Scaling has not yet been convincingly demonstrated for pure self-supervised learning from video. However, prior work has focused evaluations on semantic-related tasks $\unicode{x2013}$ action classification, ImageNet classification, etc. In this paper we focus on evaluating self-supervised learning on non-semantic vision tasks that are more spatial (3D) and temporal (+1D = 4D), such as camera pose estimation, point and object tracking, and depth estimation. We show that by learning from very large video datasets, masked auto-encoding (MAE) with transformer video models actually scales, consistently improving performance on these 4D tasks, as model size increases from 20M all the way to the largest by far reported self-supervised video model $\unicode{x2013}$ 22B parameters. Rigorous apples-to-apples comparison with many recent image and video models demonstrates the benefits of scaling 4D representations.

Self-supervision is often used for pre-training to foster performance on a downstream task by constructing meaningful representations of samples. Self-supervised learning (SSL) generally involves generating different views of the same sample and thus requires data augmentations that are challenging to construct for tabular data. This constitutes one of the main challenges of self-supervision for structured data. In the present work, we propose a novel augmentation-free SSL method for tabular data. Our approach, T-JEPA, relies on a Joint Embedding Predictive Architecture (JEPA) and is akin to mask reconstruction in the latent space. It involves predicting the latent representation of one subset of features from the latent representation of a different subset within the same sample, thereby learning rich representations without augmentations. We use our method as a pre-training technique and train several deep classifiers on the obtained representation. Our experimental results demonstrate a substantial improvement in both classification and regression tasks, outperforming models trained directly on samples in their original data space. Moreover, T-JEPA enables some methods to consistently outperform or match the performance of traditional methods likes Gradient Boosted Decision Trees. To understand why, we extensively characterize the obtained representations and show that T-JEPA effectively identifies relevant features for downstream tasks without access to the labels. Additionally, we introduce regularization tokens, a novel regularization method critical for training of JEPA-based models on structured data. Self-supervised learning has caught increasing attention in recent years due to its significant success in many applications. Self-supervision is often used for pre-training to improve models' performance on downstream tasks. In short, the objective of self-supervision for representation learning is to generate meaningful representations from unlabeled data by using pseudo-label. By pushing dissimilar samples farther away while reducing the distance between samples that are alike, self-supervised learning can facilitate learning for both supervised and unsupervised tasks.

Large Language Models (LLMs) show impressive inductive reasoning capabilities, enabling them to generate hypotheses that could generalize effectively to new instances when guided by in-context demonstrations. However, in real-world applications, LLMs' hypothesis generation is not solely determined by these demonstrations but is significantly shaped by task-specific model priors. Despite their critical influence, the distinct contributions of model priors versus demonstrations to hypothesis generation have been underexplored. This study bridges this gap by systematically evaluating three inductive reasoning strategies across five real-world tasks with three LLMs. Our empirical findings reveal that, hypothesis generation is primarily driven by the model's inherent priors; removing demonstrations results in minimal loss of hypothesis quality and downstream usage. Further analysis shows the result is consistent across various label formats with different label configurations, and prior is hard to override, even under flipped labeling. These insights advance our understanding of the dynamics of hypothesis generation in LLMs and highlight the potential for better utilizing model priors in real-world inductive reasoning tasks.

Adversarial training is a popular method to robustify models against adversarial attacks. However, it exhibits much more severe overfitting than training on clean inputs. In this work, we investigate this phenomenon from the perspective of training instances, i.e., training input-target pairs. Based on a quantitative metric measuring the relative difficulty of an instance in the training set, we analyze the model's behavior on training instances of different difficulty levels. This lets us demonstrate that the decay in generalization performance of adversarial training is a result of fitting hard adversarial instances. We theoretically verify our observations for both linear and general nonlinear models, proving that models trained on hard instances have worse generalization performance than ones trained on easy instances, and that this generalization gap increases with the size of the adversarial budget. Finally, we investigate solutions to mitigate adversarial overfitting in several scenarios, including fast adversarial training and fine-tuning a pretrained model with additional data. Our results demonstrate that using training data adaptively improves the model's robustness.

Inductive reasoning - the process of inferring general rules from a small number of observations - is a fundamental aspect of human intelligence. Recent works suggest that large language models (LLMs) can engage in inductive reasoning by sampling multiple hypotheses about the rules and selecting the one that best explains the observations. However, due to the IID sampling, semantically redundant hypotheses are frequently generated, leading to significant wastage of compute. In this paper, we 1) demonstrate that increasing the temperature to enhance the diversity is limited due to text degeneration issue, and 2) propose a novel method to improve the diversity while maintaining text quality. We first analyze the effect of increasing the temperature parameter, which is regarded as the LLM's diversity control, on IID hypotheses. Our analysis shows that as temperature rises, diversity and accuracy of hypotheses increase up to a certain point, but this trend saturates due to text degeneration. To generate hypotheses that are more semantically diverse and of higher quality, we propose a novel approach inspired by human inductive reasoning, which we call Mixture of Concepts (MoC). When applied to several inductive reasoning benchmarks, MoC demonstrated significant performance improvements compared to standard IID sampling and other approaches.

