Current RGB-D salient object detection (SOD) models are plagued by issues, including excessive model parameters and high computational complexity. These drawbacks impede the model's efficient deployment and constrain the enhancement of model performance. This paper introduces an efficient and lightweight cross-modal feature cross-fusion network, termed CMFNet. In particular, we design an efficient model utilizing MobileViT as the dual-stream backbone network, thereby significantly reducing computational complexity while maintaining robust feature extraction capabilities. Firstly, we propose a Cross-Fusion Module (CFM) designed to integrate multi-scale semantic information from RGB and depth features dynamically. Furthermore, we design a Lightweight-Mixture-of-Experts Module (L-MoE) for multi-modal features, which enhances the representational capacity of fused features at various levels by employing a dynamic routing mechanism and balanced constraint strategy to allocate appropriate expert processing units to features at different scales. Additionally, we design a Multi-scale Feature Refinement Module (MSFR) that captures multi-scale context through a combination of channel-spatial attention mechanisms and depthwise separable convolution, and gradually eliminates feature distribution differences between modalities using a dual-path residual learning strategy. Abundant experimental findings verify that the proposed CMFNet outperforms the 25 existing State-of-the-art (SOTA) methods.

As diffusion models are deployed in real-world settings and their performance driven by training data, appraising the contribution of data contributors is crucial to creating incentives for sharing quality data and to implementing policies for data compensation. Depending on the use case, model performance corresponds to various global properties of the distribution learned by a diffusion model (e.g., overall aesthetic quality). Hence, here we address the problem of attributing global properties of diffusion models to data contributors. The Shapley value provides a principled approach to valuation by uniquely satisfying game-theoretic axioms of fairness. However, estimating Shapley values for diffusion models is computationally impractical because it requires retraining and rerunning inference on many subsets of data contributors. We introduce a method to efficiently retrain and rerun inference for Shapley value estimation, by leveraging model pruning and fine-tuning. We evaluate the utility of our method with three use cases: (i) image quality for a DDPM trained on a CIFAR dataset, (ii) demographic diversity for an LDM trained on CelebA-HQ, and (iii) aesthetic quality for a Stable Diffusion model LoRA-finetuned on Post-Impressionist artworks. Our results empirically demonstrate that our framework can identify important data contributors across global properties, outperforming existing attribution methods for diffusion models.

The inconsistency between classification and localization brings a challenge in object detection. Fully supervised object detection (FSOD) benefits from bounding-box regression networks to alleviate that, which is absent in weakly supervised object detection (WSOD). Consequently, there is a significant performance gap between the two paradigms. To bridge the performance and technical gaps between WSOD and FSOD, this paper proposes a novel weakly supervised object detection framework based on classification-localization consistency. We propose Max Score Pooling (MSP), which compels features relevant to classification to align with features relevant to localization, thereby achieving consistency between classification and localization. Additionally, we propose a Proposal Fusion Mechanism (PFM) to generate pseudo-supervision for training the bounding box regression network, further reducing the impact of classification-localization inconsistency. Extensive experiments are conducted on PASCAL VOC 2007, PASCAL VOC 2012 and MS COCO 2017 datasets, demonstrating our framework’s superior performance.

Recent advancements in radiology report generation have utilized deep neural networks such as CNNs and Transformers, achieving notable improvements in generating accurate and detailed reports. However, their practical adoption is hindered by the challenge of balancing global dependency modeling with computational efficiency. The state space model, particularly its enhanced variant Mamba, offers promising linear-complexity solutions for long-range dependency modeling. Despite its strengths, Mamba’s fixed positional encoding limits its ability to effectively capture complex spatial dependencies. To address this gap, we propose RRG-Mamba, an advanced framework for efficient radiology report generation. Within the RRGMamba, we enhance the vanilla Mamba by integrating rotary position encoding (RoPE), enabling dynamic modeling of relative positional information in visual feature sequences. Furthermore, we design a global dependency learning module to optimize long-range visual feature sequence modeling. Extensive experiments on publicly available datasets, including IU X-Ray and MIMIC-CXR, demonstrate that RRG-Mamba achieves a 3.7% improvement in BLEU-4 score over existing models, along with significant gains in computational and memory efficiency. Our code is available at https://github.com/Eleanorhxd/RRG-Mamba.

