Prototype-based federated learning has emerged as a promising approach that shares lightweight prototypes to transfer knowledge among clients with data heterogeneity in a model-agnostic manner. However, existing methods often collect prototypes directly from local models, which inevitably introduce inconsistencies into representation learning due to the biased data distributions and differing model architectures among clients. In this paper, we identify that both statistical and model heterogeneity create a vicious cycle of representation inconsistency, classifier divergence, and skewed prototype alignment, which negatively impacts the performance of clients. To break the vicious cycle, we propose a novel framework named Federated Learning via Semantic Anchors (FedSA) to decouple the generation of prototypes from local representation learning. We introduce a novel perspective that uses simple yet effective semantic anchors serving as prototypes to guide local models in learning consistent representations. By incorporating semantic anchors, we further propose anchor-based regularization with margin-enhanced contrastive learning and anchor-based classifier calibration to correct feature extractors and calibrate classifiers across clients, achieving intra-class compactness and inter-class separability of prototypes while ensuring consistent decision boundaries. We then update the semantic anchors with these consistent and discriminative prototypes, which iteratively encourage clients to collaboratively learn a unified data representation with robust generalization. Extensive experiments under both statistical and model heterogeneity settings show that FedSA significantly outperforms existing prototype-based FL methods on various classification tasks.
Distance-based anomaly detection methods rely on compact and separable in-distribution (ID) embeddings to effectively delineate anomaly boundaries. Single-positive contrastive formulations suffer from class collision, promoting unnecessary intra-class variance within ID samples. While multi-positive formulations can improve inlier compactness, they fail to preserve the diversity among synthetic outliers. We address these limitations by proposing a contrastive pretext task for anomaly detection that enforces three key properties: (1) compact ID clustering to reduce intra-class variance, (2) inlier-outlier separation to enhance inter-class separation, and (3) outlier-outlier separation to maintain diversity among synthetic outliers and prevent representation collapse. These properties work together to ensure a more robust and discriminative feature space for anomaly detection. Our approach achieves approximately 12x faster convergence than NT-Xent and 7x faster than Rot-SupCon, with superior performance. On CIFAR-10, it delivers an average performance boost of 6.2% over NT-Xent and 2% over Rot-SupCon, with class-specific improvements of up to 16.9%. Our code is available at https://anonymous.4open.science/r/firm-98B6.
Knowledge Tracing (KT) is crucial in education assessment, which focuses on depicting students' learning states and assessing students' mastery of subjects. With the rise of modern online learning platforms, particularly massive open online courses (MOOCs), an abundance of interaction data has greatly advanced the development of the KT technology. Previous research commonly adopts deterministic representation to capture students' knowledge states, which neglects the uncertainty during student interactions and thus fails to model the true knowledge state in learning process. In light of this, we propose an Uncertainty-Aware Knowledge Tracing model (UKT) which employs stochastic distribution embeddings to represent the uncertainty in student interactions, with a Wasserstein self-attention mechanism designed to capture the transition of state distribution in student learning behaviors. Additionally, we introduce the aleatory uncertainty-aware contrastive learning loss, which strengthens the model's robustness towards different types of uncertainties. Extensive experiments on six real-world datasets demonstrate that UKT not only significantly surpasses existing deep learning-based models in KT prediction, but also shows unique advantages in handling the uncertainty of student interactions.
