Search

Your search keyword '"Liu, Qiao"' showing total 2,185 results
Start Over Author "Liu, Qiao" Search Limiters Available in Library Collection
2,185 results on '"Liu, Qiao"'

1. Progressive Domain Adaptation for Thermal Infrared Object Tracking

Author

Li, Qiao, Tan, Kanlun, Liu, Qiao, Yuan, Di, Li, Xin, and Liu, Yunpeng

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Due to the lack of large-scale labeled Thermal InfraRed (TIR) training datasets, most existing TIR trackers are trained directly on RGB datasets. However, tracking methods trained on RGB datasets suffer a significant drop-off in TIR data due to the domain shift issue. To this end, in this work, we propose a Progressive Domain Adaptation framework for TIR Tracking (PDAT), which transfers useful knowledge learned from RGB tracking to TIR tracking. The framework makes full use of large-scale labeled RGB datasets without requiring time-consuming and labor-intensive labeling of large-scale TIR data. Specifically, we first propose an adversarial-based global domain adaptation module to reduce domain gap on the feature level coarsely. Second, we design a clustering-based subdomain adaptation method to further align the feature distributions of the RGB and TIR datasets finely. These two domain adaptation modules gradually eliminate the discrepancy between the two domains, and thus learn domain-invariant fine-grained features through progressive training. Additionally, we collect a largescale TIR dataset with over 1.48 million unlabeled TIR images for training the proposed domain adaptation framework. Experimental results on five TIR tracking benchmarks show that the proposed method gains a nearly 6% success rate, demonstrating its effectiveness., Comment: 10 pages, 8 figures

Published
2024

2. A novel class of inhibitors that disrupts the stability of integrin heterodimers identified by CRISPR-tiling-instructed genetic screens

Author

Mattson, Nicole M, Chan, Anthony KN, Miyashita, Kazuya, Mukhaleva, Elizaveta, Chang, Wen-Han, Yang, Lu, Ma, Ning, Wang, Yingyu, Pokharel, Sheela Pangeni, Li, Mingli, Liu, Qiao, Xu, Xiaobao, Chen, Renee, Singh, Priyanka, Zhang, Leisi, Elsayed, Zeinab, Chen, Bryan, Keen, Denise, Pirrotte, Patrick, Rosen, Steven T, Chen, Jianjun, LaBarge, Mark A, Shively, John E, Vaidehi, Nagarajan, Rockne, Russell C, Feng, Mingye, and Chen, Chun-Wei

Subjects
Biochemistry and Cell Biology, Biological Sciences, Cancer, Development of treatments and therapeutic interventions, 5.1 Pharmaceuticals, Humans, Clustered Regularly Interspaced Short Palindromic Repeats, Cell Membrane, Chemical Sciences, Medical and Health Sciences, Biophysics, Developmental Biology, Biological sciences, Biomedical and clinical sciences, Chemical sciences
Abstract

The plasma membrane is enriched for receptors and signaling proteins that are accessible from the extracellular space for pharmacological intervention. Here we conducted a series of CRISPR screens using human cell surface proteome and integrin family libraries in multiple cancer models. Our results identified ITGAV (integrin αV) and its heterodimer partner ITGB5 (integrin β5) as the essential integrin α/β pair for cancer cell expansion. High-density CRISPR gene tiling further pinpointed the integral pocket within the β-propeller domain of ITGAV for integrin αVβ5 dimerization. Combined with in silico compound docking, we developed a CRISPR-Tiling-Instructed Computer-Aided (CRISPR-TICA) pipeline for drug discovery and identified Cpd_AV2 as a lead inhibitor targeting the β-propeller central pocket of ITGAV. Cpd_AV2 treatment led to rapid uncoupling of integrin αVβ5 and cellular apoptosis, providing a unique class of therapeutic action that eliminates the integrin signaling via heterodimer dissociation. We also foresee the CRISPR-TICA approach to be an accessible method for future drug discovery studies.