High-content screening (HCS) assays based on high-throughput microscopy techniques such as Cell Painting have enabled the interrogation of cells' morphological responses to perturbations at an unprecedented scale. The collection of such data promises to facilitate a better understanding of the relationships between different perturbations and their effects on cellular state. Towards achieving this goal, recent advances in cross-modal contrastive learning could, in theory, be leveraged to learn a unified latent space that aligns perturbations with their corresponding morphological effects. However, the application of such methods to HCS data is not straightforward due to substantial differences in the semantics of Cell Painting images compared to natural images, and the difficulty of representing different classes of perturbations (e.g. small molecule vs CRISPR gene knockout) in a single latent space. In response to these challenges, here we introduce CellCLIP, a cross-modal contrastive learning framework for HCS data. CellCLIP leverages pre-trained image encoders coupled with a novel channel encoding scheme to better capture relationships between different microscopy channels in image embeddings, along with natural language encoders for representing perturbations. Our framework outperforms current open-source models, demonstrating the best performance in both cross-modal retrieval and biologically meaningful downstream tasks while also achieving significant reductions in computation time. Code for our reproducing our experiments is available at https://github.com/suinleelab/CellCLIP.

Abstract Large pre‐trained models (PTMs) have shown their powerful ability in multiple natural language processing tasks. However, using them in practical application remains a challenge due to the significant computational cost and memory requirements. In order to achieve the balance of computational cost and accuracy, MLP architecture can be used as an alternative to the self‐attention module, such as pNLP‐Mixer and Hyper‐Mixer. Experiments indicate that, MLP‐based models can attain competitive performance with low cost. They maintain the balance of computation cost and accuracy successfully, yet, this is at the expense of not being able to capture short‐range dependencies. In this paper, a novel MLP‐based model, termed TS‐Mixer, is proposed which can capture local dependencies by shifting operation. Compared with other MLP‐based models, the parameters of TS‐Mixer are decoupled from the sequence length, hence it has a smaller model size in long sequence tasks. In addition, TS‐Mixer has linear computational complexity, therefore it can be used as a lightweight alternative to the self‐attention model. Experiments show that the TS‐Mixer outperforms other MLP‐based models, which achieves higher accuracy with fewer parameters in multiple downstream tasks. Notably, compared with pre‐trained models, TS‐Mixer can reach more than 90% of their accuracy with 1% or even one thousandth of the parameters (0.174 ~ 1.2 M). These results demonstrate that TS‐Mixer can achieve a better balance between the computing resources and accuracy. Code is available at: https://github.com/wyl-privacy-project/TS-Mixer .

Synthetic lethality (SL) has emerged as a promising strategy in cancer medicine. However, complex biomolecular interactions make wet lab methods time-consuming and expensive. Machine learning methods have gained widespread adoption for SL prediction in recent years. Although these methods have demonstrated particular effectiveness, they suffer from weak interpretability, making it difficult for users to understand the specific reasoning processes of the models. Also, they typically focus on a simple gene pair, thus struggling with more meaningful reasoning tasks involving other medical factors as in real life. To address these gaps, we propose an explainable multi-hop reasoning model EFOL-SL based on first-order logic queries. We first construct query graphs with triplet transformations for different tasks. Node embeddings are then fed into a sparse Transformer encoder and a visualized graph attention decoder to generate comprehensive multi-hop logical reasoning chains. By masking nodes in intermediate reasoning steps, our model can explicitly predict each node, allowing observation of its exact reasoning process. Additionally, we conduct extensive experiments on two widely used benchmarks with complex SL prediction tasks involving diverse medical entities. Evaluations demonstrate superior performance of our model over state-of-the-art methods on various tasks. Notably, EFOL-SL provides specific multi-hop logical reasoning chains behind its predictions, offering meaningful insights into the model's reasoning process.

