Graph Contrastive Learning (GCL) has recently emerged as a promising graph self-supervised learning framework for learning discriminative node representations without labels. The widely adopted objective function of GCL benefits from two key properties: \emph{alignment} and \emph{uniformity}, which align representations of positive node pairs while uniformly distributing all representations on the hypersphere. The uniformity property plays a critical role in preventing representation collapse and is achieved by pushing apart augmented views of different nodes (negative pairs). As such, existing GCL methods inherently rely on increasing the quantity and quality of negative samples, resulting in heavy computational demands, memory overhead, and potential class collision issues. In this study, we propose a negative-free objective to achieve uniformity, inspired by the fact that points distributed according to a normalized isotropic Gaussian are uniformly spread across the unit hypersphere. Therefore, we can minimize the distance between the distribution of learned representations and the isotropic Gaussian distribution to promote the uniformity of node representations. Our method also distinguishes itself from other approaches by eliminating the need for a parameterized mutual information estimator, an additional projector, asymmetric structures, and, crucially, negative samples. Extensive experiments over seven graph benchmarks demonstrate that our proposal achieves competitive performance with fewer parameters, shorter training times, and lower memory consumption compared to existing GCL methods.
Data augmentations play an important role in the recent success of Self-Supervised Learning (SSL). While commonly viewed as encoding invariances into the learned representations, this interpretation overlooks the impact of the pretraining architecture and suggests that SSL would require diverse augmentations which resemble the data to work well. However, these assumptions do not align with empirical evidence, encouraging further theoretical understanding to guide the principled design of augmentations in new domains. To this end, we use kernel theory to derive analytical expressions for data augmentations that achieve desired target representations after pretraining. We consider two popular non-contrastive losses, VICReg and Barlow Twins, and provide an algorithm to construct such augmentations. Our analysis shows that augmentations need not be similar to the data to learn useful representations, nor be diverse, and that the architecture has a significant impact on the optimal augmentations.
Prediction is a fundamental capability of all living organisms, and has been proposed as an objective for learning sensory representations. Recent work demonstrates that in primate visual systems, prediction is facilitated by neural representations that follow straighter temporal trajectories than their initial photoreceptor encoding, which allows for prediction by linear extrapolation. Inspired by these experimental findings, we develop a self-supervised learning (SSL) objective that explicitly quantifies and promotes straightening. We demonstrate the power of this objective in training deep feedforward neural networks on smoothly-rendered synthetic image sequences that mimic commonly-occurring properties of natural videos. The learned model contains neural embeddings that are predictive, but also factorize the geometric, photometric, and semantic attributes of objects. The representations also prove more robust to noise and adversarial attacks compared to previous SSL methods that optimize for invariance to random augmentations. Moreover, these beneficial properties can be transferred to other training procedures by using the straightening objective as a regularizer, suggesting a broader utility for straightening as a principle for robust unsupervised learning.
Self-supervised learning (SSL) has rapidly advanced in recent years, approaching the performance of its supervised counterparts through the extraction of representations from unlabeled data. However, dimensional collapse, where a few large eigenvalues dominate the eigenspace, poses a significant obstacle for SSL. When dimensional collapse occurs on features (e.g. hidden features and representations), it prevents features from representing the full information of the data; when dimensional collapse occurs on weight matrices, their filters are self-related and redundant, limiting their expressive power. Existing studies have predominantly concentrated on the dimensional collapse of representations, neglecting whether this can sufficiently prevent the dimensional collapse of the weight matrices and hidden features. To this end, we first time propose a mitigation approach employing orthogonal regularization (OR) across the encoder, targeting both convolutional and linear layers during pretraining. OR promotes orthogonality within weight matrices, thus safeguarding against the dimensional collapse of weight matrices, hidden features, and representations. Our empirical investigations demonstrate that OR significantly enhances the performance of SSL methods across diverse benchmarks, yielding consistent gains with both CNNs and Transformer-based architectures.