Published
2024

3. Synergistic Anchored Contrastive Pre-training for Few-Shot Relation Extraction

Author

Luo, Da, Gan, Yanglei, Hou, Rui, Lin, Run, Liu, Qiao, Cai, Yuxiang, and Gao, Wannian

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence
Abstract

Few-shot Relation Extraction (FSRE) aims to extract relational facts from a sparse set of labeled corpora. Recent studies have shown promising results in FSRE by employing Pre-trained Language Models (PLMs) within the framework of supervised contrastive learning, which considers both instances and label facts. However, how to effectively harness massive instance-label pairs to encompass the learned representation with semantic richness in this learning paradigm is not fully explored. To address this gap, we introduce a novel synergistic anchored contrastive pre-training framework. This framework is motivated by the insight that the diverse viewpoints conveyed through instance-label pairs capture incomplete yet complementary intrinsic textual semantics. Specifically, our framework involves a symmetrical contrastive objective that encompasses both sentence-anchored and label-anchored contrastive losses. By combining these two losses, the model establishes a robust and uniform representation space. This space effectively captures the reciprocal alignment of feature distributions among instances and relational facts, simultaneously enhancing the maximization of mutual information across diverse perspectives within the same relation. Experimental results demonstrate that our framework achieves significant performance enhancements compared to baseline models in downstream FSRE tasks. Furthermore, our approach exhibits superior adaptability to handle the challenges of domain shift and zero-shot relation extraction. Our code is available online at https://github.com/AONE-NLP/FSRE-SaCon.

Published
2023

4. Global, regional and national trends and impacts of natural floods, 1990-2022/Tendances mondiales, regionales et nationales relatives a l'impact des inondations, 1990-2022/Tendencias e impactos mundiales, regionales y nacionales de las inundaciones naturales entre 1990 y 2022

Author

Liu, Qiao, Du, Min, Wang, Yaping, Deng, Jie, Yan, Wenxin, Qin, Chenyuan, Liu, Min, and Liu, Jue

Subjects
Floods -- China -- Analysis -- Casualties, Mortality -- China -- Analysis -- Casualties, Health
Abstract