Drugtarget interaction (DTI) prediction is a pivotal task in the realm of drug discovery. As the volume of biological data has increased rapidly, the integration of multiple data sources to increase prediction accuracy has become increasingly important. However, few methods exploit the heterogeneous information network in the drugtarget network by integrating multisource information to address the task of drugtarget interaction prediction. In this paper, we propose a multitask DTI prediction model, HIN-MTDTI, which is grounded in heterogeneous information networks (HINs). The model employs drugtarget interaction network, drugdrug similarity network and targettarget similarity network as inputs to construct a heterogeneous information network. Moreover, we apply a graph convolutional network (GCN) on the HIN to learn the representations of drugs and targets. To augment the performance further, we integrate a bilinear attention network to capture local drugtarget interaction information fully. The experimental results on several benchmark datasets demonstrate that HIN-MTDTI outperforms state-of-the-art methods for DTI prediction, confirming the effectiveness of our method.

The traditional drug development process requires a significant investment in workforce and financial resources. Drug repositioning as an efficient alternative has attracted much attention during the last few years. Despite the wide application and success of the method, there are still many shortcomings in the existing model. For example, sparse datasets will seriously affect the existing methods' performance. Additionally, these methods do not pay attention to the noise in datasets. In response to the above defects, we propose a semantic-enriched augmented graph contrastive learning with an adaptive denoising method, called SGCD. This method enhances data from the perspective of the embedding layer, deeply mines potential neighborhood relation-ships in semantic space, and combines similar drugs in the semantic neighborhoods into prototype comparison targets, thus effectively mitigating the impact of data sparsity on the model. Moreover, to enhance the model's robustness to noisy data, we use the adaptive denoising method, which can effectively identify noisy data in the training process. Exhaustive experiments on multiple real datasets show the effectiveness of the proposed model.

MOTIVATION: Drug recommendation aims to allocate safe and effective drug combinations based on the patient's health status from electronic health records, which is crucial to assist clinical physicians in making decisions. However, the existing drug recommendation works face two key challenges: (i) difficulty in fully representing the patient's health status leads to biased drug representation; (ii) only focusing on diagnostic representations of multiple visits, neglecting the modeling of patient drug history. RESULTS: To address the above limitations, we propose a multi-view gating retrieval network (MGRN) for robust drug recommendation. We design visit-, sequence-, and token-level views to provide different perspectives on the interaction between patients and drugs, obtaining a more comprehensive representation of drugs. Moreover, we develop a gating drug retrieval module to capture critical drug information from multiple visits, which can assist in recommending more reasonable drug combinations for the current visit. When evaluated on publicly real-world MIMIC-III and MIMIC-IV datasets, the proposed MGRN establishes a new benchmark performance, particularly achieving improvements of 1.36%, 1.71%, 1.21% and 2.12%, 2.36%, 1.81% in Jaccard, PRAUC, and F1-score, respectively, compared to state-of-the-art models. AVAILABILITY AND IMPLEMENTATION: The code is available at: https://github.com/kyosen258/MGRN.git.

Human parsing has attracted considerable research interest due to its broad potential applications in the computer vision community. In this paper, we explore several useful properties, including high-resolution representation, auxiliary guidance, and model robustness, which collectively contribute to a novel method for accurate human parsing in both simple and complex scenes. Starting from simple scenes: we propose the boundary-aware hybrid resolution network (BHRN), an advanced human parsing network. BHRN utilizes deconvolutional layers and multi-scale supervision to generate rich high-resolution representations. Additionally, it includes an edge perceiving branch designed to enhance the fineness of part boundaries. Building on BHRN, we construct a dual-task mutual learning (DTML) framework. It not only provides implicit guidance to assist the parser by incorporating boundary features, but also explicitly maintains the high-order consistency between the parsing prediction and the ground truth. Toward complex scenes: we develop a domain transform method to enhance the model robustness. By transforming the input space from the spatial domain to the polar harmonic Fourier moment domain, the mapping relationship to the output semantic space is highly stable. This transformation yields robust representations for both clean and corrupted data. When evaluated on standard benchmark datasets, our method achieves superior performance compared to state-of-the-art human parsing methods. Furthermore, our domain transform strategy significantly improves the robustness of DTML dramatically in most complex scenes.

