local semantic information in the reference speech, which are irrelevant to speaker identity, can lead to speaker confusion within the speech extraction network. To overcome this challenge, we propose a self-supervised disentangled representation learning method. Our approach tackles this issue through a two-phase process, utilizing a reference speech encoding network and a global information disentanglement network to gradually disentangle the speaker identity information from other irrelevant factors. We exclusively employ the disentangled speaker identity information to guide the speech extraction network. Moreover, we introduce the adaptive modulation Transformer to ensure that the acoustic representation of the mixed signal remains undisturbed by the speaker embeddings. This component incorporates speaker embeddings as conditional information, facilitating natural and efficient guidance for the speech extraction network. Experimental results substantiate the effectiveness of our meticulously crafted approach, showcasing a substantial reduction in the likelihood of speaker confusion.

Every Node Is Different: Dynamically Fusing Self-Supervised Tasks for Attributed Graph Clustering

Attributed graph clustering is an unsupervised task that partitions nodes into different groups. Self-supervised learning (SSL) shows great potential in handling this task, and some recent studies simultaneously learn multiple SSL tasks to further boost performance. Currently, different SSL tasks are assigned the same set of weights for all graph nodes. However, we observe that some graph nodes whose neighbors are in different groups require significantly different emphases on SSL tasks. In this paper, we propose to dynamically learn the weights of SSL tasks for different nodes and fuse the embeddings learned from different SSL tasks to boost performance. We design an innovative graph clustering approach, namely Dynamically Fusing Self-Supervised Learning (DyFSS). Specifically, DyFSS fuses features extracted from diverse SSL tasks using distinct weights derived from a gating network. To effectively learn the gating network, we design a dual-level self-supervised strategy that incorporates pseudo labels and the graph structure. Extensive experiments on five datasets show that DyFSS outperforms the state-of-the-art multi-task SSL methods by up to 8.66% on the accuracy metric. The code of DyFSS is available at: https://github.com/q086/DyFSS.

Sterling: Synergistic Representation Learning on Bipartite Graphs

A fundamental challenge of bipartite graph representation learning is how to extract informative node embeddings. Self-Supervised Learning (SSL) is a promising paradigm to address this challenge. Most recent bipartite graph SSL methods are based on contrastive learning which learns embeddings by discriminating positive and negative node pairs. Contrastive learning usually requires a large number of negative node pairs, which could lead to computational burden and semantic errors. In this paper, we introduce a novel synergistic representation learning model (STERLING) to learn node embeddings without negative node pairs. STERLING preserves the unique local and global synergies in bipartite graphs. The local synergies are captured by maximizing the similarity of the inter-type and intra-type positive node pairs, and the global synergies are captured by maximizing the mutual information of co-clusters. Theoretical analysis demonstrates that STERLING could improve the connectivity between different node types in the embedding space. Extensive empirical evaluation on various benchmark datasets and tasks demonstrates the effectiveness of STERLING for extracting node embeddings.

SegIns: A simple extension to instance discrimination task for better localization learning

Recent self-supervised learning methods, where instance discrimination task is a fundamental way of pretraining convolutional neural networks (CNN), excel in transfer learning performance. Even though instance discrimination task is a well suited pretraining method for classification with its image-level learning, lack of dense representation learning makes it sub-optimal for localization tasks such as object detection. In this paper, we aim to mitigate this shortcoming of instance discrimination task by extending it to jointly learn dense representations alongside image-level representations. We add a segmentation branch parallel to the image-level learning to predict class-agnostic masks, enhancing location-awareness of the representations. We show the effectiveness of our pretraining approach on localization tasks by transferring the learned representations to object detection and segmentation tasks, providing relative improvements by up to 1.7% AP on PASCAL VOC and 0.8% AP on COCO object detection, 0.8% AP on COCO instance segmentation and 3.6% mIoU on PASCAL VOC semantic segmentation respectively.

Generalizing Parkinson's disease detection using keystroke dynamics: a self-supervised approach.