Objective To assess global, regional and national trends in the impact of floods from 1990 to 2022 and determine factors influencing flood-related deaths. Methods We used data on flood disasters from the International Disaster Database for 1990-2022 from 168 countries. We calculated the annual percentage change to estimate trends in the rates of people affected and killed by floods by study period, World Health Organization (WHO) region, country income level and flood type. We used multivariable logistic regression analysis to assess the factors associated with death from floods. Findings From 1990 to 2022, 4713 floods were recorded in 168 countries, which affected > 3.2 billion people, caused 218 353 deaths and were responsible for more than 1.3 trillion United States dollars of economic losses. The WHO Western Pacific Region had the most people affected by floods (> 2.0 billion), accounting for 63.19% (2 024 599 380/3 203 944 965) of all affected populations. The South-East Asia Region had the most deaths (71 713, 32.84%). The African and Eastern Mediterranean Regions had the highest number of people affected and killed by floods per 100 000 population in 2022. The odds of floods causing more than 50 deaths were significantly higher in low-income countries (adjusted odds ratio: 14.34; 95% confidence interval: 7.46 to 30.04) compared with high-income countries. Numbers of people affected and mortality due to floods declined over time. Conclusion Despite the decreases in populations affected and deaths, floods still have a serious impact on people and economies globally, particularly in lower-income countries. Action is needed to improve disaster risk management and flood mitigation. Objectif Evaluer les tendances mondiales, regionales et nationales relatives a l'impact des inondations entre 1990 et 2022, mais aussi determiner les facteurs influengant les deces causes par ces inondations. Methodes Nous avons exploite les donnees concernant 168 pays et reprises dans la base de donnees internationale sur les catastrophes pour la periode allant de 1990 a 2022. Nous avons ensuite calcule la variation annuelle en pourcentage afin de degager des tendances quant au nombre de personnes touchees, voire tuees par des inondations en fonction de la periode d'etude, de la region telle que definie par l'Organisation mondiale de la Sante (OMS), du niveau de revenu du pays et du type d'inondation. Enfin, nous avons employe une analyse multivariee de regression logistique pour determiner les facteurs associes aux deces en cas d'inondations. Resultats Entre 1990 et 2022, 4713 inondations ont ete recensees dans 168 pays. Elles ont touche plus de 3,2 milliards de personnes, cause 218 353 morts et entraine des pertes economiques superieures a 1,3 billion de dollars americains. La region du Pacifique occidental de l'OMS est celle ou le plus grand nombre d'habitants ont ete affectes par ces phenomenes (> 2 milliards), representant 63,19% (2 024 599 380 / 3 203 944 965) de l'ensemble des populations impactees. C'est la region d'Asie du Sud-Est qui comptait le plus grand nombre de deces (71 713, soit 32,84%). La region Africaine et la region de la Mediterranee orientale denombraient toutes deux la plus grande proportion de personnes touchees et tuees lors d'inondations par 100 000 habitants en 2022. La probability que les inondations soient a l'origine de plus de 50 morts etait nettement plus importante dans les pays a revenu faible (odds ratio ajuste : 14,34; intervalle de confiance a 95% : 7,46 a 30,04) que dans les pays a revenu eleve. Letendue des repercussions et le taux de mortalite des inondations diminuaient au fil du temps. Conclusion Malgre une baisse du nombre de personnes touchees et tuees, les inondations continuent a avoir un impact considerable sur les populations et les economies a travers le monde, surtout dans les pays aux revenus plus faibles. Des actions sont requises pour ameliorer la gestion des risques de catastrophes et la lutte contre les inondations. Objetivo Evaluar las tendencias mundiales, regionales y nacionales del impacto de las inundaciones de 1990 a 2022 y determinar los factores que influyen en las muertes relacionadas con las inundaciones. Metodos Se utilizaron datos sobre catastrofes por inundaciones de la Base de Datos Internacional sobre catastrofes para 1990-2022 de 168 paises. Se calculo la variacion porcentual anual para estimar las tendencias de las tasas de personas afectadas y fallecidas por inundaciones segun el periodo de estudio, la region de la Organizacion Mundial de la Salud (OMS), el nivel de ingresos del pais y el tipo de inundacion. Se utilizo un analisis de regresion logistica multivariable para evaluar los factores asociados a la muerte por inundaciones. Resultados Entre 1990 y 2022, se registraron 4713 inundaciones en 168 paises, que afectaron a >3200 millones de personas, causaron 218 353 muertes y fueron responsables de mas de 1,3 billones de dolares estadounidenses de perdidas economicas. La Region del Pacifico Occidental de la OMS fue la mas afectada por las inundaciones (>2000 millones), con un 63,19% (2 024 599 380/3 203 944 965) de todas las poblaciones afectadas. La Region de Asia Sudoriental fue la que registro mas muertes (71 713, 32,84%). Las regiones de Africa y el Mediterraneo Oriental registraron el mayor numero de personas afectadas y fallecidas por inundaciones por cada 100 000 habitantes en 2022. Las probabilidades de que las inundaciones causaran mas de 50 muertes fueron significativamente mayores en los paises de ingresos bajos (razon de probabilidades ajustada: 14,34; intervalo de confianza del 95%: 7,46 a 30,04) en comparacion con los paises de ingresos altos. El numero de personas afectadas y la mortalidad por inundaciones disminuyeron con el tiempo. Conclusion A pesar de la disminucion de las poblaciones afectadas y de las muertes, las inundaciones siguen teniendo un grave impacto en las personas y las economias de todo el mundo, en especial en los paises con ingresos mas bajos. Es preciso actuar para mejorar la gestion del riesgo de catastrofes y la mitigacion de las inundaciones. [phrase omitted], Introduction Natural disasters are catastrophic events of atmospheric, geological and hydrological origins. These disasters include earthquakes, floods, volcanic eruptions, landslides, tsunamis and droughts. (1) From 1998 to 2017, 1.3 million [...]