As opposed to human drivers, current autonomous driving systems still require vast amounts of labeled data to train. Recently, world models have been proposed to simultaneously enhance autonomous driving capabilities by improving the way these systems understand complex real-world environments and reduce their data demands via self-supervised pre-training. In this paper, we present AD-L-JEPA (aka Autonomous Driving with LiDAR data via a Joint Embedding Predictive Architecture), a novel self-supervised pre-training framework for autonomous driving with LiDAR data that, as opposed to existing methods, is neither generative nor contrastive. Our method learns spatial world models with a joint embedding predictive architecture. Instead of explicitly generating masked unknown regions, our self-supervised world models predict Bird's Eye View (BEV) embeddings to represent the diverse nature of autonomous driving scenes. Our approach furthermore eliminates the need to manually create positive and negative pairs, as is the case in contrastive learning. AD-L-JEPA leads to simpler implementation and enhanced learned representations. We qualitatively and quantitatively demonstrate high-quality of embeddings learned with AD-L-JEPA. We furthermore evaluate the accuracy and label efficiency of AD-L-JEPA on popular downstream tasks such as LiDAR 3D object detection and associated transfer learning. Our experimental evaluation demonstrates that AD-L-JEPA is a plausible approach for self-supervised pre-training in autonomous driving applications and is the best available approach outperforming SOTA, including most recently proposed Occupancy-MAE [1] and ALSO [2]. The source code of AD-L-JEPA is available at https://github.com/HaoranZhuExplorer/AD-L-JEPA-Release.
Over the last decade, representation learning, which embeds complex information extracted from large amounts of data into dense vector spaces, has emerged as a key technique in machine learning. Among other applications, it has been a key building block for large language models and advanced computer vision systems based on contrastive learning. A core component of representation learning systems is the projection head, which maps the original embeddings into different, often compressed spaces, while preserving the similarity relationship between vectors. In this paper, we propose a quantum-inspired projection head that includes a corresponding quantum-inspired similarity metric. Specifically, we map classical embeddings onto quantum states in Hilbert space and introduce a quantum circuit-based projection head to reduce embedding dimensionality. To evaluate the effectiveness of this approach, we extended the BERT language model by integrating our projection head for embedding compression. We compared the performance of embeddings, which were compressed using our quantum-inspired projection head, with those compressed using a classical projection head on information retrieval tasks using the TREC 2019 and TREC 2020 Deep Learning benchmarks. The results demonstrate that our quantum-inspired method achieves competitive performance relative to the classical method while utilizing 32 times fewer parameters. Furthermore, when trained from scratch, it notably excels, particularly on smaller datasets. This work not only highlights the effectiveness of the quantum-inspired approach but also emphasizes the utility of efficient, ad hoc low-entanglement circuit simulations within neural networks as a powerful quantum-inspired technique.
ChartQA presents significant challenges due to the complex distribution of chart elements and the implicit patterns embedded within the underlying data. In this chapter, we have developed a joint multimodal scene graph for charts, explicitly representing the relationships between chart elements and their associated patterns. Our proposed multimodal scene graph consists of two components: a visual graph and a textual graph, each designed to capture the structural and semantic information within the chart. To unify representations across these different modalities, we introduce a multimodal graph contrastive learning approach that learns unified representations by maximizing similarity between nodes representing the same object across multimodal graphs. The learned graph representations can be seamlessly incorporated into a transformer decoder as a soft prompt. Additionally, given the growing need for Multimodal Large Language Models (MLLMs) in zero-shot scenarios, we have designed Chain-of-Thought (CoT) prompts for MLLMs to reduce hallucinations. We tested both methods on public benchmarks such as ChartQA, OpenCQA, and ChartX, demonstrating improved performance and validating the effectiveness of our proposed methods.
Recent advancements in multimodal models have shown a strong ability in visual perception, reasoning abilities, and vision-language understanding. However, studies on visual matching ability are missing, where finding the visual correspondence of objects is essential in vision research. Our research reveals that the matching capabilities in recent multimodal LLMs (MLLMs) still exhibit systematic shortcomings, even with current strong MLLMs models, GPT-4o. In particular, we construct a Multimodal Visual Matching (MMVM) benchmark to fairly benchmark over 30 different MLLMs. The MMVM benchmark is built from 15 open-source datasets and Internet videos with manual annotation. We categorize the data samples of MMVM benchmark into eight aspects based on the required cues and capabilities to more comprehensively evaluate and analyze current MLLMs. In addition, we have designed an automatic annotation pipeline to generate the MMVM SFT dataset, including 220K visual matching data with reasoning annotation. Finally, we present CoLVA, a novel contrastive MLLM with two novel technical designs: fine-grained vision expert with object-level contrastive learning and instruction augmentation strategy. CoLVA achieves 51.06\% overall accuracy (OA) on the MMVM benchmark, surpassing GPT-4o and baseline by 8.41\% and 23.58\% OA, respectively. The results show the effectiveness of our MMVM SFT dataset and our novel technical designs. Code, benchmark, dataset, and models are available at https://github.com/zhouyiks/CoLVA.
