Discrete representations of speech, obtained from Self-Supervised Learning (SSL) foundation models, are widely used, especially where there are limited data for the downstream task, such as for a low-resource language. Typically, discretization of speech into a sequence of symbols is achieved by unsupervised clustering of the latents from an SSL model. Our study evaluates whether discrete symbols - found using k-means - adequately capture tone in two example languages, Mandarin and Yoruba. We compare latent vectors with discrete symbols, obtained from HuBERT base, MandarinHuBERT, or XLS-R, for vowel and tone classification. We find that using discrete symbols leads to a substantial loss of tone information, even for language-specialised SSL models. We suggest that discretization needs to be task-aware, particularly for tone-dependent downstream tasks.
We present a self-supervised U-Net-based masked autoencoder and noise removal model designed to reconstruct original images. Once adequately trained, this model extracts high-level features, which are then combined with features from the EfficientNet B7 model. These integrated features are subsequently fed into dense layers for classification. Among the approaches of masked input and Gaussian noise removal, we selected the best U-Net reconstruction model. Additionally, we explored various configurations, including EfficientNet with attention, attention fusion of the autoencoder, and classification utilizing U-Net encoder features. The best performance was achieved with EfficientNet B7 combined with U-Net encoder features. We employed the Adam optimizer with a learning rate of 0.0001, achieving a top accuracy of 0.94 on the validation set.
Treatment selection in breast cancer is guided by molecular subtypes and clinical characteristics. However, current tools including genomic assays lack the accuracy required for optimal clinical decision-making. We developed a novel artificial intelligence (AI)-based approach that integrates digital pathology images with clinical data, providing a more robust and effective method for predicting the risk of cancer recurrence in breast cancer patients. Specifically, we utilized a vision transformer pan-cancer foundation model trained with self-supervised learning to extract features from digitized H&E-stained slides. These features were integrated with clinical data to form a multi-modal AI test predicting cancer recurrence and death. The test was developed and evaluated using data from a total of 8,161 female breast cancer patients across 15 cohorts originating from seven countries. Of these, 3,502 patients from five cohorts were used exclusively for evaluation, while the remaining patients were used for training. Our test accurately predicted our primary endpoint, disease-free interval, in the five evaluation cohorts (C-index: 0.71 [0.68-0.75], HR: 3.63 [3.02-4.37, p<0.001]). In a direct comparison (n=858), the AI test was more accurate than Oncotype DX, the standard-of-care 21-gene assay, achieving a C-index of 0.67 [0.61-0.74] versus 0.61 [0.49-0.73], respectively. Additionally, the AI test added independent prognostic information to Oncotype DX in a multivariate analysis (HR: 3.11 [1.91-5.09, p<0.001)]). The test demonstrated robust accuracy across major molecular breast cancer subtypes, including TNBC (C-index: 0.71 [0.62-0.81], HR: 3.81 [2.35-6.17, p=0.02]), where no diagnostic tools are currently recommended by clinical guidelines. These results suggest that our AI test improves upon the accuracy of existing prognostic tests, while being applicable to a wider range of patients.
This paper presents a comparative study of near-duplicate image detection techniques in a real-world use case scenario, where a document management company is commissioned to manually annotate a collection of scanned photographs. Detecting duplicate and near-duplicate photographs can reduce the time spent on manual annotation by archivists. This real use case differs from laboratory settings as the deployment dataset is available in advance, allowing the use of transductive learning. We propose a transductive learning approach that leverages state-of-the-art deep learning architectures such as convolutional neural networks (CNNs) and Vision Transformers (ViTs). Our approach involves pre-training a deep neural network on a large dataset and then fine-tuning the network on the unlabeled target collection with self-supervised learning. The results show that the proposed approach outperforms the baseline methods in the task of near-duplicate image detection in the UKBench and an in-house private dataset.