Published
2024
Full Text
View/download PDF

5. Aspect-oriented Opinion Alignment Network for Aspect-Based Sentiment Classification

Author

Liu, Xueyi, Hou, Rui, Gan, Yanglei, Luo, Da, Li, Changlin, Shi, Xiaojun, and Liu, Qiao

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence, I.2.7
Abstract

Aspect-based sentiment classification is a crucial problem in fine-grained sentiment analysis, which aims to predict the sentiment polarity of the given aspect according to its context. Previous works have made remarkable progress in leveraging attention mechanism to extract opinion words for different aspects. However, a persistent challenge is the effective management of semantic mismatches, which stem from attention mechanisms that fall short in adequately aligning opinions words with their corresponding aspect in multi-aspect sentences. To address this issue, we propose a novel Aspect-oriented Opinion Alignment Network (AOAN) to capture the contextual association between opinion words and the corresponding aspect. Specifically, we first introduce a neighboring span enhanced module which highlights various compositions of neighboring words and given aspects. In addition, we design a multi-perspective attention mechanism that align relevant opinion information with respect to the given aspect. Extensive experiments on three benchmark datasets demonstrate that our model achieves state-of-the-art results. The source code is available at https://github.com/AONE-NLP/ABSA-AOAN., Comment: 8 pages, 5 figure, ECAI 2023

Published
2023

6. Application of PARP inhibitors combined with immune checkpoint inhibitors in ovarian cancer

Author

Fen Xiao, ZhiBin Wang, Liu Qiao, Xiu Zhang, NaYiYuan Wu, Jing Wang, and Xing Yu

Subjects
Ovarian cancer, Immune checkpoint inhibitor, PARP inhibitors, Immunotherapy, Medicine
Abstract

Abstract The advent of polyadenosine diphosphate ribose polymerase inhibitors (PARPi) has brought about significant changes in the field of ovarian cancer treatment. However, in 2022, Rucaparib, Olaparib, and Niraparib, had their marketing approval revoked for third-line and subsequent therapies due to an increased potential for adverse events. Consequently, the exploration of new treatment modalities remains imperative. Recently, the integration of PARPi with immune checkpoint inhibitors (ICIs) has emerged as a potential remedy option within the context of ovarian cancer. This article offers a comprehensive examination of the mechanisms and applications of PARPi and ICIs in the treatment of ovarian cancer. It synthesizes the existing evidence supporting their combined use and discusses key considerations that merit attention in ongoing development efforts.

Published
2024
Full Text
View/download PDF

7. Language-Enhanced Session-Based Recommendation with Decoupled Contrastive Learning

Author

Zhang, Zhipeng, Tong, Piao, Ma, Yingwei, Liu, Qiao, Liu, Xujiang, and Luo, Xu

Subjects
Computer Science - Information Retrieval
Abstract

Session-based recommendation techniques aim to capture dynamic user behavior by analyzing past interactions. However, existing methods heavily rely on historical item ID sequences to extract user preferences, leading to challenges such as popular bias and cold-start problems. In this paper, we propose a hybrid multimodal approach for session-based recommendation to address these challenges. Our approach combines different modalities, including textual content and item IDs, leveraging the complementary nature of these modalities using CatBoost. To learn universal item representations, we design a language representation-based item retrieval architecture that extracts features from the textual content utilizing pre-trained language models. Furthermore, we introduce a novel Decoupled Contrastive Learning method to enhance the effectiveness of the language representation. This technique decouples the sequence representation and item representation space, facilitating bidirectional alignment through dual-queue contrastive learning. Simultaneously, the momentum queue provides a large number of negative samples, effectively enhancing the effectiveness of contrastive learning. Our approach yielded competitive results, securing a 5th place ranking in KDD CUP 2023 Task 1. We have released the source code and pre-trained models associated with this work.

