We introduce NeCo: Patch Neighbor Consistency, a novel self-supervised training loss that enforces patch-level nearest neighbor consistency across a student and teacher model. Compared to contrastive approaches that only yield binary learning signals, i.e., 'attract' and 'repel', this approach benefits from the more fine-grained learning signal of sorting spatially dense features relative to reference patches. Our method leverages differentiable sorting applied on top of pretrained representations, such as DINOv2-registers to bootstrap the learning signal and further improve upon them. This dense post-pretraining leads to superior performance across various models and datasets, despite requiring only 19 hours on a single GPU. This method generates high-quality dense feature encoders and establishes several new state-of-the-art results such as +5.5% and +6% for non-parametric in-context semantic segmentation on ADE20k and Pascal VOC, +7.2% and +5.7% for linear segmentation evaluations on COCO-Things and -Stuff and improvements in the 3D understanding of multi-view consistency on SPair-71k, by more than 1.5%.
Self-supervision has the potential to transform reinforcement learning (RL), paralleling the breakthroughs it has enabled in other areas of machine learning. While self-supervised learning in other domains aims to find patterns in a fixed dataset, self-supervised goal-conditioned reinforcement learning (GCRL) agents discover new behaviors by learning from the goals achieved during unstructured interaction with the environment. However, these methods have failed to see similar success, both due to a lack of data from slow environment simulations as well as a lack of stable algorithms. We take a step toward addressing both of these issues by releasing a high-performance codebase and benchmark (JaxGCRL) for self-supervised GCRL, enabling researchers to train agents for millions of environment steps in minutes on a single GPU. By utilizing GPU-accelerated replay buffers, environments, and a stable contrastive RL algorithm, we reduce training time by up to $22\times$. Additionally, we assess key design choices in contrastive RL, identifying those that most effectively stabilize and enhance training performance. With this approach, we provide a foundation for future research in self-supervised GCRL, enabling researchers to quickly iterate on new ideas and evaluate them in diverse and challenging environments. Website + Code: https://github.com/MichalBortkiewicz/JaxGCRL
There is growing interest towards the use of AI directly onboard satellites for quick analysis and rapid response to critical events such as natural disasters. This paper presents a blueprint to the mission designer for the development of a modular and efficient deep learning payload to address multiple onboard inference tasks. In particular, we design a self-supervised lightweight backbone that provides features to efficient task-specific heads. The latter can be developed independently and with reduced data labeling requirements thanks to the frozen backbone. Experiments on three sample tasks of cloud segmentation, flood detection, and marine debris classification on a 7W embedded system show competitive results with inference quality close to high-complexity state-of-the-art models and high throughput in excess of 8 Mpx/s.
High-quality representation of transactional sequences is vital for modern banking applications, including risk management, churn prediction, and personalized customer offers. Different tasks require distinct representation properties: local tasks benefit from capturing the client's current state, while global tasks rely on general behavioral patterns. Previous research has demonstrated that various self-supervised approaches yield representations that better capture either global or local qualities. This study investigates the integration of two self-supervised learning techniques - instance-wise contrastive learning and a generative approach based on restoring masked events in latent space. The combined approach creates representations that balance local and global transactional data characteristics. Experiments conducted on several public datasets, focusing on sequence classification and next-event type prediction, show that the integrated method achieves superior performance compared to individual approaches and demonstrates synergistic effects. These findings suggest that the proposed approach offers a robust framework for advancing event sequences representation learning in the financial sector.
Fine-grained air quality (AQ) mapping is made possible by the proliferation of cheap AQ micro-stations (MSs). However, their measurements are often inaccurate and sensitive to local disturbances, in contrast to standardized stations (SSs) that provide accurate readings but fall short in number. To simultaneously address the issues of low data quality (MSs) and high label sparsity (SSs), a multi-task spatio-temporal network (MTSTN) is proposed, which employs self-supervised learning to utilize massive unlabeled data, aided by seasonal and trend decomposition of MS data offering reliable information as features. The MTSTN is applied to infer NO$_2$, O$_3$ and PM$_{2.5}$ concentrations in a 250 km$^2$ area in Chengdu, China, at a resolution of 500m$\times$500m$\times$1hr. Data from 55 SSs and 323 MSs were used, along with meteorological, traffic, geographic and timestamp data as features. The MTSTN excels in accuracy compared to several benchmarks, and its performance is greatly enhanced by utilizing low-quality MS data. A series of ablation and pressure tests demonstrate the results' robustness and interpretability, showcasing the MTSTN's practical value for accurate and affordable AQ inference.