In recent advancements in unsupervised visual representation learning, the Joint-Embedding Predictive Architecture (JEPA) has emerged as a significant method for extracting visual features from unlabeled imagery through an innovative masking strategy. Despite its success, two primary limitations have been identified: the inefficacy of Exponential Moving Average (EMA) from I-JEPA in preventing entire collapse and the inadequacy of I-JEPA prediction in accurately learning the mean of patch representations. Addressing these challenges, this study introduces a novel framework, namely C-JEPA (Contrastive-JEPA), which integrates the Image-based Joint-Embedding Predictive Architecture with the Variance-Invariance-Covariance Regularization (VICReg) strategy. This integration is designed to effectively learn the variance/covariance for preventing entire collapse and ensuring invariance in the mean of augmented views, thereby overcoming the identified limitations. Through empirical and theoretical evaluations, our work demonstrates that C-JEPA significantly enhances the stability and quality of visual representation learning. When pre-trained on the ImageNet-1K dataset, C-JEPA exhibits rapid and improved convergence in both linear probing and fine-tuning performance metrics.
Atrial fibrillation is a commonly encountered clinical arrhythmia associated with stroke and increased mortality. Since professional medical knowledge is required for annotation, exploiting a large corpus of ECGs to develop accurate supervised learning-based atrial fibrillation algorithms remains challenging. Self-supervised learning (SSL) is a promising recipe for generalized ECG representation learning, eliminating the dependence on expensive labeling. However, without well-designed incorporations of knowledge related to atrial fibrillation, existing SSL approaches typically suffer from unsatisfactory capture of robust ECG representations. In this paper, we propose an inter-intra period-aware ECG representation learning approach. Considering ECGs of atrial fibrillation patients exhibit the irregularity in RR intervals and the absence of P-waves, we develop specific pre-training tasks for interperiod and intraperiod representations, aiming to learn the single-period stable morphology representation while retaining crucial interperiod features. After further fine-tuning, our approach demonstrates remarkable AUC performances on the BTCH dataset, \textit{i.e.}, 0.953/0.996 for paroxysmal/persistent atrial fibrillation detection. On commonly used benchmarks of CinC2017 and CPSC2021, the generalization capability and effectiveness of our methodology are substantiated with competitive results.
Carrying conversations in multi-sound environments is one of the more challenging tasks, since the sounds overlap across time and frequency making it difficult to understand a single sound source. One proposed approach to help isolate an attended speech source is through decoding the electroencephalogram (EEG) and identifying the attended audio source using statistical or machine learning techniques. However, the limited amount of data in comparison to other machine learning problems and the distributional shift between different EEG recordings emphasizes the need for a self supervised approach that works with limited data to achieve a more robust solution. In this paper, we propose a method based on self supervised learning to minimize the difference between the latent representations of an attended speech signal and the corresponding EEG signal. This network is further finetuned for the auditory attention classification task. We compare our results with previously published methods and achieve state-of-the-art performance on the validation set.
Accurate time series forecasting is a highly valuable endeavour with applications across many industries. Despite recent deep learning advancements, increased model complexity, and larger model sizes, many state-of-the-art models often perform worse or on par with simpler models. One of those cases is a recently proposed model, FITS, claiming competitive performance with significantly reduced parameter counts. By training a one-layer neural network in the complex frequency domain, we are able to replicate these results. Our experiments on a wide range of real-world datasets further reveal that FITS especially excels at capturing periodic and seasonal patterns, but struggles with trending, non-periodic, or random-resembling behavior. With our two novel hybrid approaches, where we attempt to remedy the weaknesses of FITS by combining it with DLinear, we achieve the best results of any known open-source model on multivariate regression and promising results in multiple/linear regression on price datasets, on top of vastly improving upon what FITS achieves as a standalone model.
The challenges faced by neural networks on tabular data are well-documented and have hampered the progress of tabular foundation models. Techniques leveraging in-context learning (ICL) have shown promise here, allowing for dynamic adaptation to unseen data. ICL can provide predictions for entirely new datasets without further training or hyperparameter tuning, therefore providing very fast inference when encountering a novel task. However, scaling ICL for tabular data remains an issue: approaches based on large language models cannot efficiently process numeric tables, and tabular-specific techniques have not been able to effectively harness the power of real data to improve performance and generalization. We are able to overcome these challenges by training tabular-specific ICL-based architectures on real data with self-supervised learning and retrieval, combining the best of both worlds. Our resulting model -- the Tabular Discriminative Pre-trained Transformer (TabDPT) -- achieves state-of-the-art performance on the CC18 (classification) and CTR23 (regression) benchmarks with no task-specific fine-tuning, demonstrating the adapatability and speed of ICL once the model is pre-trained. TabDPT also demonstrates strong scaling as both model size and amount of available data increase, pointing towards future improvements simply through the curation of larger tabular pre-training datasets and training larger models.