Published
2023

8. Global existence and optimal decay of solutions to the incompressible Oldroyd-B model with only stress tensor dissipation and without damping mechanism

Author

Chen, Zhi, Feng, Weixun, and Liu, Qiao

Subjects
Mathematics - Analysis of PDEs, 35Q35, 35A01, 35A02, 35B45, F.2.2, I.2.7
Abstract

We study the $d$-dimensional ($d\geq2$) incompressible Oldroyd-B model with only stress tensor diffusion and without velocity dissipation as well as the damping mechanism on the stress tensor. Firstly, based upon some new observations on the model, we develope the pure energy argument (independent of spectral analysis) in general $L^p$ framework, and present a small initial data global existence and uniqueness of solutions to the model. Our results yield that the coupling and interaction of the velocity and the non-Newtonian stress actually enhances the regularity of the system. Later, by adding some additional $L^2$ type conditions on the low frequencies of the initial data $(u_0,\tau_0)$, %but without any more smallness restrictions, we obtain the optimal time-decay rates of the global solution $(u,\tau)$. Our result solves the problem proposed in Wang, Wu, Xu and Zhong \cite{Wang-Wu-Xu-Zhong} ({\it J. Funct. Anal.}, 282 (2022), 109332.)., Comment: we improve the result, we will give a new result

Published
2023

26. Identification of Bletilla striata rust pathogen and resistance resources screening

Author

WU Qiaofen, MA Xiaoya, XIA Ke, LU Xi, LIU Qiao, ZHAO Zhiguo, ZHENG Wenjun, and QIU Shuo

Subjects
bletilla striata, rust, pathogen identification, coleosporium sp., resistance resource, Botany, QK1-989
Abstract

In order to identify the pathogen that caused rust of Bletilla striata in Guangxi Zhuang Autonomous Region and to screen resistance resources for B. striata, the rust pathogen was isolated from infected leaves of B. striata and identified using morphology and molecular methods. At the same time, the resistance of 23 B. striata from different regions was evaluated by artificial inoculation of pathogen. The results were as follows: (1) The urediopores of strain X2 isolated from diseased leaves in B. striata were golden yellow, oval, (21.43-30.95) μm × (13.10-19.05) μm. The teleutospores were golden yellow, obovate or clavate, (17.25-30) μm × (5.5-6.65) μm. (2) The length of ITS sequence of strain X2(OQ826009) was 689 bp, which were compared with other ITS sequences in the GenBank, and the similarity reached 95.86% compared with Coleosporium sp. (KY783686.1). But the phylogenetic tree showed that the sequence was clustered together with two sequences of Coleosporium bletiae (MN108161.1, OP363680.1). The strain X2 was identified as Coleosporium bletiae by combining morphology and molecular methods. (3) After 14 days of artificial inoculation with strain X2, the disease index of 23 Bletilla striata ranged from 0 to 70.7. And the resistance of strain X2 from 23 B. striata were divided into six different levels according to the disease index. Among them, one immune materials was identified for the disease index was 0. Four highly resistant materials were identified for the disease index was 1.7-4.7. Six disease-resistant materials were identified for the disease index was 5.6-9.4. Five medium-resistant materials were identified for the disease index was 12.7-18.3. Five susceptible materials were identified for the disease index was 32.0-49.1 and two highly susceptible materials were identified for the disease index was 62.2-70.7. In conclusion, five materials (one immune material and four highly resistant materials) of B. striata from different places (Honghe City in Yunnan Province, Gongcheng County and Baise City in Guangxi Zhuang Autonomous Region, Zunyi City in Guizhou Province and Yichang City in Hubei Province, respectively) can be promoted directly or applied as parent materials for creating excellent germplasm to resist the rust in B. striata for they express immune or highly resistance. This study provides the basis for further research on pathogenesis and control of rust in B. striata.