Artificial Intelligence is revolutionizing medical practice, enhancing diagnostic accuracy and healthcare delivery. However, its adaptation in medical settings still faces significant challenges, related to data availability and privacy constraints. Synthetic data has emerged as a promising solution to mitigate these issues, addressing data scarcity while preserving privacy. Recently, Latent Diffusion Models have emerged as a powerful tool for generating high-quality synthetic data. Meanwhile, the integration of different modalities has gained interest, emphasizing the need of models capable of handle multimodal medical data. Existing approaches struggle to integrate complementary information and lack the ability to generate modalities simultaneously. To address this challenge, we present MedCoDi-M, a 6.77-billion-parameter model, designed for multimodal medical data generation, that, following Foundation Model paradigm, exploits contrastive learning and large quantity of data to build a shared latent space which capture the relationships between different data modalities. Further, we introduce the Multi-Prompt training technique, which significantly boosts MedCoDi-M's generation under different settings. We extensively validate MedCoDi-M: first we benchmark it against five competitors on the MIMIC-CXR dataset, a state-of-the-art dataset for Chest X-ray and radiological report generation. Secondly, we perform a Visual Turing Test with expert radiologists to assess the realism and clinical relevance of the generated data, ensuring alignment with real-world scenarios. Finally, we assess the utility of MedCoDi-M in addressing key challenges in the medical field, such as anonymization, data scarcity and imbalance learning. The results are promising, demonstrating the applicability of MedCoDi-M in medical contexts. Project page is at https://cosbidev.github.io/MedCoDi-M/.
Contrastive learning has gained significant attention in short text clustering, yet it has an inherent drawback of mistakenly identifying samples from the same category as negatives and then separating them in the feature space (false negative separation), which hinders the generation of superior representations. To generate more discriminative representations for efficient clustering, we propose a novel short text clustering method, called Discriminative Representation learning via \textbf{A}ttention-\textbf{E}nhanced \textbf{C}ontrastive \textbf{L}earning for Short Text Clustering (\textbf{AECL}). The \textbf{AECL} consists of two modules which are the pseudo-label generation module and the contrastive learning module. Both modules build a sample-level attention mechanism to capture similarity relationships between samples and aggregate cross-sample features to generate consistent representations. Then, the former module uses the more discriminative consistent representation to produce reliable supervision information for assist clustering, while the latter module explores similarity relationships and consistent representations optimize the construction of positive samples to perform similarity-guided contrastive learning, effectively addressing the false negative separation issue. Experimental results demonstrate that the proposed \textbf{AECL} outperforms state-of-the-art methods. If the paper is accepted, we will open-source the code.
Evaluating image captions typically relies on reference captions, which are costly to obtain and exhibit significant diversity and subjectivity. While reference-free evaluation metrics have been proposed, most focus on cross-modal evaluation between captions and images. Recent research has revealed that the modality gap generally exists in the representation of contrastive learning-based multi-modal systems, undermining the reliability of cross-modality metrics like CLIPScore. In this paper, we propose CAMScore, a cyclic reference-free automatic evaluation metric for image captioning models. To circumvent the aforementioned modality gap, CAMScore utilizes a text-to-image model to generate images from captions and subsequently evaluates these generated images against the original images. Furthermore, to provide fine-grained information for a more comprehensive evaluation, we design a three-level evaluation framework for CAMScore that encompasses pixel-level, semantic-level, and objective-level perspectives. Extensive experiment results across multiple benchmark datasets show that CAMScore achieves a superior correlation with human judgments compared to existing reference-based and reference-free metrics, demonstrating the effectiveness of the framework.
