Due to the scarcity of labeled data, self-supervised learning (SSL) has gained much attention in 3D medical image segmentation, by extracting semantic representations from unlabeled data. Among SSL strategies, Masked image modeling (MIM) has shown effectiveness by reconstructing randomly masked images to learn detailed representations. However, conventional MIM methods require extensive training data to achieve good performance, which still poses a challenge for medical imaging. Since random masking uniformly samples all regions within medical images, it may overlook crucial anatomical regions and thus degrade the pretraining efficiency. We propose AnatoMask, a novel MIM method that leverages reconstruction loss to dynamically identify and mask out anatomically significant regions to improve pretraining efficacy. AnatoMask takes a self-distillation approach, where the model learns both how to find more significant regions to mask and how to reconstruct these masked regions. To avoid suboptimal learning, Anatomask adjusts the pretraining difficulty progressively using a masking dynamics function. We have evaluated our method on 4 public datasets with multiple imaging modalities (CT, MRI, and PET). AnatoMask demonstrates superior performance and scalability compared to existing SSL methods. The code is available at https://github.com/ricklisz/AnatoMask.
Acquiring annotations for whole slide images (WSIs)-based deep learning tasks, such as creating tissue segmentation masks or detecting mitotic figures, is a laborious process due to the extensive image size and the significant manual work involved in the annotation. This paper focuses on identifying and annotating specific image regions that optimize model training, given a limited annotation budget. While random sampling helps capture data variance by collecting annotation regions throughout the WSIs, insufficient data curation may result in an inadequate representation of minority classes. Recent studies proposed diversity sampling to select a set of regions that maximally represent unique characteristics of the WSIs. This is done by pretraining on unlabeled data through self-supervised learning and then clustering all regions in the latent space. However, establishing the optimal number of clusters can be difficult and not all clusters are task-relevant. This paper presents prototype sampling, a new method for annotation region selection. It discovers regions exhibiting typical characteristics of each task-specific class. The process entails recognizing class prototypes from extensive histopathology image-caption databases and detecting unlabeled image regions that resemble these prototypes. Our results show that prototype sampling is more effective than random and diversity sampling in identifying annotation regions with valuable training information, resulting in improved model performance in semantic segmentation and mitotic figure detection tasks. Code is available at https://github.com/DeepMicroscopy/Prototype-sampling.
The use of self-supervised learning (SSL) to train pathology foundation models has increased substantially in the past few years. Notably, several models trained on large quantities of clinical data have been made publicly available in recent months. This will significantly enhance scientific research in computational pathology and help bridge the gap between research and clinical deployment. With the increase in availability of public foundation models of different sizes, trained using different algorithms on different datasets, it becomes important to establish a benchmark to compare the performance of such models on a variety of clinically relevant tasks spanning multiple organs and diseases. In this work, we present a collection of pathology datasets comprising clinical slides associated with clinically relevant endpoints including cancer diagnoses and a variety of biomarkers generated during standard hospital operation from two medical centers. We leverage these datasets to systematically assess the performance of public pathology foundation models and provide insights into best practices for training new foundation models and selecting appropriate pretrained models.
Self-supervised learning (SSL) has revolutionized visual representation learning, but has not achieved the robustness of human vision. A reason for this could be that SSL does not leverage all the data available to humans during learning. When learning about an object, humans often purposefully turn or move around objects and research suggests that these interactions can substantially enhance their learning. Here we explore whether such object-related actions can boost SSL. For this, we extract the actions performed to change from one ego-centric view of an object to another in four video datasets. We then introduce a new loss function to learn visual and action embeddings by aligning the performed action with the representations of two images extracted from the same clip. This permits the performed actions to structure the latent visual representation. Our experiments show that our method consistently outperforms previous methods on downstream category recognition. In our analysis, we find that the observed improvement is associated with a better viewpoint-wise alignment of different objects from the same category. Overall, our work demonstrates that embodied interactions with objects can improve SSL of object categories.