Published
2024
Full Text
View/download PDF

27. Comprehensive tissue deconvolution of cell-free DNA by deep learning for disease diagnosis and monitoring

Author

Li, Shuo, Zeng, Weihua, Ni, Xiaohui, Liu, Qiao, Li, Wenyuan, Stackpole, Mary L, Zhou, Yonggang, Gower, Arjan, Krysan, Kostyantyn, Ahuja, Preeti, Lu, David S, Raman, Steven S, Hsu, William, Aberle, Denise R, Magyar, Clara E, French, Samuel W, Han, Steven-Huy B, Garon, Edward B, Agopian, Vatche G, Wong, Hung, Dubinett, Steven M, and Zhou, Xianghong Jasmine

Subjects
Human Genome, Bioengineering, Genetics, Detection, screening and diagnosis, 4.1 Discovery and preclinical testing of markers and technologies, Generic health relevance, Humans, Cell-Free Nucleic Acids, Deep Learning, DNA Methylation, Biomarkers, Promoter Regions, Genetic, Biomarkers, Tumor, cell-free DNA, DNA methylation, tissue deconvolution, disease diagnosis, disease monitoring
Abstract

Plasma cell-free DNA (cfDNA) is a noninvasive biomarker for cell death of all organs. Deciphering the tissue origin of cfDNA can reveal abnormal cell death because of diseases, which has great clinical potential in disease detection and monitoring. Despite the great promise, the sensitive and accurate quantification of tissue-derived cfDNA remains challenging to existing methods due to the limited characterization of tissue methylation and the reliance on unsupervised methods. To fully exploit the clinical potential of tissue-derived cfDNA, here we present one of the largest comprehensive and high-resolution methylation atlas based on 521 noncancer tissue samples spanning 29 major types of human tissues. We systematically identified fragment-level tissue-specific methylation patterns and extensively validated them in orthogonal datasets. Based on the rich tissue methylation atlas, we develop the first supervised tissue deconvolution approach, a deep-learning-powered model, cfSort, for sensitive and accurate tissue deconvolution in cfDNA. On the benchmarking data, cfSort showed superior sensitivity and accuracy compared to the existing methods. We further demonstrated the clinical utilities of cfSort with two potential applications: aiding disease diagnosis and monitoring treatment side effects. The tissue-derived cfDNA fraction estimated from cfSort reflected the clinical outcomes of the patients. In summary, the tissue methylation atlas and cfSort enhanced the performance of tissue deconvolution in cfDNA, thus facilitating cfDNA-based disease detection and longitudinal treatment monitoring.

Published
2023

28. CausalEGM: a general causal inference framework by encoding generative modeling

Author

Liu, Qiao, Chen, Zhongren, and Wong, Wing Hung

Subjects
Statistics - Machine Learning, Computer Science - Machine Learning
Abstract

Although understanding and characterizing causal effects have become essential in observational studies, it is challenging when the confounders are high-dimensional. In this article, we develop a general framework $\textit{CausalEGM}$ for estimating causal effects by encoding generative modeling, which can be applied in both binary and continuous treatment settings. Under the potential outcome framework with unconfoundedness, we establish a bidirectional transformation between the high-dimensional confounders space and a low-dimensional latent space where the density is known (e.g., multivariate normal distribution). Through this, CausalEGM simultaneously decouples the dependencies of confounders on both treatment and outcome and maps the confounders to the low-dimensional latent space. By conditioning on the low-dimensional latent features, CausalEGM can estimate the causal effect for each individual or the average causal effect within a population. Our theoretical analysis shows that the excess risk for CausalEGM can be bounded through empirical process theory. Under an assumption on encoder-decoder networks, the consistency of the estimate can be guaranteed. In a series of experiments, CausalEGM demonstrates superior performance over existing methods for both binary and continuous treatments. Specifically, we find CausalEGM to be substantially more powerful than competing methods in the presence of large sample sizes and high dimensional confounders. The software of CausalEGM is freely available at https://github.com/SUwonglab/CausalEGM.

Published
2022

29. Mutual Information Learned Regressor: an Information-theoretic Viewpoint of Training Regression Systems

Author

Yi, Jirong, Zhang, Qiaosheng, Chen, Zhen, Liu, Qiao, Shao, Wei, He, Yusen, and Wang, Yaohua

Subjects
Statistics - Machine Learning, Computer Science - Information Theory, Computer Science - Machine Learning, Mathematics - Optimization and Control
Abstract

