In recent years, contrastive learning has achieved state-of-the-art performance in the territory of self-supervised representation learning. Many previous works have attempted to provide the theoretical understanding underlying the success of contrastive learning. Almost all of them rely on a default assumption, i.e., the label consistency assumption, which may not hold in practice (the probability of failure is called labeling error) due to the strength and randomness of common augmentation strategies, such as random resized crop (RRC). This paper investigates the theoretical impact of labeling error on the downstream classification performance of contrastive learning. We first reveal several significant negative impacts of labeling error on downstream classification risk. To mitigate these impacts, data dimensionality reduction method (e.g., singular value decomposition, SVD) is applied on original data to reduce false positive samples, and establish both theoretical and empirical evaluations. Moreover, it is also found that SVD acts as a double-edged sword, which may lead to the deterioration of downstream classification accuracy due to the reduced connectivity of the augmentation graph. Based on the above observations, we give the augmentation suggestion that we should use some moderate embedding dimension (such as $512, 1024$ in our experiments), data inflation, weak augmentation, and SVD to ensure large graph connectivity and small labeling error to improve model performance.

In the context of the health monitoring for the next generation of reusable space launchers, we outline a first step toward developing an onboard fault detection and diagnostic capability for the electrical system that controls the engine valves. Unlike existing approaches in the literature, our solution is designed to meet a broader range of key requirements. This includes estimating confidence levels for predictions, detecting out-of-distribution (OOD) cases, and controlling false alarms. The proposed solution is based on a temporal convolutional autoencoder to automatically extract low-dimensional features from raw sensor data. Fault detection and diagnosis are respectively carried out using a binary and a multiclass classifier trained on the autoencoder latent and residual spaces. The classifiers are histogram-based gradient boosting models calibrated to output probabilities that can be interpreted as confidence levels. A relatively simple technique, based on inductive conformal anomaly detection, is used to identify OOD data. We leverage other simple yet effective techniques, such as cumulative sum control chart (CUSUM) to limit the false alarms, and threshold moving to address class imbalance in fault detection. The proposed framework is highly configurable and has been evaluated on simulated data, covering both nominal and anomalous operational scenarios. The results indicate that our solution is a promising first step, though testing with real data will be necessary to ensure that it achieves the required maturity level for operational use.

Pigmented skin lesions represent localized areas of increased melanin and can indicate serious conditions like melanoma, a major contributor to skin cancer mortality. The MedMNIST v2 dataset, inspired by MNIST, was recently introduced to advance research in biomedical imaging and includes DermaMNIST, a dataset for classifying pigmented lesions based on the HAM10000 dataset. This study assesses ResNet-50 and EfficientNetV2L models for multi-class classification using DermaMNIST, employing transfer learning and various layer configurations. One configuration achieves results that match or surpass existing methods. This study suggests that convolutional neural networks (CNNs) can drive progress in biomedical image analysis, significantly enhancing diagnostic accuracy.

With the continuous evolution of China's multi-level capital market, the National Equities Exchange and Quotations (NEEQ), also known as the "New Third Board," has become a critical financing platform for small and medium-sized enterprises (SMEs). However, due to their limited scale and financial resilience, many NEEQ-listed companies face elevated risks of financial distress. To address this issue, we propose a multi-channel deep learning framework that integrates structured financial indicators, textual disclosures, and enterprise relationship data for comprehensive financial risk prediction. Specifically, we design a Triple-Channel Graph Isomorphism Network (GIN) that processes numeric, textual, and graph-based inputs separately. These modality-specific representations are fused using an attention-based mechanism followed by a gating unit to enhance robustness and prediction accuracy. Experimental results on data from 7,731 real-world NEEQ companies demonstrate that our model significantly outperforms traditional machine learning methods and single-modality baselines in terms of AUC, Precision, Recall, and F1 Score. This work provides theoretical and practical insights into risk modeling for SMEs and offers a data-driven tool to support financial regulators and investors.

Developing precise and computationally efficient traffic accident anticipation system is crucial for contemporary autonomous driving technologies, enabling timely intervention and loss prevention. In this paper, we propose an accident anticipation framework employing a dual-branch architecture that effectively integrates visual information from dashcam videos with structured textual data derived from accident reports. Furthermore, we introduce a feature aggregation method that facilitates seamless integration of multimodal inputs through large models (GPT-4o, Long-CLIP), complemented by targeted prompt engineering strategies to produce actionable feedback and standardized accident archives. Comprehensive evaluations conducted on benchmark datasets (DAD, CCD, and A3D) validate the superior predictive accuracy, enhanced responsiveness, reduced computational overhead, and improved interpretability of our approach, thus establishing a new benchmark for state-of-the-art performance in traffic accident anticipation.