Alzheimer's Dementia (AD) represents one of the most pressing challenges in the field of neurodegenerative disorders, with its progression analysis being crucial for understanding disease dynamics and developing targeted interventions. Recent advancements in deep learning and various representation learning strategies, including self-supervised learning (SSL), have shown significant promise in enhancing medical image analysis, providing innovative ways to extract meaningful patterns from complex data. Notably, the computer vision literature has demonstrated that incorporating supervisory signals into SSL can further augment model performance by guiding the learning process with additional relevant information. However, the application of such supervisory signals in the context of disease progression analysis remains largely unexplored. This gap is particularly pronounced given the inherent challenges of incorporating both event and time-to-event information into the learning paradigm. Addressing this, we propose a novel framework, Time and Even-aware SSL (TE-SSL), which integrates time-to-event and event data as supervisory signals to refine the learning process. Our comparative analysis with existing SSL-based methods in the downstream task of survival analysis shows superior performance across standard metrics.
Convolutional Neural Networks (CNNs) are known for their ability to learn hierarchical structures, naturally developing detectors for objects, and semantic concepts within their deeper layers. Activation maps (AMs) reveal these saliency regions, which are crucial for many Explainable AI (XAI) methods. However, the direct exploitation of raw AMs in CNNs for feature attribution remains underexplored in literature. This work revises Class Activation Map (CAM) methods by introducing the Label-free Activation Map (LaFAM), a streamlined approach utilizing raw AMs for feature attribution without reliance on labels. LaFAM presents an efficient alternative to conventional CAM methods, demonstrating particular effectiveness in saliency map generation for self-supervised learning while maintaining applicability in supervised learning scenarios.
Contrastive learning (CL) for Vision Transformers (ViTs) in image domains has achieved performance comparable to CL for traditional convolutional backbones. However, in 3D point cloud pretraining with ViTs, masked autoencoder (MAE) modeling remains dominant. This raises the question: Can we take the best of both worlds? To answer this question, we first empirically validate that integrating MAE-based point cloud pre-training with the standard contrastive learning paradigm, even with meticulous design, can lead to a decrease in performance. To address this limitation, we reintroduce CL into the MAE-based point cloud pre-training paradigm by leveraging the inherent contrastive properties of MAE. Specifically, rather than relying on extensive data augmentation as commonly used in the image domain, we randomly mask the input tokens twice to generate contrastive input pairs. Subsequently, a weight-sharing encoder and two identically structured decoders are utilized to perform masked token reconstruction. Additionally, we propose that for an input token masked by both masks simultaneously, the reconstructed features should be as similar as possible. This naturally establishes an explicit contrastive constraint within the generative MAE-based pre-training paradigm, resulting in our proposed method, Point-CMAE. Consequently, Point-CMAE effectively enhances the representation quality and transfer performance compared to its MAE counterpart. Experimental evaluations across various downstream applications, including classification, part segmentation, and few-shot learning, demonstrate the efficacy of our framework in surpassing state-of-the-art techniques under standard ViTs and single-modal settings. The source code and trained models are available at: https://github.com/Amazingren/Point-CMAE.
Speech encoders pretrained through self-supervised learning (SSL) have demonstrated remarkable performance in various downstream tasks, including Spoken Language Understanding (SLU) and Automatic Speech Recognition (ASR). For instance, fine-tuning SSL models for such tasks has shown significant potential, leading to improvements in the SOTA performance across challenging datasets. In contrast to existing research, this paper contributes by comparing the effectiveness of SSL approaches in the context of (i) the low-resource spoken Tunisian Arabic dialect and (ii) its combination with a low-resource SLU and ASR scenario, where only a few semantic annotations are available for fine-tuning. We conduct experiments using many SSL speech encoders on the TARIC-SLU dataset. We use speech encoders that were pre-trained on either monolingual or multilingual speech data. Some of them have also been refined without in-domain nor Tunisian data through multimodal supervised teacher-student paradigm. This study yields numerous significant findings that we are discussing in this paper.