As one of the central tasks in machine learning, regression finds lots of applications in different fields. An existing common practice for solving regression problems is the mean square error (MSE) minimization approach or its regularized variants which require prior knowledge about the models. Recently, Yi et al., proposed a mutual information based supervised learning framework where they introduced a label entropy regularization which does not require any prior knowledge. When applied to classification tasks and solved via a stochastic gradient descent (SGD) optimization algorithm, their approach achieved significant improvement over the commonly used cross entropy loss and its variants. However, they did not provide a theoretical convergence analysis of the SGD algorithm for the proposed formulation. Besides, applying the framework to regression tasks is nontrivial due to the potentially infinite support set of the label. In this paper, we investigate the regression under the mutual information based supervised learning framework. We first argue that the MSE minimization approach is equivalent to a conditional entropy learning problem, and then propose a mutual information learning formulation for solving regression problems by using a reparameterization technique. For the proposed formulation, we give the convergence analysis of the SGD algorithm for solving it in practice. Finally, we consider a multi-output regression data model where we derive the generalization performance lower bound in terms of the mutual information associated with the underlying data distribution. The result shows that the high dimensionality can be a bless instead of a curse, which is controlled by a threshold. We hope our work will serve as a good starting point for further research on the mutual information based regression., Comment: 28 pages, 2 figures, presubmitted to AISTATS2023 for reviewing

Published
2022

30. Mutual Information Learned Classifiers: an Information-theoretic Viewpoint of Training Deep Learning Classification Systems

Author

Yi, Jirong, Zhang, Qiaosheng, Chen, Zhen, Liu, Qiao, and Shao, Wei

Subjects
Computer Science - Machine Learning, Computer Science - Information Theory
Abstract

Deep learning systems have been reported to acheive state-of-the-art performances in many applications, and one of the keys for achieving this is the existence of well trained classifiers on benchmark datasets which can be used as backbone feature extractors in downstream tasks. As a main-stream loss function for training deep neural network (DNN) classifiers, the cross entropy loss can easily lead us to find models which demonstrate severe overfitting behavior when no other techniques are used for alleviating it such as data augmentation. In this paper, we prove that the existing cross entropy loss minimization for training DNN classifiers essentially learns the conditional entropy of the underlying data distribution of the dataset, i.e., the information or uncertainty remained in the labels after revealing the input. In this paper, we propose a mutual information learning framework where we train DNN classifiers via learning the mutual information between the label and input. Theoretically, we give the population error probability lower bound in terms of the mutual information. In addition, we derive the mutual information lower and upper bounds for a concrete binary classification data model in $\mbR^n$, and also the error probability lower bound in this scenario. Besides, we establish the sample complexity for accurately learning the mutual information from empirical data samples drawn from the underlying data distribution. Empirically, we conduct extensive experiments on several benchmark datasets to support our theory. Without whistles and bells, the proposed mutual information learned classifiers (MILCs) acheive far better generalization performances than the state-of-the-art classifiers with an improvement which can exceed more than 10\% in testing accuracy., Comment: 42 pages, 15 figures, 4 tables, 14 lemmas, theorems, or corollaries. arXiv admin note: text overlap with arXiv:2209.10058

Published
2022

31. Pose-Aided Video-based Person Re-Identification via Recurrent Graph Convolutional Network

Author

Pan, Honghu, Liu, Qiao, Chen, Yongyong, He, Yunqi, Zheng, Yuan, Zheng, Feng, and He, Zhenyu

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Existing methods for video-based person re-identification (ReID) mainly learn the appearance feature of a given pedestrian via a feature extractor and a feature aggregator. However, the appearance models would fail when different pedestrians have similar appearances. Considering that different pedestrians have different walking postures and body proportions, we propose to learn the discriminative pose feature beyond the appearance feature for video retrieval. Specifically, we implement a two-branch architecture to separately learn the appearance feature and pose feature, and then concatenate them together for inference. To learn the pose feature, we first detect the pedestrian pose in each frame through an off-the-shelf pose detector, and construct a temporal graph using the pose sequence. We then exploit a recurrent graph convolutional network (RGCN) to learn the node embeddings of the temporal pose graph, which devises a global information propagation mechanism to simultaneously achieve the neighborhood aggregation of intra-frame nodes and message passing among inter-frame graphs. Finally, we propose a dual-attention method consisting of node-attention and time-attention to obtain the temporal graph representation from the node embeddings, where the self-attention mechanism is employed to learn the importance of each node and each frame. We verify the proposed method on three video-based ReID datasets, i.e., Mars, DukeMTMC and iLIDS-VID, whose experimental results demonstrate that the learned pose feature can effectively improve the performance of existing appearance models.

