Intracerebral hemorrhage (ICH) is a type of stroke that, although less common than ischemic stroke, has higher mortality and disability rates. Among its complications, hematoma expansion (HE) is a major determinant of poor outcomes. Accurate segmentation of ICH lesions and timely prediction of HE are crucial for effective intervention. Existing methods mainly rely on the clinical experience of neurosurgeons and typically treat these two tasks independently, fundamentally overlooking the powerful, synergistic information shared between a hematoma’s morphology and its propensity for expansion, leading to inaccurate HE prediction. Furthermore, conventional deep learning models often struggle to capture irregular hematoma boundaries and fail to reconcile the feature conflict between pixel-level segmentation and image-level prediction. To comprehensively address these limitations, we propose a dual-stream guided conditional diffusion framework called ICHSC-Diff. It facilitates a symbiotic interplay by leveraging the intrinsic generative capabilities of diffusion, coupling generative features with explicit radiological data to form a multi-modal representation for robust early HE prediction. In particular, we design a Task-aware Conditioning Module that utilizes a Style Preservation Refinement Module (SPRM) for adaptive preliminary denoising, dynamically adjusting the filters to match different hematomas. Thereafter, a Synergistic Representation Generator captures multi-scale local features by separating noise from the signal, preserving edge and texture details to guide the generation of accurate hematoma masks. Moreover, we develop a Multi-Scale Hybrid Fusion (MSHF) module that uses an attention-driven mechanism to reconstruct and fuse radiological features with generative features. This process reduces redundant information unrelated to hematoma characteristics, leading to robust early HE prediction. We further accelerate the inference process using the DPM-Solver strategy, enhancing the model’s efficiency. Extensive experiments on real-world intracerebral hemorrhage datasets demonstrate that our method excels in overall prediction performance, which in turn enhances ICH segmentation performance. The source codes of our framework are publicly available at https://github.com/tarkmfcv/ICHSC-Diff .

Cine-MRI establishes transformative potential in postoperative swallowing assessment of tongue cancer patients through highly resolved dynamic images. Yet current qualitative methods are based on isolated anatomical landmarks, demanding specialized neural architectures and labor-intensive manual annotations. To overcome these limitations, we propose a segmentation anything model (SAM) grounded quantitative framework that enables automated and precise swallowing function assessment. Firstly, we leverage the teacher-student paradigm by fine-tuning SAM to create an anatomical motion expert from its diverse visual priors. Secondly, several lightweight state-of-the-art student networks are optimized and pre-trained to adapt to the distilled knowledge from these priors. Experiments validate that this SAM-based teacher supervision exhibits wide compatibility across diverse architectures, with U-Net achieving the most substantial improvements$\boldsymbol{(} \mathbf{+ 6. 1 2 \%}$Dice,$\boldsymbol{+ 6. 6 6 \%}$clDice), while maintaining computational efficiency for plug-and-play integration. Moreover, its convenient deployment can make clinical assessment of such functions more straightforward and reliable.

Understanding the conformational landscape of proteins is essential for elucidating protein function and facilitating drug design. However, existing protein conformation benchmarks fail to capture the full energy landscape, limiting their ability to evaluate the diversity and physical plausibility of AI-generated structures. We introduce ProteinConformers, a large-scale benchmark dataset comprising over 381,000 physically realistic conformations for 87 CASP targets. These were derived from more than 40,000 structural decoys via extensive all-atom molecular dynamics simulations totaling over 6 million CPU hours. Using this dataset, we propose novel metrics to evaluate conformational diversity and plausibility, and systematically benchmark six protein conformation generative models. Our results highlight that leveraging large-scale protein sequence data can enhance a model’s ability to explore conformational space, potentially reducing reliance on MD-derived data. Additionally, we find that PDB and MD datasets influence model performance differently, current models perform well on inter-atomic distance prediction but struggle with inter-residue orientation generation. Overall, our dataset, evaluation metrics, and benchmarking results provide the first comprehensive foundation for assessing generative models in protein conformational modeling. Dataset and instructions are available at https://huggingface.co/ datasets/Jim990908/ProteinConformers/tree/main. Codes are stored at https://github.com/auroua/ProteinConformers. An interactive website locates at https://zhanggroup.org/ProteinConformers.

In e-commerce recommender systems, providing product suggestions to customers that are often bought together, which is called “complementary recommendation,” not only improves customer experience but also boosts business impact. However, in practice, it is highly challenging to efficiently extract the complementary relations between the items due to noisy and low coverage of the co-purchased records in transaction datasets. To address these challenges, graph neural networks (GNN) have been increasingly adopted in complementary item recommendations thanks to their capabilities in integrating side-information and topological structures to extract these complex item relationships. However, most existing GNN-based methods fall short in learning better product complementary representation since they often utilize a simple one-to-one product-to-vector mapping strategy, which fails to describe the transitive logic of complementary items. To overcome this challenge, we propose a new GNN model called transitivity-encoded graph attention networks (TransGAT). To our knowledge, TransGAT is the first method that extends representation space by encoding the behavioral direction into embedding space in GNN and enabling mutual relationship extraction between complementary items. In order to better extract customer's intrinsic behavioral information, we further adopt the substitute information as the guidance by jointly learning complements and substitutes graphs and coupling them together. Moreover, several self-supervised data augmentation strategies are incorporated in our approach. Through evaluations on three real-world datasets, TransGAT consistently surpasses contemporary benchmarks, showcasing its prowess in complementary item recommendations.

A key challenge of big data analytics is how to collect a large volume of (labeled) data. Crowdsourcing aims to address this challenge via aggregating and estimating high-quality data (e.g., sentiment label for text) from pervasive clients/users. Existing studies on crowdsourcing focus on designing new methods to improve the aggregated data quality from unreliable/noisy clients. However, the security aspects of such crowdsourcing systems remain under-explored to date. We aim to bridge this gap in this work. Specifically, we show that crowdsourcing is vulnerable to data poisoning attacks, in which malicious clients provide carefully crafted data to corrupt the aggregated data. We formulate our proposed data poisoning attacks as an optimization problem that maximizes the error of the aggregated data. Our evaluation results on one synthetic and two real-world benchmark datasets demonstrate that the proposed attacks can substantially increase the estimation errors of the aggregated data. We also propose two defenses to reduce the impact of malicious clients. Our empirical results show that the proposed defenses can substantially reduce the estimation errors of the data poisoning attacks.

In this paper, we propose a novel visual tracking algorithm combining appearance feature and spatial information. These two aspects extract the constant elements of the target object. To represent information of an object, we utilize a block-local model and a structured visual histogram learned from an object's spatial information, which can adapt to changes in the appearance of the target and environment of the background. During tracking, the proposed algorithm introduces an occlusion model to prevent the environment change from affecting the target's posterior probability. Besides, the online updating is based on incremental algorithms for principal component analysis and the renewal of a spatial histogram queue. Experimental results show that our method outperforms relative trackers.