People who have dementia show a decline in their speech abilities. In speech-based dementia detection, the difficulty has remained the representation of an individual's sequential temporal variation of speech is related to dementia symptoms with fix-length features. In this paper, a novel feature extrac- tion method is proposed for extracting fix-length features from unfixed-length audio recordings for dementia detection. When diagnosing dementia, an automatic speech recognition (ASR) system is necessary for extracting linguistic information when constructing an automatic dementia detection system. This paper uses wav2vec2.0, a self-supervised end-to-end ASR system, to achieve such a goal. Similar to the pipeline ASR system, which has been used for extracting the sequential speak-and-pause patterns related to dementia using estimated time alignment information, we propose using character-level transcripts to extract speak-and-pause patterns. Path signature technology, which can represent a sequential feature with a trajectory in the un-parameterised path space, is proposed to describe speak- and-pause patterns embedded in character-level transcripts into character path signatures. Similarly, the variable-length embed- ding matrices extracted from wav2vec2.0's contextual layers are also represented with their acoustic path signatures. The exper- iments are designed based on three publicly available datasets: DementiaBank, ADReSS and ADReSSo. The results show that: (1). The distinguished information embedded in the character path signature is visualised for dementia detection; (2). The acoustic path signature and character path signature individually can show superior performance on all three publicly available datasets. (3). Combining the character path signature with the acoustic path signature can considerably increase performance over the ADReSSo dataset.

Despite significant progress in medical image analysis using deep learning, predicting incidental prostate cancer (iPCa) remains challenging due to subtle differences in multiparametric magnetic resonance imaging (mpMRI) and a lower incidence rate. To address these challenges, we propose iPCa-Former, a transformer-based framework designed to enhance iPCa prediction within prostate mpMRI slices. Firstly, built on an encoder-decoder architecture, our iPCa-Former facilitates the simultaneous optimization of two tasks through mutual learning: prostate transition zone segmentation and iPCa prediction. Secondly, we introduce a joint optimization function that combines focal loss and boundary-based mutual information (BMI) loss, effectively addressing the imbalance of positive and negative samples in classification and the challenge posed by a small proportion of the foreground region in segmentation. Moreover, we construct an iPCa mpMRI dataset comprising 10,276 prostate mpMRI slices from 485 patients clinically diagnosed with benign prostatic hyperplasia, however, 27 out of these patients are identified as iPCa. When evaluated on this benchmark dataset, our iPCa-Former outperforms state-of-the-art methods, demonstrating the superior performance of our approach.

Traditional drug development is often high-risk and time-consuming. A promising alternative is to reuse or relocate approved drugs. Recently, some methods based on graph representation learning have started to be used for drug repositioning. These models learn the low dimensional embeddings of drug and disease nodes from the drug-disease interaction network to predict the potential association between drugs and diseases. However, these methods have strict requirements for the dataset, and if the dataset is sparse, the performance of these methods will be severely affected. At the same time, these methods have poor robustness to noise in the dataset. In response to the above challenges, we propose a drug repositioning model based on self-supervised graph learning with adptive denoising, called SADR. SADR uses data augmentation and contrastive learning strategies to learn feature representations of nodes, which can effectively solve the problems caused by sparse datasets. SADR includes an adaptive denoising training (ADT) component that can effectively identify noisy data during the training process and remove the impact of noise on the model. We have conducted comprehensive experiments on three datasets and have achieved better prediction accuracy compared to multiple baseline models. At the same time, we propose the top 10 new predictive approved drugs for treating two diseases. This demonstrates the ability of our model to identify potential drug candidates for disease indications.