This paper explores the applications and challenges of graph neural networks (GNNs) in processing complex graph data brought about by the rapid development of the Internet. Given the heterogeneity and redundancy problems that graph data often have, traditional GNN methods may be overly dependent on the initial structure and attribute information of the graph, which limits their ability to accurately simulate more complex relationships and patterns in the graph. Therefore, this study proposes a graph neural network model under a self-supervised learning framework, which can flexibly combine different types of additional information of the attribute graph and its nodes, so as to better mine the deep features in the graph data. By introducing a self-supervisory mechanism, it is expected to improve the adaptability of existing models to the diversity and complexity of graph data and improve the overall performance of the model.
Self-supervised learning has become a cornerstone in various areas, particularly histopathological image analysis. Image augmentation plays a crucial role in self-supervised learning, as it generates variations in image samples. However, traditional image augmentation techniques often overlook the unique characteristics of histopathological images. In this paper, we propose a new histopathology-specific image augmentation method called stain reconstruction augmentation (SRA). We integrate our SRA with MoCo v3, a leading model in self-supervised contrastive learning, along with our additional contrastive loss terms, and call the new model SRA-MoCo v3. We demonstrate that our SRA-MoCo v3 always outperforms the standard MoCo v3 across various downstream tasks and achieves comparable or superior performance to other foundation models pre-trained on significantly larger histopathology datasets.
This paper demonstrates that spatial information can be used to learn interpretable representations in medical images using Self-Supervised Learning (SSL). Our proposed method, ISImed, is based on the observation that medical images exhibit a much lower variability among different images compared to classic data vision benchmarks. By leveraging this resemblance of human body structures across multiple images, we establish a self-supervised objective that creates a latent representation capable of capturing its location in the physical realm. More specifically, our method involves sampling image crops and creating a distance matrix that compares the learned representation vectors of all possible combinations of these crops to the true distance between them. The intuition is, that the learned latent space is a positional encoding for a given image crop. We hypothesize, that by learning these positional encodings, comprehensive image representations have to be generated. To test this hypothesis and evaluate our method, we compare our learned representation with two state-of-the-art SSL benchmarking methods on two publicly available medical imaging datasets. We show that our method can efficiently learn representations that capture the underlying structure of the data and can be used to transfer to a downstream classification task.
While image-text representation learning has become very popular in recent years, existing models tend to lack spatial awareness and have limited direct applicability for dense understanding tasks. For this reason, self-supervised image-only pretraining is still the go-to method for many dense vision applications (e.g. depth estimation, semantic segmentation), despite the lack of explicit supervisory signals. In this paper, we close this gap between image-text and self-supervised learning, by proposing a novel general-purpose image-text model, which can be effectively used off-the-shelf for dense and global vision tasks. Our method, which we refer to as Text-Image Pretraining with Spatial awareness (TIPS), leverages two simple and effective insights. First, on textual supervision: we reveal that replacing noisy web image captions by synthetically generated textual descriptions boosts dense understanding performance significantly, due to a much richer signal for learning spatially aware representations. We propose an adapted training method that combines noisy and synthetic captions, resulting in improvements across both dense and global understanding tasks. Second, on the learning technique: we propose to combine contrastive image-text learning with self-supervised masked image modeling, to encourage spatial coherence, unlocking substantial enhancements for downstream applications. Building on these two ideas, we scale our model using the transformer architecture, trained on a curated set of public images. Our experiments are conducted on 8 tasks involving 16 datasets in total, demonstrating strong off-the-shelf performance on both dense and global understanding, for several image-only and image-text tasks.
