Pre-training on image-text colonoscopy records offers substantial potential for improving endoscopic image analysis, but faces challenges including non-informative background images, complex medical terminology, and ambiguous multi-lesion descriptions. We introduce Endo-CLIP, a novel self-supervised framework that enhances Contrastive Language-Image Pre-training (CLIP) for this domain. Endo-CLIP's three-stage framework--cleansing, attunement, and unification--addresses these challenges by (1) removing background frames, (2) leveraging large language models to extract clinical attributes for fine-grained contrastive learning, and (3) employing patient-level cross-attention to resolve multi-polyp ambiguities. Extensive experiments demonstrate that Endo-CLIP significantly outperforms state-of-the-art pre-training methods in zero-shot and few-shot polyp detection and classification, paving the way for more accurate and clinically relevant endoscopic analysis.
Current applications of self-supervised learning to wireless channel representation often borrow paradigms developed for text and image processing, without fully addressing the unique characteristics and constraints of wireless communications. Aiming to fill this gap, we first propose WiMAE (Wireless Masked Autoencoder), a transformer-based encoder-decoder foundation model pretrained on a realistic open-source multi-antenna wireless channel dataset. Building upon this foundation, we develop ContraWiMAE, which enhances WiMAE by incorporating a contrastive learning objective alongside the reconstruction task in a unified multi-task framework. By warm-starting from pretrained WiMAE weights and generating positive pairs via noise injection, the contrastive component enables the model to capture both structural and discriminative features, enhancing representation quality beyond what reconstruction alone can achieve. Through extensive evaluation on unseen scenarios, we demonstrate the effectiveness of both approaches across multiple downstream tasks, with ContraWiMAE showing further improvements in linear separability and adaptability in diverse wireless environments. Comparative evaluations against a state-of-the-art wireless channel foundation model confirm the superior performance and data efficiency of our models, highlighting their potential as powerful baselines for future research in self-supervised wireless channel representation learning.
While pre-trained language models excel at semantic understanding, they often struggle to capture nuanced affective information critical for affective recognition tasks. To address these limitations, we propose a novel framework for enhancing emotion-aware embeddings in transformer-based models. Our approach introduces a continuous valence-arousal labeling system to guide contrastive learning, which captures subtle and multi-dimensional emotional nuances more effectively. Furthermore, we employ a dynamic token perturbation mechanism, using gradient-based saliency to focus on sentiment-relevant tokens, improving model sensitivity to emotional cues. The experimental results demonstrate that the proposed framework outperforms existing methods, achieving up to 15.5% improvement in the emotion classification benchmark, highlighting the importance of employing continuous labels. This improvement demonstrates that the proposed framework is effective in affective representation learning and enables precise and contextually relevant emotional understanding.
Multifarious intent detection predictors are developed for different languages, including English, Chinese and French, however, the field remains underdeveloped for Urdu, the 10th most spoken language. In the realm of well-known languages, intent detection predictors utilize the strategy of few-shot learning and prediction of unseen classes based on the model training on seen classes. However, Urdu language lacks few-shot strategy based intent detection predictors and traditional predictors are focused on prediction of the same classes which models have seen in the train set. To empower Urdu language specific intent detection, this introduces a unique contrastive learning approach that leverages unlabeled Urdu data to re-train pre-trained language models. This re-training empowers LLMs representation learning for the downstream intent detection task. Finally, it reaps the combined potential of pre-trained LLMs and the prototype-informed attention mechanism to create a comprehensive end-to-end LLMPIA intent detection pipeline. Under the paradigm of proposed predictive pipeline, it explores the potential of 6 distinct language models and 13 distinct similarity computation methods. The proposed framework is evaluated on 2 public benchmark datasets, namely ATIS encompassing 5836 samples and Web Queries having 8519 samples. Across ATIS dataset under 4-way 1 shot and 4-way 5 shot experimental settings LLMPIA achieved 83.28% and 98.25% F1-Score and on Web Queries dataset produced 76.23% and 84.42% F1-Score, respectively. In an additional case study on the Web Queries dataset under same classes train and test set settings, LLMPIA outperformed state-of-the-art predictor by 53.55% F1-Score.