Deep learning based object detection methods have made significant progress in recent years. However, these methods often suffer from a substantial performance drop when domain shifts occur, making it difficult to generalize a source domain trained object detector to a new target domain. To address this problem, we propose an Online Meta Learning Framework (OMLF) for unsupervised domain adaptive object detection. In our proposed framework, we adopt the Polar Harmonic Fourier Moment (PHFM) to generate target-like intermediate data. The purpose is to construct a two-pair framework that learns meta knowledge (i.e. model initial parameters) from the pair of “source-to-intermediate” to assist another pair of “intermediate-to-target”. Moreover, the optimizing process requires a heavy computational load due to triggering higher-order gradients. To alleviate this problem, we introduce a shortest-path update strategy that accelerates optimization. When evaluated on several benchmark adaptation scenarios (i.e. normal-to-foggy weather, cross cameras, synthetic-to-real, and real-to-artistic), our OMLF achieves state-of-the-art results, demonstrating its effectiveness.

Drone-based vehicle detection is a critical task within intelligent transportation systems. The existing methods that rely solely on single visible or infrared modalities often struggle to achieve both precise and robust detection. Effectively integrating cross-modal information to assist in vehicle detection remains a significant challenge. In this article, we propose a mask-guided Mamba fusion (MGMF) method for visible-infrared vehicle detection in aerial scenes. The proposed MGMF framework consists of two key components: the masked regularization constraint module (MRCM) and the state-space fusion module (SSFM). First, in MAEM, we use candidate regions from one modality to cover corresponding regions of intermediate-level features from another modality, while a regularization constraint extracts cross-modal guidance. This design allows cross-modal features focused on vehicle areas to be extracted from both modalities for fusion. Second, in SSFM, we propose mapping cross-modal features into a shared hidden state for interaction. This reduces disparities between the cross-modal features and enhances the representation, enabling better perception of intermodal correlations. When evaluated on the DroneVehicle dataset, our MGMF achieves an 80.24% with respect to mAP, establishing a new benchmark for state-of-the-art performance. Ablation studies further demonstrate the effectiveness of our MAEM and SSFM in enhancing visible-infrared fusion for vehicle detection.

Predicting drug-drug interaction (DDI) plays a crucial role in drug recommendation and discovery. However, wet lab methods are prohibitively expensive and time-consuming due to drug interactions. In recent years, deep learning methods have gained widespread use in drug reasoning. Although these methods have demonstrated effectiveness, they can only predict the interaction between a drug pair and do not contain any other information. However, DDI is greatly affected by various other biomedical factors (such as the dose of the drug). As a result, it is challenging to apply them to more complex and meaningful reasoning tasks. Therefore, this study regards DDI as a link prediction problem on knowledge graphs and proposes a DDI prediction model based on Cross-Transformer and Graph Convolutional Networks (GCNs) in first-order logical query form, TransFOL. In the model, a biomedical query graph is first built to learn the embedding representation. Subsequently, an enhancement module is designed to aggregate the semantics of entities and relations. Cross-Transformer is used for encoding to obtain semantic information between nodes, and GCN is used to gather neighbour information further and predict inference results. To evaluate the performance of TransFOL on common DDI tasks, we conduct experiments on two benchmark datasets. The experimental results indicate that our model outperforms state-of-the-art methods on traditional DDI tasks. Additionally, we introduce different biomedical information in the other two experiments to make the settings more realistic. Experimental results verify the strong drug reasoning ability and generalization of TransFOL in complex settings.

Feature selection helps reduce data acquisition costs in ML, but the standard approach is to train models with static feature subsets. Here, we consider the dynamic feature selection (DFS) problem where a model sequentially queries features based on the presently available information. DFS is often addressed with reinforcement learning, but we explore a simpler approach of greedily selecting features based on their conditional mutual information. This method is theoretically appealing but requires oracle access to the data distribution, so we develop a learning approach based on amortized optimization. The proposed method is shown to recover the greedy policy when trained to optimality, and it outperforms numerous existing feature selection methods in our experiments, thus validating it as a simple but powerful approach for this problem.