Although supervised image denoising networks have shown remarkable performance on synthesized noisy images, they often fail in practice due to the difference between real and synthesized noise. Since clean-noisy image pairs from the real world are extremely costly to gather, self-supervised learning, which utilizes noisy input itself as a target, has been studied. To prevent a self-supervised denoising model from learning identical mapping, each output pixel should not be influenced by its corresponding input pixel; This requirement is known as J-invariance. Blind-spot networks (BSNs) have been a prevalent choice to ensure J-invariance in self-supervised image denoising. However, constructing variations of BSNs by injecting additional operations such as downsampling can expose blinded information, thereby violating J-invariance. Consequently, convolutions designed specifically for BSNs have been allowed only, limiting architectural flexibility. To overcome this limitation, we propose PUCA, a novel J-invariant U-Net architecture, for self-supervised denoising. PUCA leverages patch-unshuffle/shuffle to dramatically expand receptive fields while maintaining J-invariance and dilated attention blocks (DABs) for global context incorporation. Experimental results demonstrate that PUCA achieves state-of-the-art performance, outperforming existing methods in self-supervised image denoising.
The 3D reconstruction of plants is challenging due to their complex shape causing many occlusions. Next-Best-View (NBV) methods address this by iteratively selecting new viewpoints to maximize information gain (IG). Deep-learning-based NBV (DL-NBV) methods demonstrate higher computational efficiency over classic voxel-based NBV approaches but current methods require extensive training using ground-truth plant models, making them impractical for real-world plants. These methods, moreover, rely on offline training with pre-collected data, limiting adaptability in changing agricultural environments. This paper proposes a self-supervised learning-based NBV method (SSL-NBV) that uses a deep neural network to predict the IG for candidate viewpoints. The method allows the robot to gather its own training data during task execution by comparing new 3D sensor data to the earlier gathered data and by employing weakly-supervised learning and experience replay for efficient online learning. Comprehensive evaluations were conducted in simulation and real-world environments using cross-validation. The results showed that SSL-NBV required fewer views for plant reconstruction than non-NBV methods and was over 800 times faster than a voxel-based method. SSL-NBV reduced training annotations by over 90% compared to a baseline DL-NBV. Furthermore, SSL-NBV could adapt to novel scenarios through online fine-tuning. Also using real plants, the results showed that the proposed method can learn to effectively plan new viewpoints for 3D plant reconstruction. Most importantly, SSL-NBV automated the entire network training and uses continuous online learning, allowing it to operate in changing agricultural environments.
Self-supervised learning (SSL) leverages large amounts of unlabelled data to learn rich speech representations, fostering improvements in automatic speech recognition (ASR), even when only a small amount of labelled data is available for fine-tuning. Despite the advances in SSL, a significant challenge remains when the data used for pre-training (source domain) mismatches the fine-tuning data (target domain). To tackle this domain mismatch challenge, we propose a new domain adaptation method for low-resource ASR focused on contrastive mixup for joint-embedding architectures named AC-Mix (agnostic contrastive mixup). In this approach, the SSL model is adapted through additional pre-training using mixed data views created by interpolating samples from the source and the target domains. Our proposed adaptation method consistently outperforms the baseline system, using approximately 11 hours of adaptation data and requiring only 1 hour of adaptation time on a single GPU with WavLM-Large.
Diagnosing and treating skin diseases require advanced visual skills across domains and the ability to synthesize information from multiple imaging modalities. While current deep learning models excel at specific tasks like skin cancer diagnosis from dermoscopic images, they struggle to meet the complex, multimodal requirements of clinical practice. Here, we introduce PanDerm, a multimodal dermatology foundation model pretrained through self-supervised learning on over 2 million real-world skin disease images from 11 clinical institutions across 4 imaging modalities. We evaluated PanDerm on 28 diverse benchmarks, including skin cancer screening, risk stratification, differential diagnosis of common and rare skin conditions, lesion segmentation, longitudinal monitoring, and metastasis prediction and prognosis. PanDerm achieved state-of-the-art performance across all evaluated tasks, often outperforming existing models when using only 10% of labeled data. We conducted three reader studies to assess PanDerm's potential clinical utility. PanDerm outperformed clinicians by 10.2% in early-stage melanoma detection through longitudinal analysis, improved clinicians' skin cancer diagnostic accuracy by 11% on dermoscopy images, and enhanced non-dermatologist healthcare providers' differential diagnosis by 16.5% across 128 skin conditions on clinical photographs. These results demonstrate PanDerm's potential to improve patient care across diverse clinical scenarios and serve as a model for developing multimodal foundation models in other medical specialties, potentially accelerating the integration of AI support in healthcare.