In the Internet of Vehicles (IoV), Federated Learning (FL) provides a privacy-preserving solution by aggregating local models without sharing data. Traditional supervised learning requires image data with labels, but data labeling involves significant manual effort. Federated Self-Supervised Learning (FSSL) utilizes Self-Supervised Learning (SSL) for local training in FL, eliminating the need for labels while protecting privacy. Compared to other SSL methods, Momentum Contrast (MoCo) reduces the demand for computing resources and storage space by creating a dictionary. However, using MoCo in FSSL requires uploading the local dictionary from vehicles to Base Station (BS), which poses a risk of privacy leakage. Simplified Contrast (SimCo) addresses the privacy leakage issue in MoCo-based FSSL by using dual temperature instead of a dictionary to control sample distribution. Additionally, considering the negative impact of motion blur on model aggregation, and based on SimCo, we propose a motion blur-resistant FSSL method, referred to as BFSSL. Furthermore, we address energy consumption and delay in the BFSSL process by proposing a Deep Reinforcement Learning (DRL)-based resource allocation scheme, called DRL-BFSSL. In this scheme, BS allocates the Central Processing Unit (CPU) frequency and transmission power of vehicles to minimize energy consumption and latency, while aggregating received models based on the motion blur level. Simulation results validate the effectiveness of our proposed aggregation and resource allocation methods.
Freeform handwriting authentication verifies a person's identity from their writing style and habits in messy handwriting data. This technique has gained widespread attention in recent years as a valuable tool for various fields, e.g., fraud prevention and cultural heritage protection. However, it still remains a challenging task in reality due to three reasons: (i) severe damage, (ii) complex high-dimensional features, and (iii) lack of supervision. To address these issues, we propose SherlockNet, an energy-oriented two-branch contrastive self-supervised learning framework for robust and fast freeform handwriting authentication. It consists of four stages: (i) pre-processing: converting manuscripts into energy distributions using a novel plug-and-play energy-oriented operator to eliminate the influence of noise; (ii) generalized pre-training: learning general representation through two-branch momentum-based adaptive contrastive learning with the energy distributions, which handles the high-dimensional features and spatial dependencies of handwriting; (iii) personalized fine-tuning: calibrating the learned knowledge using a small amount of labeled data from downstream tasks; and (iv) practical application: identifying individual handwriting from scrambled, missing, or forged data efficiently and conveniently. Considering the practicality, we construct EN-HA, a novel dataset that simulates data forgery and severe damage in real applications. Finally, we conduct extensive experiments on six benchmark datasets including our EN-HA, and the results prove the robustness and efficiency of SherlockNet.
Searching on bipartite graphs serves as a fundamental task for various real-world applications, such as recommendation systems, database retrieval, and document querying. Conventional approaches rely on similarity matching in continuous Euclidean space of vectorized node embeddings. To handle intensive similarity computation efficiently, hashing techniques for graph-structured data have emerged as a prominent research direction. However, despite the retrieval efficiency in Hamming space, previous studies have encountered catastrophic performance decay. To address this challenge, we investigate the problem of hashing with Graph Convolutional Network for effective Top-N search. Our findings indicate the learning effectiveness of incorporating hashing techniques within the exploration of bipartite graph reception fields, as opposed to simply treating hashing as post-processing to output embeddings. To further enhance the model performance, we advance upon these findings and propose Bipartite Graph Contrastive Hashing (BGCH+). BGCH+ introduces a novel dual augmentation approach to both intermediate information and hash code outputs in the latent feature spaces, thereby producing more expressive and robust hash codes within a dual self-supervised learning paradigm. Comprehensive empirical analyses on six real-world benchmarks validate the effectiveness of our dual feature contrastive learning in boosting the performance of BGCH+ compared to existing approaches.