Medicine recommendation assists physicians in automatically providing medicine combinations, which is critical in health care. Existing efforts focus on making medicine recommendations based on the patient’s electronic health record(EHR). However, they ignore external medicine knowledge and are vulnerable to the missing EHR. In this paper, a knowledge-enhanced dual graph neural network (KDGN) is proposed to recommend medicine sets. KDGN combines diagnosis-level and procedure-level attention mechanisms to encode multiple types of medical codes. In order to mine medicine from medical knowledge, KDGN further designs a dual-graph neural network, which constructs a medicine co-occurrence graph and molecular connection graph, and retrieves potential therapeutic drugs. Furthermore, during the training phase, we introduce the automatic correction loss based on maximum likelihood estimation to mitigate the impact of missing EHR and enhance the robustness of KDGN. We evaluate the proposed model on the public MIMIC-III dataset, and experimental results show that KDGN outperforms the state-of-the-art model in 4 out of 5 evaluation metrics. Our dataset and code are available at: https://github.com/Benjamin-cell/KDGN.

The identification of protein complexes can help understand cellular organization principles and the mechanism of biological evolution. In recent years, researchers have proposed numerous computational methods to identify protein complexes through their interaction networks. Most of these methods identify protein complexes based on the topological structure of the PPI network. However, the topological structure contained in the PPI network is very complicated, and the applicability of advanced representation learning methods has not been researched in depth. This paper proposes a contrastive self-supervised representation learning method to identify protein complexes. Our method uses a mix-hop aggregator based on graph neural network (GNN) to capture high-order interaction in the PPI network and leverage a contrastive self-supervised method to train our model without introducing protein labels. Then, we get the vector representation for each protein and construct a weighted PPI network based on the vector representation similarity. Finally, we apply clustering aggregation to identify protein complexes based on a weighted PPI network. In order to access our method, different PPI networks, DIP, Kroganl4k and Biogrid, are used as datasets. By comparing the competing methods including COACH, CMC, MCODE, ClusterONE, GANE and COAN, experimental results show that our method outperforms classic and state-of-the-art methods.

BACKGROUND: Clinical notes are documents that contain detailed information about the health status of patients. Medical codes generally accompany them. However, the manual diagnosis is costly and error-prone. Moreover, large datasets in clinical diagnosis are susceptible to noise labels because of erroneous manual annotation. Therefore, machine learning has been utilized to perform automatic diagnoses. Previous state-of-the-art (SOTA) models used convolutional neural networks to build document representations for predicting medical codes. However, the clinical notes are usually long-tailed. Moreover, most models fail to deal with the noise during code allocation. Therefore, denoising mechanism and long-tailed classification are the keys to automated coding at scale. RESULTS: In this paper, a new joint learning model is proposed to extend our attention model for predicting medical codes from clinical notes. On the MIMIC-III-50 dataset, our model outperforms all the baselines and SOTA models in all quantitative metrics. On the MIMIC-III-full dataset, our model outperforms in the macro-F1, micro-F1, macro-AUC, and precision at eight compared to the most advanced models. In addition, after introducing the denoising mechanism, the convergence speed of the model becomes faster, and the loss of the model is reduced overall. CONCLUSIONS: The innovations of our model are threefold: firstly, the code-specific representation can be identified by adopted the self-attention mechanism and the label attention mechanism. Secondly, the performance of the long-tailed distributions can be boosted by introducing the joint learning mechanism. Thirdly, the denoising mechanism is suitable for reducing the noise effects in medical code prediction. Finally, we evaluate the effectiveness of our model on the widely-used MIMIC-III datasets and achieve new SOTA results.

In this paper, we focus on the Protein-Protein Interaction Article Classification (PPIAC) problem. In order to make better use of domain knowledge, we propose a Hierarchical Knowledge-aware Hybrid Neural Network (HKaHNN) model to classify PPI articles. Inspired by existing work, we introduce two kinds of knowledge embeddings and design a Hierarchical Knowledge-aware Attention (HKaATT) component which implements the interaction between the original token representations and external knowledge from the token-level and sentence-level respectively. In addition, in order to improve the anti-interference ability of the model, we adopt the adversarial training strategy. Our model achieves competitive performance on BioCreative II and BioCreative III corpora, with Fl-scores of 83.67% and 68.86%, respectively.