We present RelCon, a novel self-supervised Relative Contrastive learning approach that uses a learnable distance measure in combination with a softened contrastive loss for training an motion foundation model from wearable sensors. The learned distance provides a measurement of semantic similarity between a pair of accelerometer time-series segments, which is used to measure the distance between an anchor and various other sampled candidate segments. The self-supervised model is trained on 1 billion segments from 87,376 participants from a large wearables dataset. The model achieves strong performance across multiple downstream tasks, encompassing both classification and regression. To our knowledge, we are the first to show the generalizability of a self-supervised learning model with motion data from wearables across distinct evaluation tasks. Advances in self-supervised learning (SSL) combined with the availability of large-scale datasets have resulted in a proliferation of foundation models (FMs) in computer vision (Oquab et al., 2023), NLP (OpenAI et al., 2023), and speech understanding (Yang et al., 2024). These models provide powerful, general-purpose representations for a particular domain of data, and support generalization to a broad set of downstream tasks without the need for finetuning. For example, the image representation contained in the DINOv2 (Oquab et al., 2023) model was trained in an entirely selfsupervised way and achieves state-of-the-art performance on multiple dense image prediction tasks such as depth estimation and semantic segmentation, by decoding a frozen base representation with task-specific heads. In contrast to these advances, the times-series have not yet benefited from the foundation model approach, with a few exceptions (Abbaspourazad et al., 2024; Das et al., 2023). This is particularly unfortunate for problems in mobile health (mHealth) signal analysis, which encompasses data modalities such as accelerometry, PPG, and ECG (Rehg et al., 2017), as the collection of mHealth data from participants can be time-consuming and expensive. However, recent advances in self-supervised learning for mHealth signals (Abbaspourazad et al., 2024; Yuan et al., 2024; Xu et al., 2024) have shown promising performance, raising the question of whether it is now feasible to train foundation models for mHealth signals. In this paper, we demonstrate, for the first time, the feasibility of adopting a foundation model approach for the analysis of accelerometry data across tasks. Accelerometry is an important mHealth signal modality that is used in human activity recognition (HAR) (Haresamudram et al., 2022), physical health status assessment (Xu et al., 2022), energy expenditure estimation (Stutz et al., 2024), and gait assessment (Apple, 2021), among many other tasks.

Tabular data poses unique challenges due to its heterogeneous nature, combining both continuous and categorical variables. Existing approaches often struggle to effectively capture the underlying structure and relationships within such data. We propose GFTab (Geodesic Flow Kernels for Semi- Supervised Learning on Mixed-Variable Tabular Dataset), a semi-supervised framework specifically designed for tabular datasets. GFTab incorporates three key innovations: 1) Variable-specific corruption methods tailored to the distinct properties of continuous and categorical variables, 2) A Geodesic flow kernel based similarity measure to capture geometric changes between corrupted inputs, and 3) Tree-based embedding to leverage hierarchical relationships from available labeled data. To rigorously evaluate GFTab, we curate a comprehensive set of 21 tabular datasets spanning various domains, sizes, and variable compositions. Our experimental results show that GFTab outperforms existing ML/DL models across many of these datasets, particularly in settings with limited labeled data.

Text representation learning is significant as the cornerstone of natural language processing. In recent years, graph contrastive learning (GCL) has been widely used in text representation learning due to its ability to represent and capture complex text information in a self-supervised setting. However, current mainstream graph contrastive learning methods often require the incorporation of domain knowledge or cumbersome computations to guide the data augmentation process, which significantly limits the application efficiency and scope of GCL. Additionally, many methods learn text representations only by constructing word-document relationships, which overlooks the rich contextual semantic information in the text. To address these issues and exploit representative textual semantics, we present an event-based, simple, and effective graph contrastive learning (SE-GCL) for text representation. Precisely, we extract event blocks from text and construct internal relation graphs to represent inter-semantic interconnections, which can ensure that the most critical semantic information is preserved. Then, we devise a streamlined, unsupervised graph contrastive learning framework to leverage the complementary nature of the event semantic and structural information for intricate feature data capture. In particular, we introduce the concept of an event skeleton for core representation semantics and simplify the typically complex data augmentation techniques found in existing graph contrastive learning to boost algorithmic efficiency. We employ multiple loss functions to prompt diverse embeddings to converge or diverge within a confined distance in the vector space, ultimately achieving a harmonious equilibrium. We conducted experiments on the proposed SE-GCL on four standard data sets (AG News, 20NG, SougouNews, and THUCNews) to verify its effectiveness in text representation learning.

Image Aesthetic Assessment (IAA) is a vital and intricate task that entails analyzing and assessing an image's aesthetic values, and identifying its highlights and areas for improvement. Traditional methods of IAA often concentrate on a single aesthetic task and suffer from inadequate labeled datasets, thus impairing in-depth aesthetic comprehension. Despite efforts to overcome this challenge through the application of Multi-modal Large Language Models (MLLMs), such models remain underdeveloped for IAA purposes. To address this, we propose a comprehensive aesthetic MLLM capable of nuanced aesthetic insight. Central to our approach is an innovative multi-scale text-guided self-supervised learning technique. This technique features a multi-scale feature alignment module and capitalizes on a wealth of unlabeled data in a self-supervised manner to structurally and functionally enhance aesthetic ability. The empirical evidence indicates that accompanied with extensive instruct-tuning, our model sets new state-of-the-art benchmarks across multiple tasks, including aesthetic scoring, aesthetic commenting, and personalized image aesthetic assessment. Remarkably, it also demonstrates zero-shot learning capabilities in the emerging task of aesthetic suggesting. Furthermore, for personalized image aesthetic assessment, we harness the potential of in-context learning and showcase its inherent advantages.

Understanding the role of data augmentations is critical for applying Self-Supervised Learning (SSL) methods in new domains. Data augmentations are commonly understood as encoding invariances into the learned representations. This interpretation suggests that SSL would require diverse augmentations that resemble the original data. However, in practice, augmentations do not need to be similar to the original data nor be diverse, and can be neither at the same time. We provide a theoretical insight into this phenomenon. We show that for different SSL losses, any non-redundant representation can be learned with a single suitable augmentation. We provide an algorithm to reconstruct such augmentations and give insights into augmentation choices in SSL.