Emotion recognition using electroencephalography (EEG) signals has garnered widespread attention in recent years. However, existing studies have struggled to develop a sufficiently generalized model suitable for different datasets without re-training (cross-corpus). This difficulty arises because distribution differences across datasets far exceed the intra-dataset variability. To solve this problem, we propose a novel Soft Contrastive Masked Modeling (SCMM) framework. Inspired by emotional continuity, SCMM integrates soft contrastive learning with a new hybrid masking strategy to effectively mine the "short-term continuity" characteristics inherent in human emotions. During the self-supervised learning process, soft weights are assigned to sample pairs, enabling adaptive learning of similarity relationships across samples. Furthermore, we introduce an aggregator that weightedly aggregates complementary information from multiple close samples based on pairwise similarities among samples to enhance fine-grained feature representation, which is then used for original sample reconstruction. Extensive experiments on the SEED, SEED-IV and DEAP datasets show that SCMM achieves state-of-the-art (SOTA) performance, outperforming the second-best method by an average accuracy of 4.26% under two types of cross-corpus conditions (same-class and different-class) for EEG-based emotion recognition.
Learning a data representation for downstream supervised learning tasks under unlabeled scenario is both critical and challenging. In this paper, we propose a novel unsupervised transfer learning approach using adversarial contrastive training (ACT). Our experimental results demonstrate outstanding classification accuracy with both fine-tuned linear probe and K-NN protocol across various datasets, showing competitiveness with existing state-of-the-art self-supervised learning methods. Moreover, we provide an end-to-end theoretical guarantee for downstream classification tasks in a misspecified, over-parameterized setting, highlighting how a large amount of unlabeled data contributes to prediction accuracy. Our theoretical findings suggest that the testing error of downstream tasks depends solely on the efficiency of data augmentation used in ACT when the unlabeled sample size is sufficiently large. This offers a theoretical understanding of learning downstream tasks with a small sample size.
Geometric trees are characterized by their tree-structured layout and spatially constrained nodes and edges, which significantly impacts their topological attributes. This inherent hierarchical structure plays a crucial role in domains such as neuron morphology and river geomorphology, but traditional graph representation methods often overlook these specific characteristics of tree structures. To address this, we introduce a new representation learning framework tailored for geometric trees. It first features a unique message passing neural network, which is both provably geometrical structure-recoverable and rotation-translation invariant. To address the data label scarcity issue, our approach also includes two innovative training targets that reflect the hierarchical ordering and geometric structure of these geometric trees. This enables fully self-supervised learning without explicit labels. We validate our method's effectiveness on eight real-world datasets, demonstrating its capability to represent geometric trees.
Masked autoencoder has been widely explored in point cloud self-supervised learning, whereby the point cloud is generally divided into visible and masked parts. These methods typically include an encoder accepting visible patches (normalized) and corresponding patch centers (position) as input, with the decoder accepting the output of the encoder and the centers (position) of the masked parts to reconstruct each point in the masked patches. Then, the pre-trained encoders are used for downstream tasks. In this paper, we show a motivating empirical result that when directly feeding the centers of masked patches to the decoder without information from the encoder, it still reconstructs well. In other words, the centers of patches are important and the reconstruction objective does not necessarily rely on representations of the encoder, thus preventing the encoder from learning semantic representations. Based on this key observation, we propose a simple yet effective method, i.e., learning to Predict Centers for Point Masked AutoEncoders (PCP-MAE) which guides the model to learn to predict the significant centers and use the predicted centers to replace the directly provided centers. Specifically, we propose a Predicting Center Module (PCM) that shares parameters with the original encoder with extra cross-attention to predict centers. Our method is of high pre-training efficiency compared to other alternatives and achieves great improvement over Point-MAE, particularly surpassing it by 5.50% on OBJ-BG, 6.03% on OBJ-ONLY, and 5.17% on PB-T50-RS for 3D object classification on the ScanObjectNN dataset. The code is available at https://github.com/aHapBean/PCP-MAE.