Due to significant variations in the projection of the same object from different viewpoints, machine learning algorithms struggle to recognize the same object across various perspectives. In contrast, toddlers quickly learn to recognize objects from different viewpoints with almost no supervision. Recent works argue that toddlers develop this ability by mapping close-in-time visual inputs to similar representations while interacting with objects. High acuity vision is only available in the central visual field, which may explain why toddlers (much like adults) constantly move their gaze around during such interactions. It is unclear whether/how much toddlers curate their visual experience through these eye movements to support learning object representations. In this work, we explore whether a bio inspired visual learning model can harness toddlers' gaze behavior during a play session to develop view-invariant object recognition. Exploiting head-mounted eye tracking during dyadic play, we simulate toddlers' central visual field experience by cropping image regions centered on the gaze location. This visual stream feeds a time-based self-supervised learning algorithm. Our experiments demonstrate that toddlers' gaze strategy supports the learning of invariant object representations. Our analysis also reveals that the limited size of the central visual field where acuity is high is crucial for this. We further find that toddlers' visual experience elicits more robust representations compared to adults' mostly because toddlers look at objects they hold themselves for longer bouts. Overall, our work reveals how toddlers' gaze behavior supports self-supervised learning of view-invariant object recognition.
Group Recommendation (GR), which aims to recommend items to groups of users, has become a promising and practical direction for recommendation systems. This paper points out two issues of the state-of-the-art GR models. (1) The pre-defined and fixed number of user groups is inadequate for real-time industrial recommendation systems, where the group distribution can shift dynamically. (2) The training schema of existing GR methods is supervised, necessitating expensive user-group and group-item labels, leading to significant annotation costs. To this end, we present a novel unsupervised group recommendation framework named \underline{I}dentify \underline{T}hen \underline{R}ecommend (\underline{ITR}), where it first identifies the user groups in an unsupervised manner even without the pre-defined number of groups, and then two pre-text tasks are designed to conduct self-supervised group recommendation. Concretely, at the group identification stage, we first estimate the adaptive density of each user point, where areas with higher densities are more likely to be recognized as group centers. Then, a heuristic merge-and-split strategy is designed to discover the user groups and decision boundaries. Subsequently, at the self-supervised learning stage, the pull-and-repulsion pre-text task is proposed to optimize the user-group distribution. Besides, the pseudo group recommendation pre-text task is designed to assist the recommendations. Extensive experiments demonstrate the superiority and effectiveness of ITR on both user recommendation (e.g., 22.22\% NDCG@5 $\uparrow$) and group recommendation (e.g., 22.95\% NDCG@5 $\uparrow$). Furthermore, we deploy ITR on the industrial recommender and achieve promising results.
Self-supervised learning (SSL) models have become crucial in speech processing, with recent advancements concentrating on developing architectures that capture representations across multiple timescales. The primary goal of these multi-scale architectures is to exploit the hierarchical nature of speech, where lower-resolution components aim to capture representations that align with increasingly abstract concepts (e.g., from phones to words to sentences). Although multi-scale approaches have demonstrated some improvements over single-scale models, the precise reasons for these enhancements have poor empirical support. In this study, we present an initial analysis of layer-wise representations in multi-scale architectures, with a focus on Canonical Correlation Analysis (CCA) and Mutual Information (MI). We apply this analysis to Multi-Resolution HuBERT (MR-HuBERT) and find that (1) the improved performance on SUPERB tasks is primarily due to the auxiliary low-resolution loss rather than the downsampling itself, and (2) downsampling to lower resolutions neither improves downstream performance nor correlates with higher-level information (e.g., words), though it does improve computational efficiency. These findings challenge assumptions about the multi-scale nature of MR-HuBERT and motivate the importance of disentangling computational efficiency from learning better representations.
Foundation models have recently gained significant attention because of their generalizability and adaptability across multiple tasks and data distributions. Although medical foundation models have emerged, solutions for cardiac imaging, especially echocardiography videos, are still unexplored. In this paper, we introduce EchoFM, a foundation model specifically designed to represent and analyze echocardiography videos. In EchoFM, we propose a self-supervised learning framework that captures both spatial and temporal variability patterns through a spatio-temporal consistent masking strategy and periodic-driven contrastive learning. This framework can effectively capture the spatio-temporal dynamics of echocardiography and learn the representative video features without any labels. We pre-train our model on an extensive dataset comprising over 290,000 echocardiography videos covering 26 scan views across different imaging modes, with up to 20 million frames of images. The pre-trained EchoFM can then be easily adapted and fine-tuned for a variety of downstream tasks, serving as a robust backbone model. Our evaluation was systemically designed for four downstream tasks after the echocardiography examination routine. Experiment results show that EchoFM surpasses state-of-the-art methods, including specialized echocardiography methods, self-supervised pre-training models, and general-purposed pre-trained foundation models, across all downstream tasks.