Based on the concept that ventral visual stream (VVS) mainly functions for object recognition, current unsupervised task-driven methods model VVS by contrastive learning, and have achieved good brain similarity. However, we believe functions of VVS extend beyond just object recognition. In this paper, we introduce an additional function involving VVS, named relative position (RP) prediction. We first theoretically explain contrastive learning may be unable to yield the model capability of RP prediction. Motivated by this, we subsequently integrate RP learning with contrastive learning, and propose a new unsupervised task-driven method to model VVS, which is more inline with biological reality. We conduct extensive experiments, demonstrating that: (i) our method significantly improves downstream performance of object recognition while enhancing RP predictivity; (ii) RP predictivity generally improves the model brain similarity. Our results provide strong evidence for the involvement of VVS in location perception (especially RP prediction) from a computational perspective.
Benefiting from the effectiveness of graph neural networks (GNNs) and contrastive learning, GNN-based contrastive learning has become mainstream for knowledge-aware recommendation. However, most existing contrastive learning-based methods have difficulties in effectively capturing the underlying hierarchical structure within user-item bipartite graphs and knowledge graphs. Moreover, they commonly generate positive samples for contrastive learning by perturbing the graph structure, which may lead to a shift in user preference learning. To overcome these limitations, we propose hyperbolic contrastive learning with model-augmentation for knowledge-aware recommendation. To capture the intrinsic hierarchical graph structures, we first design a novel Lorentzian knowledge aggregation mechanism, which enables more effective representations of users and items. Then, we propose three model-level augmentation techniques to assist Hyperbolic contrastive learning. Different from the classical structure-level augmentation (e.g., edge dropping), the proposed model-augmentations can avoid preference shifts between the augmented positive pair. Finally, we conduct extensive experiments to demonstrate the superiority (maximum improvement of $11.03\%$) of proposed methods over existing baselines.
Recent proliferation of generative AI tools for visual content creation-particularly in the context of visual artworks-has raised serious concerns about copyright infringement and forgery. The large-scale datasets used to train these models often contain a mixture of copyrighted and non-copyrighted artworks. Given the tendency of generative models to memorize training patterns, they are susceptible to varying degrees of copyright violation. Building on the recently proposed DeepfakeArt Challenge benchmark, this work introduces DFA-CON, a contrastive learning framework designed to detect copyright-infringing or forged AI-generated art. DFA-CON learns a discriminative representation space, posing affinity among original artworks and their forged counterparts within a contrastive learning framework. The model is trained across multiple attack types, including inpainting, style transfer, adversarial perturbation, and cutmix. Evaluation results demonstrate robust detection performance across most attack types, outperforming recent pretrained foundation models. Code and model checkpoints will be released publicly upon acceptance.
Dense retrievers utilize pre-trained backbone language models (e.g., BERT, LLaMA) that are fine-tuned via contrastive learning to perform the task of encoding text into sense representations that can be then compared via a shallow similarity operation, e.g. inner product. Recent research has questioned the role of fine-tuning vs. that of pre-training within dense retrievers, specifically arguing that retrieval knowledge is primarily gained during pre-training, meaning knowledge not acquired during pre-training cannot be sub-sequentially acquired via fine-tuning. We revisit this idea here as the claim was only studied in the context of a BERT-based encoder using DPR as representative dense retriever. We extend the previous analysis by testing other representation approaches (comparing the use of CLS tokens with that of mean pooling), backbone architectures (encoder-only BERT vs. decoder-only LLaMA), and additional datasets (MSMARCO in addition to Natural Questions). Our study confirms that in DPR tuning, pre-trained knowledge underpins retrieval performance, with fine-tuning primarily adjusting neuron activation rather than reorganizing knowledge. However, this pattern does not hold universally, such as in mean-pooled (Contriever) and decoder-based (LLaMA) models. We ensure full reproducibility and make our implementation publicly available at https://github.com/ielab/DenseRetriever-Knowledge-Acquisition.