Foundation models have triggered a paradigm shift in computer vision and are increasingly being adopted in remote sensing, particularly for multispectral imagery. Yet, their potential in hyperspectral imaging (HSI) remains untapped due to the absence of comprehensive and globally representative hyperspectral datasets. To close this gap, we introduce SpectralEarth, a large-scale multi-temporal dataset designed to pretrain hyperspectral foundation models leveraging data from the Environmental Mapping and Analysis Program (EnMAP). SpectralEarth comprises 538,974 image patches covering 415,153 unique locations from more than 11,636 globally distributed EnMAP scenes spanning two years of archive. Additionally, 17.5% of these locations include multiple timestamps, enabling multi-temporal HSI analysis. Utilizing state-of-the-art self-supervised learning (SSL) algorithms, we pretrain a series of foundation models on SpectralEarth. We integrate a spectral adapter into classical vision backbones to accommodate the unique characteristics of HSI. In tandem, we construct four downstream datasets for land-cover and crop-type mapping, providing benchmarks for model evaluation. Experimental results support the versatility of our models, showcasing their generalizability across different tasks and sensors. We also highlight computational efficiency during model fine-tuning. The dataset, models, and source code will be made publicly available.
Recently, the state space model Mamba has demonstrated efficient long-sequence modeling capabilities, particularly for addressing long-sequence visual tasks in 3D medical imaging. However, existing generative self-supervised learning methods have not yet fully unleashed Mamba's potential for handling long-range dependencies because they overlook the inherent causal properties of state space sequences in masked modeling. To address this challenge, we propose a general-purpose pre-training framework called MambaMIM, a masked image modeling method based on a novel TOKen-Interpolation strategy (TOKI) for the selective structure state space sequence, which learns causal relationships of state space within the masked sequence. Further, MambaMIM introduces a bottom-up 3D hybrid masking strategy to maintain a masking consistency across different architectures and can be used on any single or hybrid Mamba architecture to enhance its multi-scale and long-range representation capability. We pre-train MambaMIM on a large-scale dataset of 6.8K CT scans and evaluate its performance across eight public medical segmentation benchmarks. Extensive downstream experiments reveal the feasibility and advancement of using Mamba for medical image pre-training. In particular, when we apply the MambaMIM to a customized architecture that hybridizes MedNeXt and Vision Mamba, we consistently obtain the state-of-the-art segmentation performance. The code is available at: https://github.com/FengheTan9/MambaMIM.
Despite significant research efforts and advancements, cancer remains a leading cause of mortality. Early cancer prediction has become a crucial focus in cancer research to streamline patient care and improve treatment outcomes. Manual tumor detection by histopathologists can be time consuming, prompting the need for computerized methods to expedite treatment planning. Traditional approaches to tumor detection rely on supervised learning, necessitates a large amount of annotated data for model training. However, acquiring such extensive labeled data can be laborious and time\-intensive. This research examines the three learning environments: supervised learning (SL), semi\-supervised learning (Semi\-SL), and self\-supervised learning (Self\-SL): to predict kidney, lung, and breast cancer. Three pre\-trained deep learning models (Residual Network\-50, Visual Geometry Group\-16, and EfficientNetB0) are evaluated based on these learning settings using seven carefully curated training sets. To create the first training set (TS1), SL is applied to all annotated image samples. Five training sets (TS2\-TS6) with different ratios of labeled and unlabeled cancer images are used to evaluateSemi\-SL. Unlabeled cancer images from the final training set (TS7) are utilized for Self\-SL assessment. Among different learning environments, outcomes from the Semi\-SL setting show a strong degree of agreement with the outcomes achieved in the SL setting. The uniform pattern of observations from the pre\-trained models across all three datasets validates the methodology and techniques of the research. Based on modest number of labeled samples and minimal computing cost, our study suggests that the Semi\-SL option can be a highly viable replacement for the SL option under label annotation constraint scenarios.