Multimodal representation learning seeks to relate and decompose information inherent in multiple modalities. By disentangling modality-specific information from information that is shared across modalities, we can improve interpretability and robustness and enable downstream tasks such as the generation of counterfactual outcomes. Separating the two types of information is challenging since they are often deeply entangled in many real-world applications. We propose Disentangled Self-Supervised Learning (DisentangledSSL), a novel self-supervised approach for learning disentangled representations. We present a comprehensive analysis of the optimality of each disentangled representation, particularly focusing on the scenario not covered in prior work where the so-called Minimum Necessary Information (MNI) point is not attainable. We demonstrate that DisentangledSSL successfully learns shared and modality-specific features on multiple synthetic and real-world datasets and consistently outperforms baselines on various downstream tasks, including prediction tasks for vision-language data, as well as molecule-phenotype retrieval tasks for biological data. The code is available at https://github.com/uhlerlab/DisentangledSSL.
Our work tackles the computational challenges of contrastive learning methods, particularly for the pretraining of Vision Transformers (ViTs). Despite the effectiveness of contrastive learning, the substantial computational resources required for training often hinder their practical application. To mitigate this issue, we propose an acceleration framework, leveraging ViT's unique ability to generalize across inputs of varying sequence lengths. Our method employs a mix of sequence compression strategies, including randomized token dropout and flexible patch scaling, to reduce the cost of gradient estimation and accelerate convergence. We further provide an in-depth analysis of the gradient estimation error of various acceleration strategies as well as their impact on downstream tasks, offering valuable insights into the trade-offs between acceleration and performance. We also propose a novel procedure to identify an optimal acceleration schedule to adjust the sequence compression ratios to the training progress, ensuring efficient training without sacrificing downstream performance. Our approach significantly reduces computational overhead across various self-supervised learning algorithms on large-scale datasets. In ImageNet, our method achieves speedups of 4$\times$ in MoCo, 3.3$\times$ in SimCLR, and 2.5$\times$ in DINO, demonstrating substantial efficiency gains.
This paper describes sound event localization and detection (SELD) for spatial audio recordings captured by firstorder ambisonics (FOA) microphones. In this task, one may train a deep neural network (DNN) using FOA data annotated with the classes and directions of arrival (DOAs) of sound events. However, the performance of this approach is severely bounded by the amount of annotated data. To overcome this limitation, we propose a novel method of pretraining the feature extraction part of the DNN in a self-supervised manner. We use spatial audio-visual recordings abundantly available as virtual reality contents. Assuming that sound objects are concurrently observed by the FOA microphones and the omni-directional camera, we jointly train audio and visual encoders with contrastive learning such that the audio and visual embeddings of the same recording and DOA are made close. A key feature of our method is that the DOA-wise audio embeddings are jointly extracted from the raw audio data, while the DOA-wise visual embeddings are separately extracted from the local visual crops centered on the corresponding DOA. This encourages the latent features of the audio encoder to represent both the classes and DOAs of sound events. The experiment using the DCASE2022 Task 3 dataset of 20 hours shows non-annotated audio-visual recordings of 100 hours reduced the error score of SELD from 36.4 pts to 34.9 pts.
While large models trained with self-supervised learning on offline datasets have shown remarkable capabilities in text and image domains, achieving the same generalisation for agents that act in sequential decision problems remains an open challenge. In this work, we take a step towards this goal by procedurally generating tens of millions of 2D physics-based tasks and using these to train a general reinforcement learning (RL) agent for physical control. To this end, we introduce Kinetix: an open-ended space of physics-based RL environments that can represent tasks ranging from robotic locomotion and grasping to video games and classic RL environments, all within a unified framework. Kinetix makes use of our novel hardware-accelerated physics engine Jax2D that allows us to cheaply simulate billions of environment steps during training. Our trained agent exhibits strong physical reasoning capabilities, being able to zero-shot solve unseen human-designed environments. Furthermore, fine-tuning this general agent on tasks of interest shows significantly stronger performance than training an RL agent *tabula rasa*. This includes solving some environments that standard RL training completely fails at. We believe this demonstrates the feasibility of large scale, mixed-quality pre-training for online RL and we hope that Kinetix will serve as a useful framework to investigate this further.