As the biomedical literature increases exponentially, biomedical named entity recognition (BNER) has become an important task in biomedical information extraction. In the previous studies based on deep learning, pretrained word embedding becomes an indispensable part of the neural network models, effectively improving their performance. However, the biomedical literature typically contains numerous polysemous and ambiguous words. Using fixed pretrained word representations is not appropriate. Therefore, this paper adopts the pretrained embeddings from language models (ELMo) to generate dynamic word embeddings according to context. In addition, in order to avoid the problem of insufficient training data in specific fields and introduce richer input representations, we propose a multitask learning multichannel bidirectional gated recurrent unit (BiGRU) model. Multiple feature representations (e.g., word-level, contextualized word-level, character-level) are, respectively, or collectively fed into the different channels. Manual participation and feature engineering can be avoided through automatic capturing features in BiGRU. In merge layer, multiple methods are designed to integrate the outputs of multichannel BiGRU. We combine BiGRU with the conditional random field (CRF) to address labels’ dependence in sequence labeling. Moreover, we introduce the auxiliary corpora with same entity types for the main corpora to be evaluated in multitask learning framework, then train our model on these separate corpora and share parameters with each other. Our model obtains promising results on the JNLPBA and NCBI-disease corpora, with F1-scores of 76.0% and 88.7%, respectively. The latter achieves the best performance among reported existing feature-based models.

Fundus images are one of the main methods for diagnosing eye diseases in modern medicine.The vascular segmentation of fundus images is an essential step in quantitative disease analysis.Based on the previous studies, we found that the category imbalance is one of the main reasons that restrict the improvement of segmentation accuracy.This paper presents a new method for supervised retinal vessel segmentation that can effectively solve the above problems.In recent years, it is a popular method that using deep learning to solve retinal vessel segmentation.We have improved the loss function for deep learning in order to better handle category imbalances.By using a multi-scale convolutional neural network structure and label processing approach, our results have reached the most advanced level.Our approach is a meaningful attempt to improve blood vessel segmentation and further improve the diagnostic level of eye diseases.

In the summer of 2015, a five-day professional development workshop was held at West Virginia University Institute of Technology, located in Montgomery, West Virginia, with the objective of providing systematic training of project-based learning to high school math teachers. Twenty-two teachers participated in the workshop. Instructors of the workshop were faculty members from West Virginia University Institute of Technology, West Virginia University, and West Virginia State University. The workshop's focus was project-based learning, which employs projects closely related to real-world applications to facilitate delivering abstract concepts. Specifically during the workshop, the participating high school math teachers learned designing engineering projects, mapping engineering projects to Next Generation math standards/objectives, and assessing the outcomes of project-based learning. Each participating teacher is required to implement at least one engineering project in his/her math class and the results will be collected by the superintendents of the three participating school districts. The workshop has two primary hypotheses: (i) teachers who participated in the workshop will increase their self-efficacy toward implementing project-based learning, applying engineering and technology to address content standards and objectives, and using assessments to inform instruction, and (ii) project-based learning will improve students' self-efficacy and learning effectiveness in math, and in turn, will increase their interest/intention to pursue STEM disciplines. The impact of the workshop on teachers is determined through surveys and interviews. Social Cognitive Career Theory is applied to evaluate the impact of project-based learning on the participating teachers' students. Results of the surveys, interviews, and student performance will be presented at the conference.

Efficient visual tracking is a challenging task in the computer vision community due to its large motion uncertainty induced by occlusion, abrupt motion or appearance changes. In this paper, we propose a Hamiltonian Markov Chain Monte Carlo (MCMC) based tracking scheme for efficient tracking within the Bayesian filtering framework, aiming at handling full or partial occlusions, abrupt motion and appearance changes. In this tracking scheme, no complex models are built for motion uncertainties. The object states are augmented by introducing a momentum item and the Hamiltonian dynamics (HD) is integrated into the traditional MCMC-based tracking method. A new object state is proposed by computing a trajectory according to HD, implemented with the Leapfrog method. The new state can be distant from the current object state but, nevertheless, has a high acceptance probability, which consequently bypasses the slow exploration of the state space suffered by traditional random-walk proposal distribution. In addition, the proposed tracking algorithm can avoid being trapped in local maxima, which is suffered by conventional MCMC-based tracking algorithms. Experimental results reveal that our approach is efficient and effective in dealing with various types of tracking scenarios compared with several alternatives.