Recent advancements in speech-language models have yielded significant improvements in speech tokenization and synthesis. However, effectively mapping the complex, multidimensional attributes of speech into discrete tokens remains challenging. This process demands acoustic, semantic, and contextual information for precise speech representations. Existing speech representations generally fall into two categories: acoustic tokens from audio codecs and semantic tokens from speech self-supervised learning models. Although recent efforts have unified acoustic and semantic tokens for improved performance, they overlook the crucial role of contextual representation in comprehensive speech modeling. Our empirical investigations reveal that the absence of contextual representations results in elevated Word Error Rate (WER) and Word Information Lost (WIL) scores in speech transcriptions. To address these limitations, we propose two novel distillation approaches: (1) a language model (LM)-guided distillation method that incorporates contextual information, and (2) a combined LM and self-supervised speech model (SM)-guided distillation technique that effectively distills multimodal representations (acoustic, semantic, and contextual) into a comprehensive speech tokenizer, termed DM-Codec. The DM-Codec architecture adopts a streamlined encoder-decoder framework with a Residual Vector Quantizer (RVQ) and incorporates the LM and SM during the training process. Experiments show DM-Codec significantly outperforms state-of-the-art speech tokenization models, reducing WER by up to 13.46%, WIL by 9.82%, and improving speech quality by 5.84% and intelligibility by 1.85% on the LibriSpeech benchmark dataset. The code, samples, and model checkpoints are available at https://github.com/mubtasimahasan/DM-Codec.
Graph foundation models (GFMs) have recently gained significant attention. However, the unique data processing and evaluation setups employed by different studies hinder a deeper understanding of their progress. Additionally, current research tends to focus on specific subsets of graph learning tasks, such as structural tasks, node-level tasks, or classification tasks. As a result, they often incorporate specialized modules tailored to particular task types, losing their applicability to other graph learning tasks and contradicting the original intent of foundation models to be universal. Therefore, to enhance consistency, coverage, and diversity across domains, tasks, and research interests within the graph learning community in the evaluation of GFMs, we propose GFMBench-a systematic and comprehensive benchmark comprising 26 datasets. Moreover, we introduce LangGFM, a novel GFM that relies entirely on large language models. By revisiting and exploring the effective graph textualization principles, as well as repurposing successful techniques from graph augmentation and graph self-supervised learning within the language space, LangGFM achieves performance on par with or exceeding the state of the art across GFMBench, which can offer us new perspectives, experiences, and baselines to drive forward the evolution of GFMs.
WiFi sensing is an emerging technology that utilizes wireless signals for various sensing applications. However, the reliance on supervised learning and the scarcity of labelled data and the incomprehensible channel state information (CSI) data pose significant challenges. These issues affect deep learning models’ performance and generalization across different environments. Consequently, self-supervised learning (SSL) is emerging as a promising strategy to extract meaningful data representations with minimal reliance on labelled samples. In this paper, we introduce a novel SSL framework called Context-Aware Predictive Coding (CAPC), which effectively learns from unlabelled data and adapts to diverse environments. CAPC integrates elements of Contrastive Predictive Coding (CPC) and the augmentation-based SSL method, Barlow Twins, promoting temporal and contextual consistency in data representations. This hybrid approach captures essential temporal information in CSI, crucial for tasks like human activity recognition (HAR), and ensures robustness against data distortions. Additionally, we propose a unique augmentation, employing both uplink and downlink CSI to isolate free space propagation effects and minimize the impact of electronic distortions of the transceiver. Our evaluations demonstrate that CAPC not only outperforms other SSL methods and supervised approaches, but also achieves superior generalization capabilities. Specifically, CAPC requires fewer labelled samples while significantly outperforming supervised learning by an average margin of 30.53% and surpassing SSL baselines by 6.5% on average in low-labelled data scenarios. Furthermore, our transfer learning studies on an unseen dataset with a different HAR task and environment showcase an accuracy improvement of 1.8% over other SSL baselines and 24.7% over supervised learning, emphasizing its exceptional cross-domain adaptability. These results mark a significant breakthrough in SSL applications for WiFi sensing, highlighting CAPC’s environmental adaptability and reduced dependency on labelled data.