Speech accents present a serious challenge to the performance of state-of-the-art end-to-end Automatic Speech Recognition (ASR) systems. Even with self-supervised learning and pre-training of ASR models, accent invariance is seldom achieved. In this work, we propose an accent-aware adaptation technique for self-supervised learning that introduces a trainable set of accent-specific codebooks to the self-supervised architecture. These learnable codebooks enable the model to capture accent specific information during pre-training, that is further refined during ASR finetuning. On the Mozilla Common Voice dataset, our proposed approach outperforms all other accent-adaptation approaches on both seen and unseen English accents, with up to 9% relative reduction in word error rate (WER).
In the realm of self-supervised learning (SSL), conventional wisdom has gravitated towards the utility of massive, general domain datasets for pretraining robust backbones. In this paper, we challenge this idea by exploring if it is possible to bridge the scale between general-domain datasets and (traditionally smaller) domain-specific datasets to reduce the current performance gap. More specifically, we propose Precision at Scale (PaS), a novel method for the autonomous creation of domain-specific datasets on-demand. The modularity of the PaS pipeline enables leveraging state-of-the-art foundational and generative models to create a collection of images of any given size belonging to any given domain with minimal human intervention. Extensive analysis in two complex domains, proves the superiority of PaS datasets over existing traditional domain-specific datasets in terms of diversity, scale, and effectiveness in training visual transformers and convolutional neural networks. Most notably, we prove that automatically generated domain-specific datasets lead to better pretraining than large-scale supervised datasets such as ImageNet-1k and ImageNet-21k. Concretely, models trained on domain-specific datasets constructed by PaS pipeline, beat ImageNet-1k pretrained backbones by at least 12% in all the considered domains and classification tasks and lead to better food domain performance than supervised ImageNet-21k pretrain while being 12 times smaller. Code repository: https://github.com/jesusmolrdv/Precision-at-Scale/
Self-attention performs well in long context but has quadratic complexity. Existing RNN layers have linear complexity, but their performance in long context is limited by the expressive power of their hidden state. We propose a new class of sequence modeling layers with linear complexity and an expressive hidden state. The key idea is to make the hidden state a machine learning model itself, and the update rule a step of self-supervised learning. Since the hidden state is updated by training even on test sequences, our layers are called Test-Time Training (TTT) layers. We consider two instantiations: TTT-Linear and TTT-MLP, whose hidden state is a linear model and a two-layer MLP respectively. We evaluate our instantiations at the scale of 125M to 1.3B parameters, comparing with a strong Transformer and Mamba, a modern RNN. Both TTT-Linear and TTT-MLP match or exceed the baselines. Similar to Transformer, they can keep reducing perplexity by conditioning on more tokens, while Mamba cannot after 16k context. With preliminary systems optimization, TTT-Linear is already faster than Transformer at 8k context and matches Mamba in wall-clock time. TTT-MLP still faces challenges in memory I/O, but shows larger potential in long context, pointing to a promising direction for future research.
Self-supervised learning methods for medical images primarily rely on the imaging modality during pretraining. While such approaches deliver promising results, they do not leverage associated patient or scan information collected within Electronic Health Records (EHR). Here, we propose to incorporate EHR data during self-supervised pretraining with a Masked Siamese Network (MSN) to enhance the quality of chest X-ray representations. We investigate three types of EHR data, including demographic, scan metadata, and inpatient stay information. We evaluate our approach on three publicly available chest X-ray datasets, MIMIC-CXR, CheXpert, and NIH-14, using two vision transformer (ViT) backbones, specifically ViT-Tiny and ViT-Small. In assessing the quality of the representations via linear evaluation, our proposed method demonstrates significant improvement compared to vanilla MSN and state-of-the-art self-supervised learning baselines. Our work highlights the potential of EHR-enhanced self-supervised pre-training for medical imaging. The code is publicly available at: https://github.com/nyuad-cai/CXR-EHR-MSN
This paper presents an accurate phoneme alignment model that aims for speech analysis and video content creation. We propose a variational autoencoder (VAE)-based alignment model in which a probable path is searched using encoded acoustic and linguistic embeddings in an unsupervised manner. Our proposed model is based on one TTS alignment (OTA) and extended to obtain phoneme boundaries. Specifically, we incorporate a VAE architecture to maintain consistency between the embedding and input, apply gradient annealing to avoid local optimum during training, and introduce a self-supervised learning (SSL)-based acoustic-feature input and state-level linguistic unit to utilize rich and detailed information. Experimental results show that the proposed model generated phoneme boundaries closer to annotated ones compared with the conventional OTA model, the CTC-based segmentation model, and the widely-used tool MFA.