Similarity measure is a critical component in image retrieval systems, and learning similarity measure from the relevance feedback has become a promising way to enhance retrieval performance. Existing approaches mainly focus on learning the visual similarity measure from online feedbacks or constructing the semantic similarity measure depended on historical feedbacks log. However, there is still a big room to elevate the retrieval performance, because few works take the relationship between the visual similarity and the semantic similarity into account. This paper proposes the collaborative learning similarity measure, CoSim, which focuses on the collaborative learning between the visual content of images and the hidden semantic in log. Concretely, the semantic similarity is first learned from log data and serves as prior knowledge. Then, the visual similarity is learned from a mixture of labeled and unlabeled images. In particular, unlabeled images are exploited for the relevant and irrelevant classes in different ways. Finally, the collaborative learning similarity is produced by integrating the visual similarity and the semantic similarity in a nonlinear way. An empirical study shows that the proposed CoSim is significantly more effective than some existing approaches.

Support vector machine (SVM) based active learning technique plays a key role to alleviate the burden of labeling in relevance feedback. However, most SVM-based active learning algorithms are challenged by the small example problem and the asymmetric distribution problem. This paper proposes a novel active learning scheme that deals with SVM ensemble under the semi-supervised setting to address the fist problem. For the second problem, a bias-ensemble mechanism is developed to guide the classification model to pay more attention on the positive examples than the negative ones. An empirical study shows that the proposed scheme is significantly more effective than some existing approaches.

A 3D-FDTD algorithm is developed and used to compute the emissivity of finite-size and arbitrary-shape objects. Under thermal equilibrium, the emissivity of an object is the same as its absorptivity. The absorptivity is a function of both the scattering cross section and the absorption cross section of the object; in this study, these cross sections were computed using the FDTD approach. Emissivity values for spherical, cylindrical and landmine-like objects as function of observation angle, polarization and permittivity are generated and presented in this paper.

This paper presents a novel cavity backed slot (CBS) antenna for subsurface sensing applications. The CBS antenna is designed to be "matched" in the two-half-space configuration (one half space air; and the other ground) over a relatively wide frequency band. As a result, when attached onto ground surface, it is able to efficiently couple microwave power into and out of the ground. In this study, CBS antennas with operating frequency range [8.5 GHz, 13.6 GHz] are designed to detect objects buried in sand. Data acquisition is carried out using a sandpit with size (125 cm times 100 cm times 80 cm) as the test bed. One transmitting CBS antenna is fixed at the center of the sandpit and one receiving antenna is physically moved along a rectangular grid (i.e., multi-static measurement). An inverse synthetic aperture radar (ISAR) algorithm is adopted for inverse processing. A 4-inch-diameter metallic sphere is used as calibration target; and three targets are tested, including a 3-inch-diameter metallic sphere, a T-shaped copper target, and a landmine simulant. Imaging results are presented and compared with those obtained using horn antennas (which are not "matched" to the air-ground interface). Better signal-to-clutter ratios are demonstrated by the proposed CBS antennas.

WebME is a Web mining environment (system) developed by Tsinghua University. It integrates many functions including: customization-based gathering, rough and subtle filtering, storing, indexing, recognizing, schema and data extracting, classifying, clustering and summarizing of Web pages, intelligent search engine, Web navigation, recommendation of Web information based on association of concept etc. We aim to utilize it as an experimental platform of Web mining on which we can design, implement, test and evaluate various algorithms of Web mining. In the paper, its architecture, main function and running environment are described.

Introduces an approach to recommendation of Web pages based on conceptual association. This approach can produce recommendatory Web page links for users based on the information that associates strongly with user's query, increase the chance to find more relevant links, and therefore improve the recall of a search engine. We discuss the meaning, effect and variety of conceptual association, and describe how to generate associational information related with a user's query and how to realize the association-based recommendation of Web pages. We also present our experimental result and give idea about our further work.