Early screening for these diseases is essential, yet access to medical care remains limited in low-and middle-income countries as well as in resource-limited settings. We develop InSight, an AI-based app that combines patient metadata with fundus images for accurate diagnosis of five common eye diseases to improve accessibility of screenings. Methods InSight features a three-stage pipeline: real-time image quality assessment, disease diagnosis model, and a DR grading model to assess severity. Our disease diagnosis model incorporates three key innovations: (a) Multimodal fusion technique (MetaFusion) combining clinical metadata and images; (b) Pretraining method leveraging supervised and self-supervised loss functions; and (c) Multitask model to simultaneously predict 5 diseases. We make use of BRSET (lab-captured images) and mBRSET (smartphone-captured images) datasets, both of which also contain clinical metadata for model training/evaluation. Results Trained on a dataset of BRSET and mBRSET images, the image quality checker achieves near-100% accuracy in filtering out low-quality fundus images. The multimodal pretrained disease diagnosis model outperforms models using only images by 6% in balanced accuracy for BRSET and 4% for mBRSET. Conclusions The InSight pipeline demonstrates robustness across varied image conditions and has high diagnostic accuracy across all five diseases, generalizing to both smartphone and lab captured images. The multitask model contributes to the lightweight nature of the pipeline, making it five times computationally efficient compared to having five individual models corresponding to each disease.

Test-time scaling aims to improve language model performance by leveraging additional compute during inference. While many works have empirically studied techniques like Best-of-N (BoN) and rejection sampling that make use of a verifier to enable test-time scaling, there is little theoretical understanding of how verifier imperfection affects performance. In this work, we address this gap. Specifically, we prove how instance-level accuracy of these methods is precisely characterized by the geometry of the verifier's ROC curve. Interestingly, while scaling is determined by the local geometry of the ROC curve for rejection sampling, it depends on global properties of the ROC curve for BoN. As a consequence when the ROC curve is unknown, it is impossible to extrapolate the performance of rejection sampling based on the low-compute regime. Furthermore, while rejection sampling outperforms BoN for fixed compute, in the infinite-compute limit both methods converge to the same level of accuracy, determined by the slope of the ROC curve near the origin. Our theoretical results are confirmed by experiments on GSM8K using different versions of Llama and Qwen to generate and verify solutions.

Composed Image Retrieval (CIR) retrieves relevant images based on a reference image and accompanying text describing desired modifications. However, existing CIR methods only focus on retrieving the target image and disregard the relevance of other images. This limitation arises because most methods employing contrastive learning-which treats the target image as positive and all other images in the batch as negatives-can inadvertently include false negatives. This may result in retrieving irrelevant images, reducing user satisfaction even when the target image is retrieved. To address this issue, we propose Query-Relevant Retrieval through Hard Negative Sampling (QuRe), which optimizes a reward model objective to reduce false negatives. Additionally, we introduce a hard negative sampling strategy that selects images positioned between two steep drops in relevance scores following the target image, to effectively filter false negatives. In order to evaluate CIR models on their alignment with human satisfaction, we create Human-Preference FashionIQ (HP-FashionIQ), a new dataset that explicitly captures user preferences beyond target retrieval. Extensive experiments demonstrate that QuRe achieves state-of-the-art performance on FashionIQ and CIRR datasets while exhibiting the strongest alignment with human preferences on the HP-FashionIQ dataset. The source code is available at https://github.com/jackwaky/QuRe.

We explore an ASR-free method for isolated word reading assessment in low-resource settings. Our few-shot approach compares input child speech to a small set of adult-provided reference templates. Inputs and templates are encoded using intermediate layers from large self-supervised learned (SSL) models. Using an Afrikaans child speech benchmark, we investigate design options such as discretising SSL features and barycentre averaging of the templates. Idealised experiments show reasonable performance for adults, but a substantial drop for child speech input, even with child templates. Despite the success of employing SSL representations in low-resource speech tasks, our work highlights the limitations of SSL representations for processing child data when used in a few-shot classification system.

Advancements in wireless and mobile technologies, including 5G advanced and the envisioned 6G, are driving exponential growth in wireless devices. However, this rapid expansion exacerbates spectrum scarcity, posing a critical challenge. Dynamic spectrum allocation (DSA)--which relies on sensing and dynamically sharing spectrum--has emerged as an essential solution to address this issue. While machine learning (ML) models hold significant potential for improving spectrum sensing, their adoption in centralized ML-based DSA systems is limited by privacy concerns, bandwidth constraints, and regulatory challenges. To overcome these limitations, distributed ML-based approaches such as Federated Learning (FL) offer promising alternatives. This work addresses two key challenges in FL-based spectrum sensing (FLSS). First, the scarcity of labeled data for training FL models in practical spectrum sensing scenarios is tackled with a semi-supervised FL approach, combined with energy detection, enabling model training on unlabeled datasets. Second, we examine the security vulnerabilities of FLSS, focusing on the impact of data poisoning attacks. Our analysis highlights the shortcomings of existing majority-based defenses in countering such attacks. To address these vulnerabilities, we propose a novel defense mechanism inspired by vaccination, which effectively mitigates data poisoning attacks without relying on majority-based assumptions. Extensive experiments on both synthetic and real-world datasets validate our solutions, demonstrating that FLSS can achieve near-perfect accuracy on unlabeled datasets and maintain Byzantine robustness against both targeted and untargeted data poisoning attacks, even when a significant proportion of participants are malicious.