This paper formalizes an emerging learning paradigm that uses a trained model as a reference to guide and enhance the training of a target model through strategic data selection or weighting, named $\textbf{model steering}$. While ad-hoc methods have been used in various contexts, including the training of large foundation models, its underlying principles remain insufficiently understood, leading to sub-optimal performance. In this work, we propose a theory-driven framework for model steering called $\textbf{DRRho risk minimization}$, which is rooted in Distributionally Robust Optimization (DRO). Through a generalization analysis, we provide theoretical insights into why this approach improves generalization and data efficiency compared to training without a reference model. To the best of our knowledge, this is the first time such theoretical insights are provided for the new learning paradigm, which significantly enhance our understanding and practice of model steering. Building on these insights and the connection between contrastive learning and DRO, we introduce a novel method for Contrastive Language-Image Pretraining (CLIP) with a reference model, termed DRRho-CLIP. Extensive experiments validate the theoretical insights, reveal a superior scaling law compared to CLIP without a reference model, and demonstrate its strength over existing heuristic approaches.
The task of weakly supervised temporal sentence grounding (WSTSG) aims to detect temporal intervals corresponding to a language description from untrimmed videos with only video-level video-language correspondence. For an anchor sample, most existing approaches generate negative samples either from other videos or within the same video for contrastive learning. However, some training samples are highly similar to the anchor sample, directly regarding them as negative samples leads to difficulties for optimization and ignores the correlations between these similar samples and the anchor sample. To address this, we propose Positive Sample Mining (PSM), a novel framework that mines positive samples from the training set to provide more discriminative supervision. Specifically, for a given anchor sample, we partition the remaining training set into semantically similar and dissimilar subsets based on the similarity of their text queries. To effectively leverage these correlations, we introduce a PSM-guided contrastive loss to ensure that the anchor proposal is closer to similar samples and further from dissimilar ones. Additionally, we design a PSM-guided rank loss to ensure that similar samples are closer to the anchor proposal than to the negative intra-video proposal, aiming to distinguish the anchor proposal and the negative intra-video proposal. Experiments on the WSTSG and grounded VideoQA tasks demonstrate the effectiveness and superiority of our method.
We present MarkMatch, a retrieval system for detecting whether two paper ballot marks were filled by the same hand. Unlike the previous SOTA method BubbleSig, which used binary classification on isolated mark pairs, MarkMatch ranks stylistic similarity between a query mark and a mark in the database using contrastive learning. Our model is trained with a dense batch similarity matrix and a dual loss objective. Each sample is contrasted against many negatives within each batch, enabling the model to learn subtle handwriting difference and improve generalization under handwriting variation and visual noise, while diagonal supervision reinforces high confidence on true matches. The model achieves an F1 score of 0.943, surpassing BubbleSig's best performance. MarkMatch also integrates Segment Anything Model for flexible mark extraction via box- or point-based prompts. The system offers election auditors a practical tool for visual, non-biometric investigation of suspicious ballots.
Leaf disease identification plays a pivotal role in smart agriculture. However, many existing studies still struggle to integrate image and textual modalities to compensate for each other's limitations. Furthermore, many of these approaches rely on pretraining with constrained datasets such as ImageNet, which lack domain-specific information. We propose SCOLD (Soft-target COntrastive learning for Leaf Disease identification), a context-aware vision-language foundation model tailored to address these challenges for agricultural tasks. SCOLD is developed using a diverse corpus of plant leaf images and corresponding symptom descriptions, comprising over 186,000 image-caption pairs aligned with 97 unique concepts. Through task-agnostic pretraining, SCOLD leverages contextual soft targets to mitigate overconfidence in contrastive learning by smoothing labels, thereby improving model generalization and robustness on fine-grained classification tasks. Experimental results demonstrate that SCOLD outperforms existing vision-language models such as OpenAI-CLIP-L, BioCLIP, and SigLIP2 across several benchmarks, including zero-shot and few-shot classification, image-text retrieval, and image classification, while maintaining a competitive parameter footprint. Ablation studies further highlight SCOLD's effectiveness in contrast to its counterparts. The proposed approach significantly advances the agricultural vision-language foundation model, offering strong performance with minimal or no supervised fine-tuning. This work lays a solid groundwork for future research on models trained with long-form and simplified contexts, tasks involving class ambiguity, and multi-modal systems for intelligent plant disease diagnostics. The code for this study is available at https://huggingface.co/enalis/scold
In this paper, we propose a diffusion model that integrates a representation-conditioning mechanism, where the representations derived from a Vision Transformer (ViT) are used to condition the internal process of a Transformer-based diffusion model. This approach enables representation-conditioned data generation, addressing the challenge of requiring large-scale labeled datasets by leveraging self-supervised learning on unlabeled data. We evaluate our method through a zero-shot classification task for hematoma detection in brain imaging. Compared to the strong contrastive learning baseline, DINOv2, our method achieves a notable improvement of +6.15% in accuracy and +13.60% in F1-score, demonstrating its effectiveness in image classification.