Self-Supervised Learning (SSL) presents an exciting opportunity to unlock the potential of vast, untapped clinical datasets, for various downstream applications that suffer from the scarcity of labeled data. While SSL has revolutionized fields like natural language processing and computer vision, its adoption in 3D medical image computing has been limited by three key pitfalls: Small pre-training dataset sizes, architectures inadequate for 3D medical image analysis, and insufficient evaluation practices. In this paper, we address these issues by i) leveraging a large-scale dataset of 39k 3D brain MRI volumes and ii) using a Residual Encoder U-Net architecture within the state-of-the-art nnU-Net framework. iii) A robust development framework, incorporating 5 development and 8 testing brain MRI segmentation datasets, allowed performance-driven design decisions to optimize the simple concept of Masked Auto Encoders (MAEs) for 3D CNNs. The resulting model not only surpasses previous SSL methods but also outperforms the strong nnU-Net baseline by an average of approximately 3 Dice points setting a new state-of-the-art. Our code and models are made available here.
Self-supervised learning (SSL) has shown remarkable data representation capabilities across a wide range of datasets. However, when applied to real-world datasets with long-tailed distributions, performance on multiple downstream tasks degrades significantly. Recently, the community has begun to focus more on self-supervised long-tailed learning. Some works attempt to transfer temperature mechanisms to self-supervised learning or use category-space uniformity constraints to balance the representation of different categories in the embedding space to fight against long-tail distributions. However, most of these approaches focus on the joint optimization of all samples in the dataset or on constraining the category distribution, with little attention given to whether each individual sample is optimally guided during training. To address this issue, we propose Temperature Auxiliary Sample-level Encouragement (TASE). We introduce pseudo-labels into self-supervised long-tailed learning, utilizing pseudo-label information to drive a dynamic temperature and re-weighting strategy. Specifically, We assign an optimal temperature parameter to each sample. Additionally, we analyze the lack of quantity awareness in the temperature parameter and use re-weighting to compensate for this deficiency, thereby achieving optimal training patterns at the sample level. Comprehensive experimental results on six benchmarks across three datasets demonstrate that our method achieves outstanding performance in improving long-tail recognition, while also exhibiting high robustness.
Emergent communication, driven by generative models, enables agents to develop a shared language for describing their individual views of the same objects through interactions. Meanwhile, self-supervised learning (SSL), particularly SimSiam, uses discriminative representation learning to make representations of augmented views of the same data point closer in the representation space. Building on the prior work of VI-SimSiam, which incorporates a generative and Bayesian perspective into the SimSiam framework via variational inference (VI) interpretation, we propose SimSiam+VAE, a unified approach for both representation learning and emergent communication. SimSiam+VAE integrates a variational autoencoder (VAE) into the predictor of the SimSiam network to enhance representation learning and capture uncertainty. Experimental results show that SimSiam+VAE outperforms both SimSiam and VI-SimSiam. We further extend this model into a communication framework called the SimSiam Naming Game (SSNG), which applies the generative and Bayesian approach based on VI to develop internal representations and emergent language, while utilizing the discriminative process of SimSiam to facilitate mutual understanding between agents. In experiments with established models, despite the dynamic alternation of agent roles during interactions, SSNG demonstrates comparable performance to the referential game and slightly outperforms the Metropolis-Hastings naming game.
Parkinson's disease (PD) is a progressive neurological disorder that impacts the quality of life significantly, making in-home monitoring of motor symptoms such as Freezing of Gait (FoG) critical. However, existing symptom monitoring technologies are power-hungry, rely on extensive amounts of labeled data, and operate in controlled settings. These shortcomings limit real-world deployment of the technology. This work presents LIFT-PD, a computationally-efficient self-supervised learning framework for real-time FoG detection. Our method combines self-supervised pre-training on unlabeled data with a novel differential hopping windowing technique to learn from limited labeled instances. An opportunistic model activation module further minimizes power consumption by selectively activating the deep learning module only during active periods. Extensive experimental results show that LIFT-PD achieves a 7.25% increase in precision and 4.4% improvement in accuracy compared to supervised models while using as low as 40% of the labeled training data used for supervised learning. Additionally, the model activation module reduces inference time by up to 67% compared to continuous inference. LIFT-PD paves the way for practical, energy-efficient, and unobtrusive in-home monitoring of PD patients with minimal labeling requirements.