Self-supervised learning (SSL) has significantly advanced image representation learning, yet efficiency challenges persist, particularly with adversarial training. Many SSL methods require extensive epochs to achieve convergence, a demand further amplified in adversarial settings. To address this inefficiency, we revisit the robust EMP-SSL framework, emphasizing the importance of increasing the number of crops per image to accelerate learning. Unlike traditional contrastive learning, robust EMP-SSL leverages multi-crop sampling, integrates an invariance term and regularization, and reduces training epochs, enhancing time efficiency. Evaluated with both standard linear classifiers and multi-patch embedding aggregation, robust EMP-SSL provides new insights into SSL evaluation strategies. Our results show that robust crop-based EMP-SSL not only accelerates convergence but also achieves a superior balance between clean accuracy and adversarial robustness, outperforming multi-crop embedding aggregation. Additionally, we extend this approach with free adversarial training in Multi-Crop SSL, introducing the Cost-Free Adversarial Multi-Crop Self-Supervised Learning (CF-AMC-SSL) method. CF-AMC-SSL demonstrates the effectiveness of free adversarial training in reducing training time while simultaneously improving clean accuracy and adversarial robustness. These findings underscore the potential of CF-AMC-SSL for practical SSL applications. Our code is publicly available at https://github.com/softsys4ai/CF-AMC-SSL.
Satisfactory progress has been achieved recently in universal segmentation of CT images. Following the success of vision-language methods, there is a growing trend towards utilizing text prompts and contrastive learning to develop universal segmentation models. However, there exists a significant imbalance in information density between 3D images and text prompts. Moreover, the standard fully connected layer segmentation approach faces significant challenges in handling multiple classes and exhibits poor generalizability. To address these challenges, we propose the VOxel Interacting with LAnguage method (VOILA) for universal CT image segmentation. Initially, we align voxels and language into a shared representation space and classify voxels on the basis of cosine similarity. Subsequently, we develop the Voxel-Language Interaction framework to mitigate the impact of class imbalance caused by foreground-background discrepancies and variations in target volumes. Furthermore, a Complexity-Aware Sampling method is proposed to focus on region hard to segment, achieved by generating pseudo-heatmaps from a trainable Gaussian mixture distribution. Our results indicate the proposed VOILA is capable to achieve improved performance with reduced parameters and computational cost during training. Furthermore, it demonstrates significant generalizability across diverse datasets without additional fine-tuning.
Self-supervised learning (SSL) has emerged as a crucial technique in image processing, encoding, and understanding, especially for developing today's vision foundation models that utilize large-scale datasets without annotations to enhance various downstream tasks. This study introduces a novel SSL approach, Information-Maximized Soft Variable Discretization (IMSVD), for image representation learning. Specifically, IMSVD softly discretizes each variable in the latent space, enabling the estimation of their probability distributions over training batches and allowing the learning process to be directly guided by information measures. Motivated by the MultiView assumption, we propose an information-theoretic objective function to learn transform-invariant, non-travail, and redundancy-minimized representation features. We then derive a joint-cross entropy loss function for self-supervised image representation learning, which theoretically enjoys superiority over the existing methods in reducing feature redundancy. Notably, our non-contrastive IMSVD method statistically performs contrastive learning. Extensive experimental results demonstrate the effectiveness of IMSVD on various downstream tasks in terms of both accuracy and efficiency. Thanks to our variable discretization, the embedding features optimized by IMSVD offer unique explainability at the variable level. IMSVD has the potential to be adapted to other learning paradigms. Our code is publicly available at https://github.com/niuchuangnn/IMSVD.