Passive monitoring of touchscreen interactions generates keystroke dynamic signals that can be used to detect and track neurological conditions such as Parkinson's disease (PD) and psychomotor impairment with minimal burden on the user. However, this typically requires datasets with clinically confirmed labels collected in standardized environments, which is challenging, especially for a large subject pool. This study validates the efficacy of a self-supervised learning method in reducing the reliance on labels and evaluates its generalizability. We propose a new type of self-supervised loss combining Barlow Twins loss, which attempts to create similar feature representations with reduced feature redundancy for samples coming from the same subject, and a Dissimilarity loss, which promotes uncorrelated features for samples generated by different subjects. An encoder is first pre-trained using this loss on unlabeled data from an uncontrolled setting, then fine-tuned with clinically validated data. Our experiments test the model generalizability with controls and subjects with PD on 2 independent datasets. Our approach showed better generalization compared to previous methods, including a feature engineering strategy, a deep learning model pre-trained on Parkinsonian signs, and a traditional supervised model. The absence of standardized data acquisition protocols and the limited availability of annotated datasets compromise the generalizability of supervised models. In these contexts, self-supervised models offer the advantage of learning more robust patterns from the data, bypassing the need for ground truth labels. This approach has the potential to accelerate the clinical validation of touchscreen typing software for neurodegenerative diseases.

Cellular data extraction from multiplexed brain imaging data using self-supervised Dual-loss Adaptive Masked Autoencoder

Reliable large-scale cell detection and segmentation is the fundamental first step to understanding biological processes in the brain. The ability to phenotype cells at scale can accelerate preclinical drug evaluation and system-level brain histology studies. The impressive advances in deep learning offer a practical solution to cell image detection and segmentation. Unfortunately, categorizing cells and delineating their boundaries for training deep networks is an expensive process that requires skilled biologists. This paper presents a novel self-supervised Dual-Loss Adaptive Masked Autoencoder (DAMA) for learning rich features from multiplexed immunofluorescence brain images. DAMA’s objective function minimizes the conditional entropy in pixel-level reconstruction and feature-level regression. Unlike existing self-supervised learning methods based on a random image masking strategy, DAMA employs a novel adaptive mask sampling strategy to maximize mutual information and effectively learn brain cell data. To the best of our knowledge, this is the first effort to develop a self-supervised learning method for multiplexed immunofluorescence brain images. Our extensive experiments demonstrate that DAMA features enable superior cell detection, segmentation, and classification performance without requiring many annotations. In addition, to examine the generalizability of DAMA, we also experimented on TissueNet, a multiplexed imaging dataset comprised of two-channel fluorescence images from six distinct tissue types, captured using six different imaging platforms. Our code is publicly available at https://github.com/hula-ai/DAMA.

WISE: whole-scenario embryo identification using self-supervised learning encoder in IVF.

To study the effectiveness of whole-scenario embryo identification using a self-supervised learning encoder (WISE) in in vitro fertilization (IVF) on time-lapse, cross-device, and cryo-thawed scenarios. WISE was based on the vision transformer (ViT) architecture and masked autoencoders (MAE), a self-supervised learning (SSL) method. To train WISE, we prepared three datasets including the SSL pre-training dataset, the time-lapse identification dataset, and the cross-device identification dataset. To identify whether pairs of images were from the same embryos in different scenarios in the downstream identification tasks, embryo images including time-lapse and microscope images were first pre-processed through object detection, cropping, padding, and resizing, and then fed into WISE to get predictions. WISE could accurately identify embryos in the three scenarios. The accuracy was 99.89% on the time-lapse identification dataset, and 83.55% on the cross-device identification dataset. Besides, we subdivided a cryo-thawed evaluation set from the cross-device test set to have a better estimation of how WISE performs in the real-world, and it reached an accuracy of 82.22%. There were approximately 10% improvements in cross-device and cryo-thawed identification tasks after the SSL method was applied. Besides, WISE demonstrated improvements in the accuracy of 9.5%, 12%, and 18% over embryologists in the three scenarios. SSL methods can improve embryo identification accuracy even when dealing with cross-device and cryo-thawed paired images. The study is the first to apply SSL in embryo identification, and the results show the promise of WISE for future application in embryo witnessing.