Supervised learning has become a cornerstone of modern machine learning, yet a comprehensive theory explaining its effectiveness remains elusive. Empirical phenomena, such as neural analogy-making and the linear representation hypothesis, suggest that supervised models can learn interpretable factors of variation in a linear fashion. Recent advances in self-supervised learning, particularly nonlinear Independent Component Analysis, have shown that these methods can recover latent structures by inverting the data generating process. We extend these identifiability results to parametric instance discrimination, then show how insights transfer to the ubiquitous setting of supervised learning with cross-entropy minimization. We prove that even in standard classification tasks, models learn representations of ground-truth factors of variation up to a linear transformation. We corroborate our theoretical contribution with a series of empirical studies. First, using simulated data matching our theoretical assumptions, we demonstrate successful disentanglement of latent factors. Second, we show that on DisLib, a widely-used disentanglement benchmark, simple classification tasks recover latent structures up to linear transformations. Finally, we reveal that models trained on ImageNet encode representations that permit linear decoding of proxy factors of variation. Together, our theoretical findings and experiments offer a compelling explanation for recent observations of linear representations, such as superposition in neural networks. This work takes a significant step toward a cohesive theory that accounts for the unreasonable effectiveness of supervised deep learning.
LIFT-PD is an innovative self-supervised learning framework developed for real-time detection of Freezing of Gait (FoG) in Parkinson's Disease (PD) patients, using a single triaxial accelerometer. It minimizes the reliance on large labeled datasets by applying a Differential Hopping Windowing Technique (DHWT) to address imbalanced data during training. Additionally, an Opportunistic Inference Module is used to reduce energy consumption by activating the model only during active movement periods. Extensive testing on publicly available datasets showed that LIFT-PD improved precision by 7.25% and accuracy by 4.4% compared to supervised models, while using 40% fewer labeled samples and reducing inference time by 67%. These findings make LIFT-PD a highly practical and energy-efficient solution for continuous, in-home monitoring of PD patients.
Photoplethysmography (PPG) is the leading non-invasive technique for monitoring biosignals and cardiovascular health, with widespread adoption in both clinical settings and consumer wearable devices. While machine learning models trained on PPG signals have shown promise, they tend to be task-specific and struggle with generalization. Current research is limited by the use of single-device datasets, insufficient exploration of out-of-domain generalization, and a lack of publicly available models, which hampers reproducibility. To address these limitations, we present PaPaGei, the first open foundation model for PPG signals. The model is pre-trained on over 57,000 hours of data, comprising 20 million unlabeled PPG segments from publicly available datasets. We introduce a novel representation learning approach that leverages domain knowledge of PPG signal morphology across individuals, enabling the capture of richer representations compared to traditional contrastive learning methods. We evaluate PaPaGei against state-of-the-art time-series foundation models and self-supervised learning benchmarks across 20 tasks from 10 diverse datasets, spanning cardiovascular health, sleep disorders, pregnancy monitoring, and wellbeing assessment. Our model demonstrates superior performance, improving classification and regression metrics by 6.3% and 2.9% respectively in at least 14 tasks. Notably, PaPaGei achieves these results while being more data- and parameter-efficient, outperforming models that are 70x larger. Beyond accuracy, we examine model robustness across different skin tones, establishing a benchmark for bias evaluation in future models. PaPaGei can serve as both a feature extractor and an encoder for multimodal models, opening up new opportunities for multimodal health monitoring.
Neural encoding of artificial neural networks (ANNs) links their computational representations to brain responses, offering insights into how the brain processes information. Current studies mostly use linear encoding models for clarity, even though brain responses are often nonlinear. This has sparked interest in developing nonlinear encoding models that are still interpretable. To address this problem, we propose LinBridge, a learnable and flexible framework based on Jacobian analysis for interpreting nonlinear encoding models. LinBridge posits that the nonlinear mapping between ANN representations and neural responses can be factorized into a linear inherent component that approximates the complex nonlinear relationship, and a mapping bias that captures sample-selective nonlinearity. The Jacobian matrix, which reflects output change rates relative to input, enables the analysis of sample-selective mapping in nonlinear models. LinBridge employs a self-supervised learning strategy to extract both the linear inherent component and nonlinear mapping biases from the Jacobian matrices of the test set, allowing it to adapt effectively to various nonlinear encoding models. We validate the LinBridge framework in the scenario of neural visual encoding, using computational visual representations from CLIP-ViT to predict brain activity recorded via functional magnetic resonance imaging (fMRI). Our experimental results demonstrate that: 1) the linear inherent component extracted by LinBridge accurately reflects the complex mappings of nonlinear neural encoding models; 2) the sample-selective mapping bias elucidates the variability of nonlinearity across different levels of the visual processing hierarchy. This study presents a novel tool for interpreting nonlinear neural encoding models and offers fresh evidence about hierarchical nonlinearity distribution in the visual cortex.