Recent advancements in vision foundation models (VFMs) have revolutionized visual perception in 2D, yet their potential for 3D scene understanding, particularly in autonomous driving applications, remains underexplored. In this paper, we introduce LargeAD, a versatile and scalable framework designed for large-scale 3D pretraining across diverse real-world driving datasets. Our framework leverages VFMs to extract semantically rich superpixels from 2D images, which are aligned with LiDAR point clouds to generate high-quality contrastive samples. This alignment facilitates cross-modal representation learning, enhancing the semantic consistency between 2D and 3D data. We introduce several key innovations: i) VFM-driven superpixel generation for detailed semantic representation, ii) a VFM-assisted contrastive learning strategy to align multimodal features, iii) superpoint temporal consistency to maintain stable representations across time, and iv) multi-source data pretraining to generalize across various LiDAR configurations. Our approach delivers significant performance improvements over state-of-the-art methods in both linear probing and fine-tuning tasks for both LiDAR-based segmentation and object detection. Extensive experiments on eleven large-scale multi-modal datasets highlight our superior performance, demonstrating the adaptability, efficiency, and robustness in real-world autonomous driving scenarios.
Recent research has demonstrated that Large Language Models (LLMs) are not limited to text-only tasks but can also function as multimodal models across various modalities, including audio, images, and videos. In particular, research on 3D Large Multimodal Models (3D LMMs) is making notable strides, driven by the potential of processing higher-dimensional data like point clouds. However, upon closer examination, we find that the visual and textual content within each sample of existing training datasets lacks both high informational granularity and clarity, which serve as a bottleneck for precise cross-modal understanding. To address these issues, we propose CL3DOR, Contrastive Learning for 3D large multimodal models via Odds ratio on high-Resolution point clouds, designed to ensure greater specificity and clarity in both visual and textual content. Specifically, we increase the density of point clouds per object and construct informative hard negative responses in the training dataset to penalize unwanted responses. To leverage hard negative responses, we incorporate the odds ratio as an auxiliary term for contrastive learning into the conventional language modeling loss. CL3DOR achieves state-of-the-art performance in 3D scene understanding and reasoning benchmarks. Additionally, we demonstrate the effectiveness of CL3DOR's key components through extensive experiments.
Product Attribute Value Identification (PAVI) involves identifying attribute values from product profiles, a key task for improving product search, recommendations, and business analytics on e-commerce platforms. However, existing PAVI methods face critical challenges, such as inferring implicit values, handling out-of-distribution (OOD) values, and producing normalized outputs. To address these limitations, we introduce Taxonomy-Aware Contrastive Learning Retrieval (TACLR), the first retrieval-based method for PAVI. TACLR formulates PAVI as an information retrieval task by encoding product profiles and candidate values into embeddings and retrieving values based on their similarity to the item embedding. It leverages contrastive training with taxonomy-aware hard negative sampling and employs adaptive inference with dynamic thresholds. TACLR offers three key advantages: (1) it effectively handles implicit and OOD values while producing normalized outputs; (2) it scales to thousands of categories, tens of thousands of attributes, and millions of values; and (3) it supports efficient inference for high-load industrial scenarios. Extensive experiments on proprietary and public datasets validate the effectiveness and efficiency of TACLR. Moreover, it has been successfully deployed in a real-world e-commerce platform, processing millions of product listings daily while supporting dynamic, large-scale attribute taxonomies.
Action Quality Assessment (AQA), which aims at automatic and fair evaluation of athletic performance, has gained increasing attention in recent years. However, athletes are often in rapid movement and the corresponding visual appearance variances are subtle, making it challenging to capture fine-grained pose differences and leading to poor estimation performance. Furthermore, most common AQA tasks, such as diving in sports, are usually divided into multiple sub-actions, each of which contains different durations. However, existing methods focus on segmenting the video into fixed frames, which disrupts the temporal continuity of sub-actions resulting in unavoidable prediction errors. To address these challenges, we propose a novel action quality assessment method through hierarchically pose-guided multi-stage contrastive regression. Firstly, we introduce a multi-scale dynamic visual-skeleton encoder to capture fine-grained spatio-temporal visual and skeletal features. Then, a procedure segmentation network is introduced to separate different sub-actions and obtain segmented features. Afterwards, the segmented visual and skeletal features are both fed into a multi-modal fusion module as physics structural priors, to guide the model in learning refined activity similarities and variances. Finally, a multi-stage contrastive learning regression approach is employed to learn discriminative representations and output prediction results. In addition, we introduce a newly-annotated FineDiving-Pose Dataset to improve the current low-quality human pose labels. In experiments, the results on FineDiving and MTL-AQA datasets demonstrate the effectiveness and superiority of our proposed approach. Our source code and dataset are available at https://github.com/Lumos0507/HP-MCoRe.