Due to its all-weather and day-and-night capabilities, Synthetic Aperture Radar imagery is essential for various applications such as disaster management, earth monitoring, change detection and target recognition. However, the scarcity of labeled SAR data limits the performance of most deep learning algorithms. To address this issue, we propose a novel self-supervised learning framework based on masked Siamese Vision Transformers to create a General SAR Feature Extractor coined SAFE. Our method leverages contrastive learning principles to train a model on unlabeled SAR data, extracting robust and generalizable features. SAFE is applicable across multiple SAR acquisition modes and resolutions. We introduce tailored data augmentation techniques specific to SAR imagery, such as sub-aperture decomposition and despeckling. Comprehensive evaluations on various downstream tasks, including few-shot classification, segmentation, visualization, and pattern detection, demonstrate the effectiveness and versatility of the proposed approach. Our network competes with or surpasses other state-of-the-art methods in few-shot classification and segmentation tasks, even without being trained on the sensors used for the evaluation.
The scarcity of property labels remains a key challenge in materials informatics, whereas materials data without property labels are abundant in comparison. By pretraining supervised property prediction models on self-supervised tasks that depend only on the "intrinsic information" available in any Crystallographic Information File (CIF), there is potential to leverage the large amount of crystal data without property labels to improve property prediction results on small datasets. We apply Deep InfoMax as a self-supervised machine learning framework for materials informatics that explicitly maximises the mutual information between a point set (or graph) representation of a crystal and a vector representation suitable for downstream learning. This allows the pretraining of supervised models on large materials datasets without the need for property labels and without requiring the model to reconstruct the crystal from a representation vector. We investigate the benefits of Deep InfoMax pretraining implemented on the Site-Net architecture to improve the performance of downstream property prediction models with small amounts (<10^3) of data, a situation relevant to experimentally measured materials property databases. Using a property label masking methodology, where we perform self-supervised learning on larger supervised datasets and then train supervised models on a small subset of the labels, we isolate Deep InfoMax pretraining from the effects of distributional shift. We demonstrate performance improvements in the contexts of representation learning and transfer learning on the tasks of band gap and formation energy prediction. Having established the effectiveness of Deep InfoMax pretraining in a controlled environment, our findings provide a foundation for extending the approach to address practical challenges in materials informatics.
Diffusion Models are popular generative modeling methods in various vision tasks, attracting significant attention. They can be considered a unique instance of self-supervised learning methods due to their independence from label annotation. This survey explores the interplay between diffusion models and representation learning. It provides an overview of diffusion models' essential aspects, including mathematical foundations, popular denoising network architectures, and guidance methods. Various approaches related to diffusion models and representation learning are detailed. These include frameworks that leverage representations learned from pre-trained diffusion models for subsequent recognition tasks and methods that utilize advancements in representation and self-supervised learning to enhance diffusion models. This survey aims to offer a comprehensive overview of the taxonomy between diffusion models and representation learning, identifying key areas of existing concerns and potential exploration. Github link: https://github.com/dongzhuoyao/Diffusion-Representation-Learning-Survey-Taxonomy
We present a novel speaker-independent acoustic-to-articulatory inversion (AAI) model, overcoming the limitations observed in conventional AAI models that rely on acoustic features derived from restricted datasets. To address these challenges, we leverage representations from a pre-trained self-supervised learning (SSL) model to more effectively estimate the global, local, and kinematic pattern information in Electromagnetic Articulography (EMA) signals during the AAI process. We train our model using an adversarial approach and introduce an attention-based Multi-duration phoneme discriminator (MDPD) designed to fully capture the intricate relationship among multi-channel articulatory signals. Our method achieves a Pearson correlation coefficient of 0.847, marking state-of-the-art performance in speaker-independent AAI models. The implementation details and code can be found online.