Super resolution-based methodology for self-supervised segmentation of microscopy images.

Data-driven Artificial Intelligence (AI)/Machine learning (ML) image analysis approaches have gained a lot of momentum in analyzing microscopy images in bioengineering, biotechnology, and medicine. The success of these approaches crucially relies on the availability of high-quality microscopy images, which is often a challenge due to the diverse experimental conditions and modes under which these images are obtained. In this study, we propose the use of recent ML-based image super-resolution (SR) techniques for improving the image quality of microscopy images, incorporating them into multiple ML-based image analysis tasks, and describing a comprehensive study, investigating the impact of SR techniques on the segmentation of microscopy images. The impacts of four Generative Adversarial Network (GAN)- and transformer-based SR techniques on microscopy image quality are measured using three well-established quality metrics. These SR techniques are incorporated into multiple deep network pipelines using supervised, contrastive, and non-contrastive self-supervised methods to semantically segment microscopy images from multiple datasets. Our results show that the image quality of microscopy images has a direct influence on the ML model performance and that both supervised and self-supervised network pipelines using SR images perform better by 2%-6% in comparison to baselines, not using SR. Based on our experiments, we also establish that the image quality improvement threshold range [20-64] for the complemented Perception-based Image Quality Evaluator(PIQE) metric can be used as a pre-condition by domain experts to incorporate SR techniques to significantly improve segmentation performance. A plug-and-play software platform developed to integrate SR techniques with various deep networks using supervised and self-supervised learning methods is also presented.

Physics-Guided Dual Self-Supervised Learning for Structure-Based Material Property Prediction.

Deep learning models have been widely used for high-performance material property prediction. However, training such models usually requires a large amount of labeled data, which are usually unavailable. Self-supervised learning (SSL) methods have been proposed to address this data scarcity issue. Herein, we present DSSL, a physics-guided dual SSL framework, for graph neural network-based material property prediction, which combines node masking-based generative SSL with atomic coordinate perturbation-based contrastive SSL strategies to capture local and global information about input crystals. Moreover, we achieve physics-guided pretraining by using the macroproperty (e.g., elasticity)-related microproperty prediction of atomic stiffness as an additional pretext task. We pretrain our DSSL model on the Materials Project database and fine-tune it with 10 material property data sets. The experimental results demonstrate that teaching neural networks some physics using the SSL strategy can afford ≤26.89% performance improvement compared to that of the baseline models.

A self-supervised learning framework based on masked autoencoder for complex wafer bin map classification

Wafer bin map (WBM) automatic classification is one of the critical challenges for semiconductor intelligent manufacturing. Many deep learning-based classification models have performed well in WBM classification, but all require a large amount of labeled data for training. Since real-world WBMs are highly complex and can be labeled correctly only by seasoned engineers, such requirements undermine the practical value of those methods. Several self-supervised learning methods have recently been proposed for WBM to improve classification performance. However, they still require much labeled data for fine-tuning and are only adapted for binary WBM with a single gross failure area. To address these limitations, this study introduces a self-supervised framework based on masked autoencoder (MAE) for complex WBMs with mixed bin signatures and multiple gross failure area patterns. A patchMC encoder is proposed to improve MAE’s representation ability for complex WBMs with mixed bin signatures. Moreover, the pre-trained MAE encoder with a multi-label classifier fine-tuned by labeled WBMs enables a few-shot classification of complex WBMs with multiple gross failure areas. Experimental validation of the proposed method is performed on a real-world complex WBM dataset from Intel Corporation. The results demonstrate that the proposed method can make good use of unlabeled WBMs and reduce the demand for labeled data to a few-shot level and, at the same time, guarantees a classification accuracy of more than 90%. By comparing MAE with other self-supervised learning methods, MAE outperforms other existing self-supervised methods for WBM data.