Speech emotion recognition (SER) has made significant strides with the advent of powerful self-supervised learning (SSL) models. However, the generalization of these models to diverse languages and emotional expressions remains a challenge. We propose a large-scale benchmark to evaluate the robustness and adaptability of state-of-the-art SER models in both in-domain and out-of-domain settings. Our benchmark includes a diverse set of multilingual datasets, focusing on less commonly used corpora to assess generalization to new data. We employ logit adjustment to account for varying class distributions and establish a single dataset cluster for systematic evaluation. Surprisingly, we find that the Whisper model, primarily designed for automatic speech recognition, outperforms dedicated SSL models in cross-lingual SER. Our results highlight the need for more robust and generalizable SER models, and our benchmark serves as a valuable resource to drive future research in this direction.
Whole Slide Image (WSI) classification with multiple instance learning (MIL) in digital pathology faces significant computational challenges. Current methods mostly rely on extensive self-supervised learning (SSL) for satisfactory performance, requiring long training periods and considerable computational resources. At the same time, no pre-training affects performance due to domain shifts from natural images to WSIs. We introduce Snuffy architecture, a novel MIL-pooling method based on sparse transformers that mitigates performance loss with limited pre-training and enables continual few-shot pre-training as a competitive option. Our sparsity pattern is tailored for pathology and is theoretically proven to be a universal approximator with the tightest probabilistic sharp bound on the number of layers for sparse transformers, to date. We demonstrate Snuffy's effectiveness on CAMELYON16 and TCGA Lung cancer datasets, achieving superior WSI and patch-level accuracies. The code is available on https://github.com/jafarinia/snuffy.
Electroencephalography (EEG) research typically focuses on tasks with narrowly defined objectives, but recent studies are expanding into the use of unlabeled data within larger models, aiming for a broader range of applications. This addresses a critical challenge in EEG research. For example, Kostas et al. (2021) show that self-supervised learning (SSL) outperforms traditional supervised methods. Given the high noise levels in EEG data, we argue that further improvements are possible with additional preprocessing. Current preprocessing methods often fail to efficiently manage the large data volumes required for SSL, due to their lack of optimization, reliance on subjective manual corrections, and validation processes or inflexible protocols that limit SSL. We propose a Python-based EEG preprocessing pipeline optimized for self-supervised learning, designed to efficiently process large-scale data. This optimization not only stabilizes self-supervised training but also enhances performance on downstream tasks compared to training with raw data.
Self-supervised learning (SSL) has been shown to be a powerful approach for learning visual representations. In this study, we propose incorporating ZCA whitening as the final layer of the encoder in self-supervised learning to enhance the quality of learned features by normalizing and decorrelating them. Although whitening has been utilized in SSL in previous works, its potential to universally improve any SSL model has not been explored. We demonstrate that adding whitening as the last layer of SSL pretrained encoders is independent of the self-supervised learning method and encoder architecture, thus it improves performance for a wide range of SSL methods across multiple encoder architectures and datasets. Our experiments show that whitening is capable of improving linear and k-NN probing accuracy by 1-5%. Additionally, we propose metrics that allow for a comprehensive analysis of the learned features, provide insights into the quality of the representations and help identify collapse patterns.
Self-supervised learning (SSL) has become an important approach in pretraining large neural networks, enabling unprecedented scaling of model and dataset sizes. While recent advances like I-JEPA have shown promising results for Vision Transformers, adapting such methods to Convolutional Neural Networks (CNNs) presents unique challenges. In this paper, we introduce CNN-JEPA, a novel SSL method that successfully applies the joint embedding predictive architecture approach to CNNs. Our method incorporates a sparse CNN encoder to handle masked inputs, a fully convolutional predictor using depthwise separable convolutions, and an improved masking strategy. We demonstrate that CNN-JEPA outperforms I-JEPA with ViT architectures on ImageNet-100, achieving 73.3% linear top-1 accuracy with a standard ResNet-50 encoder. Compared to other CNN-based SSL methods, CNN-JEPA requires 17-35% less training time for the same number of epochs and approaches the linear and k-NN top-1 accuracies of BYOL, SimCLR, and VICReg. Our approach offers a simpler, more efficient alternative to existing SSL methods for CNNs, requiring minimal augmentations and no separate projector network.