The European Space Agency's Copernicus Sentinel-1 (S-1) mission is a constellation of C-band synthetic aperture radar (SAR) satellites that provide unprecedented monitoring of the world's oceans. S-1's wave mode (WV) captures 20x20 km image patches at 5 m pixel resolution and is unaffected by cloud cover or time-of-day. The mission's open data policy has made SAR data easily accessible for a range of applications, but the need for manual image annotations is a bottleneck that hinders the use of machine learning methods. This study uses nearly 10 million WV-mode images and contrastive self-supervised learning to train a semantic embedding model called WV-Net. In multiple downstream tasks, WV-Net outperforms a comparable model that was pre-trained on natural images (ImageNet) with supervised learning. Experiments show improvements for estimating wave height (0.50 vs 0.60 RMSE using linear probing), estimating near-surface air temperature (0.90 vs 0.97 RMSE), and performing multilabel-classification of geophysical and atmospheric phenomena (0.96 vs 0.95 micro-averaged AUROC). WV-Net embeddings are also superior in an unsupervised image-retrieval task and scale better in data-sparse settings. Together, these results demonstrate that WV-Net embeddings can support geophysical research by providing a convenient foundation model for a variety of data analysis and exploration tasks.
Italy exhibits rich linguistic diversity across its territory due to the distinct regional languages spoken in different areas. Recent advances in self-supervised learning provide new opportunities to analyze Italy's linguistic varieties using speech data alone. This includes the potential to leverage representations learned from large amounts of data to better examine nuances between closely related linguistic varieties. In this study, we focus on automatically identifying the geographic region of origin of speech samples drawn from Italy's diverse language varieties. We leverage self-supervised learning models to tackle this task and analyze differences and similarities between Italy's regional languages. In doing so, we also seek to uncover new insights into the relationships among these diverse yet closely related varieties, which may help linguists understand their interconnected evolution and regional development over time and space. To improve the discriminative ability of learned representations, we evaluate several supervised contrastive learning objectives, both as pre-training steps and additional fine-tuning objectives. Experimental evidence shows that pre-trained self-supervised models can effectively identify regions from speech recording. Additionally, incorporating contrastive objectives during fine-tuning improves classification accuracy and yields embeddings that distinctly separate regional varieties, demonstrating the value of combining self-supervised pre-training and contrastive learning for this task.
Knowledge-enhanced pre-trained language models (KEPLMs) leverage relation triples from knowledge graphs (KGs) and integrate these external data sources into language models via self-supervised learning. Previous works treat knowledge enhancement as two independent operations, i.e., knowledge injection and knowledge integration. In this paper, we propose to learn Knowledge-Enhanced language representations with Hierarchical Reinforcement Learning (KEHRL), which jointly addresses the problems of detecting positions for knowledge injection and integrating external knowledge into the model in order to avoid injecting inaccurate or irrelevant knowledge. Specifically, a high-level reinforcement learning (RL) agent utilizes both internal and prior knowledge to iteratively detect essential positions in texts for knowledge injection, which filters out less meaningful entities to avoid diverting the knowledge learning direction. Once the entity positions are selected, a relevant triple filtration module is triggered to perform low-level RL to dynamically refine the triples associated with polysemic entities through binary-valued actions. Experiments validate KEHRL's effectiveness in probing factual knowledge and enhancing the model's performance on various natural language understanding tasks.