Most existing text recognition methods are trained on large-scale synthetic datasets due to the scarcity of labeled real-world datasets. Synthetic images, however, cannot faithfully reproduce real-world scenarios, such as uneven illumination, irregular layout, occlusion, and degradation, resulting in performance disparities when handling complex real-world images. Recent self-supervised learning techniques, notably contrastive learning and masked image modeling (MIM), narrow this domain gap by exploiting unlabeled real text images. This study first analyzes the original Masked AutoEncoder (MAE) and observes that random patch masking predominantly captures low-level textural features but misses high-level contextual representations. To fully exploit the high-level contextual representations, we introduce random blockwise and span masking in the text recognition task. These strategies can mask the continuous image patches and completely remove some characters, forcing the model to infer relationships among characters within a word. Our Multi-Masking Strategy (MMS) integrates random patch, blockwise, and span masking into the MIM frame, which jointly learns low and high-level textual representations. After fine-tuning with real data, MMS outperforms the state-of-the-art self-supervised methods in various text-related tasks, including text recognition, segmentation, and text-image super-resolution.
Multimodal news contains a wealth of information and is easily affected by deepfake modeling attacks. To combat the latest image and text generation methods, we present a new Multimodal Fake News Detection dataset (MFND) containing 11 manipulated types, designed to detect and localize highly authentic fake news. Furthermore, we propose a Shallow-Deep Multitask Learning (SDML) model for fake news, which fully uses unimodal and mutual modal features to mine the intrinsic semantics of news. Under shallow inference, we propose the momentum distillation-based light punishment contrastive learning for fine-grained uniform spatial image and text semantic alignment, and an adaptive cross-modal fusion module to enhance mutual modal features. Under deep inference, we design a two-branch framework to augment the image and text unimodal features, respectively merging with mutual modalities features, for four predictions via dedicated detection and localization projections. Experiments on both mainstream and our proposed datasets demonstrate the superiority of the model. Codes and dataset are released at https://github.com/yunan-wang33/sdml.
Graph anomaly detection is a popular and vital task in various real-world scenarios, which has been studied for several decades. Recently, many studies extending deep learning-based methods have shown preferable performance on graph anomaly detection. However, existing methods are lack of efficiency that is definitely necessary for embedded devices. Towards this end, we propose an Efficient Anomaly detection model on heterogeneous Graphs via contrastive LEarning (EAGLE) by contrasting abnormal nodes with normal ones in terms of their distances to the local context. The proposed method first samples instance pairs on meta path-level for contrastive learning. Then, a graph autoencoder-based model is applied to learn informative node embeddings in an unsupervised way, which will be further combined with the discriminator to predict the anomaly scores of nodes. Experimental results show that EAGLE outperforms the state-of-the-art methods on three heterogeneous network datasets.