With the growing significance of network security, the classification of encrypted traffic has emerged as an urgent challenge. Traditional byte-based traffic analysis methods are constrained by the rigid granularity of information and fail to fully exploit the diverse correlations between bytes. To address these limitations, this paper introduces MH-Net, a novel approach for classifying network traffic that leverages multi-view heterogeneous traffic graphs to model the intricate relationships between traffic bytes. The essence of MH-Net lies in aggregating varying numbers of traffic bits into multiple types of traffic units, thereby constructing multi-view traffic graphs with diverse information granularities. By accounting for different types of byte correlations, such as header-payload relationships, MH-Net further endows the traffic graph with heterogeneity, significantly enhancing model performance. Notably, we employ contrastive learning in a multi-task manner to strengthen the robustness of the learned traffic unit representations. Experiments conducted on the ISCX and CIC-IoT datasets for both the packet-level and flow-level traffic classification tasks demonstrate that MH-Net achieves the best overall performance compared to dozens of SOTA methods.
Spectral Graph Neural Networks effectively handle graphs with different homophily levels, with low-pass filter mining feature smoothness and high-pass filter capturing differences. When these distinct filters could naturally form two opposite views for self-supervised learning, the commonalities between the counterparts for the same node remain unexplored, leading to suboptimal performance. In this paper, a simple yet effective self-supervised contrastive framework, LOHA, is proposed to address this gap. LOHA optimally leverages low-pass and high-pass views by embracing "harmony in diversity". Rather than solely maximizing the difference between these distinct views, which may lead to feature separation, LOHA harmonizes the diversity by treating the propagation of graph signals from both views as a composite feature. Specifically, a novel high-dimensional feature named spectral signal trend is proposed to serve as the basis for the composite feature, which remains relatively unaffected by changing filters and focuses solely on original feature differences. LOHA achieves an average performance improvement of 2.8% over runner-up models on 9 real-world datasets with varying homophily levels. Notably, LOHA even surpasses fully-supervised models on several datasets, which underscores the potential of LOHA in advancing the efficacy of spectral GNNs for diverse graph structures.
With the rapid growth of the Internet of Things and Cyber-Physical Systems, widespread sensor deployment has become essential. However, the high costs of building sensor networks limit their scale and coverage, making fine-grained deployment challenging. Inductive Spatio-Temporal Kriging (ISK) addresses this issue by introducing virtual sensors. Based on graph neural networks (GNNs) extracting the relationships between physical and virtual sensors, ISK can infer the measurements of virtual sensors from physical sensors. However, current ISK methods rely on conventional message-passing mechanisms and network architectures, without effectively extracting spatio-temporal features of physical sensors and focusing on representing virtual sensors. Additionally, existing graph construction methods face issues of sparse and noisy connections, destroying ISK performance. To address these issues, we propose DarkFarseer, a novel ISK framework with three key components. First, we propose the Neighbor Hidden Style Enhancement module with a style transfer strategy to enhance the representation of virtual nodes in a temporal-then-spatial manner to better extract the spatial relationships between physical and virtual nodes. Second, we propose Virtual-Component Contrastive Learning, which aims to enrich the node representation by establishing the association between the patterns of virtual nodes and the regional patterns within graph components. Lastly, we design a Similarity-Based Graph Denoising Strategy, which reduces the connectivity strength of noisy connections around virtual nodes and their neighbors based on their temporal information and regional spatial patterns. Extensive experiments demonstrate that DarkFarseer significantly outperforms existing ISK methods.