Due to the scarcity of industrial data, individual equipment users, particularly start-ups, struggle to independently train a comprehensive fault diagnosis model; federated learning enables collaborative training while ensuring data privacy, making it an ideal solution. However, the diversity of working conditions leads to variations in fault modes, resulting in inconsistent label spaces across different clients. In federated diagnostic scenarios, label space inconsistency leads to local models focus on client-specific fault modes and causes local models from different clients to map different failure modes to similar feature representations, which weakens the aggregated global model's generalization. To tackle this issue, this article proposed a federated cross-domain diagnostic framework termed Federated Invariant Features Learning (FedIFL). In intra-client training, prototype contrastive learning mitigates intra-client domain shifts, subsequently, feature generating ensures local models can access distributions of other clients in a privacy-friendly manner. Besides, in cross-client training, a feature disentanglement mechanism is introduced to mitigate cross-client domain shifts, specifically, an instance-level federated instance consistency loss is designed to ensure the instance-level consistency of invariant features between different clients, furthermore, a federated instance personalization loss and an orthogonal loss are constructed to distinguish specific features that from the invariant features. Eventually, the aggregated model achieves promising generalization among global label spaces, enabling accurate fault diagnosis for target clients' Motor Driven Systems (MDSs) with inconsistent label spaces. Experiments on real-world MDSs validate the effectiveness and superiority of FedIFL in federated cross-domain diagnosis with inconsistent fault modes.
Few-shot text classification has important application value in low-resource environments. This paper proposes a strategy that combines adaptive fine-tuning, contrastive learning, and regularization optimization to improve the classification performance of Transformer-based models. Experiments on the FewRel 2.0 dataset show that T5-small, DeBERTa-v3, and RoBERTa-base perform well in few-shot tasks, especially in the 5-shot setting, which can more effectively capture text features and improve classification accuracy. The experiment also found that there are significant differences in the classification difficulty of different relationship categories. Some categories have fuzzy semantic boundaries or complex feature distributions, making it difficult for the standard cross entropy loss to learn the discriminative information required to distinguish categories. By introducing contrastive loss and regularization loss, the generalization ability of the model is enhanced, effectively alleviating the overfitting problem in few-shot environments. In addition, the research results show that the use of Transformer models or generative architectures with stronger self-attention mechanisms can help improve the stability and accuracy of few-shot classification.
Graph contrastive learning (GCL) has achieved remarkable success by following the computer vision paradigm of preserving absolute similarity between augmented views. However, this approach faces fundamental challenges in graphs due to their discrete, non-Euclidean nature -- view generation often breaks semantic validity and similarity verification becomes unreliable. Through analyzing 11 real-world graphs, we discover a universal pattern transcending the homophily-heterophily dichotomy: label consistency systematically diminishes as structural distance increases, manifesting as smooth decay in homophily graphs and oscillatory decay in heterophily graphs. We establish theoretical guarantees for this pattern through random walk theory, proving label distribution convergence and characterizing the mechanisms behind different decay behaviors. This discovery reveals that graphs naturally encode relative similarity patterns, where structurally closer nodes exhibit collectively stronger semantic relationships. Leveraging this insight, we propose RELGCL, a novel GCL framework with complementary pairwise and listwise implementations that preserve these inherent patterns through collective similarity objectives. Extensive experiments demonstrate that our method consistently outperforms 20 existing approaches across both homophily and heterophily graphs, validating the effectiveness of leveraging natural relative similarity over artificial absolute similarity.
Contrastive Representation Learning (CRL) has achieved impressive success in various domains in recent years. Nevertheless, the theoretical understanding of the generalization behavior of CRL is limited. Moreover, to the best of our knowledge, the current literature only analyzes generalization bounds under the assumption that the data tuples used for contrastive learning are independently and identically distributed. However, in practice, we are often limited to a fixed pool of reusable labeled data points, making it inevitable to recycle data across tuples to create sufficiently large datasets. Therefore, the tuple-wise independence condition imposed by previous works is invalidated. In this paper, we provide a generalization analysis for the CRL framework under non-$i.i.d.$ settings that adheres to practice more realistically. Drawing inspiration from the literature on U-statistics, we derive generalization bounds which indicate the required number of samples in each class scales as the logarithm of the covering number of the class of learnable feature representations associated to each class. Next, we apply our main results to derive excess risk bounds for common function classes such as linear maps and neural networks.