Published
2022

32. Effects of Nitrogen and Phosphorus Additions on Aggregate-associated Soil Carbon and Interactions with Phosphorus Fractions in Evergreen Broad-leaved Forest

Author

HE Lingyun, CHEN Fusheng, ZHENG Zhiyu, LIU Qiao, WANG Shengnan, and WANG Fangchao

Subjects
phosphorus input, nitrogen deposition, soil aggregate, phosphorus component, soil organic carbon, evergreen broad-leaved forest, Environmental sciences, GE1-350, Agriculture
Abstract

[Objective] To explore how phosphorus input regulates the effects of atmospheric nitrogen deposition on soil aggregate organic carbon content and the relationship between carbon and phosphorus. [Methods] A long-term monitoring test platform for nutrient addition was established in evergreen broad-leaved forest soil for 6 years (2015—2021), including 4 treatments: Control [P0+N0, P 0 kg/(hm2·a)+N 0 kg/(hm2·a)], nitrogen addition [P0+N100, P 0 kg/(hm2·a)+N 100 kg/(hm2·a)], phosphorus input [P50+N0, P 50 kg/(hm2·a)+N 0 kg/(hm2·a)] and nitrogen and phosphorus were simultaneously input (P50+N100, P 50 kg/(hm2·a)+N 100 kg/(hm2·a)], and each treatment was repeated 3 times, and a total of 12 plots were obtained. 0—10 cm soil samples were collected from the sample site in August 2021 to determine the basic physical and chemical properties, soil particle size distribution, phosphorus components of soil aggregates of different particle sizes, and organic carbon (SOC) content. [Results] (1) Under P0 treatment, nitrogen addition significantly increased the proportion of large aggregates, decreased the clay and powder contents, and increased the SOC content in each aggregate particle size. Nitrogen addition significantly decreased and increased the content of labile phosphorus (LP) and residual phosphorus (RP) in aggregate clay and powder, respectively. (2) Under P50 treatment, nitrogen addition significantly increased the geometric mean diameter (GMD) of soil aggregates, but had no significant effects on phosphorus components and SOC contents in aggregates of different particle sizes. (3) Under P0 treatment, soil aggregate SOC was positively correlated with refractory phosphorus. There was no significant correlation between SOC of soil aggregates and functional phosphorus components under P50 treatment. It was concluded that under P0 treatment, nitrogen addition could increase soil organic carbon retention by increasing the refractory phosphorus in soil aggregate clay and powder particle size of evergreen broad-leaved forest. Under P50 treatment, the effect of nitrogen addition on organic carbon in each particle size may be regulated by biological factors and has nothing to do with the availability of phosphorus. [Conclusion] The effect of nitrogen deposition on the internal relationship of carbon and phosphorus in soil aggregates in evergreen broad-leaved forests was regulated by phosphorus. The results of this study provide data supported for forest soil carbon cycling in response to global climate change.

Published
2024
Full Text
View/download PDF

34. Graph Convolutional Networks for Multi-modality Medical Imaging: Methods, Architectures, and Clinical Applications

Author

Ding, Kexin, Zhou, Mu, Wang, Zichen, Liu, Qiao, Arnold, Corey W., Zhang, Shaoting, and Metaxas, Dimitri N.

Subjects
Electrical Engineering and Systems Science - Image and Video Processing, Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning
Abstract

Image-based characterization and disease understanding involve integrative analysis of morphological, spatial, and topological information across biological scales. The development of graph convolutional networks (GCNs) has created the opportunity to address this information complexity via graph-driven architectures, since GCNs can perform feature aggregation, interaction, and reasoning with remarkable flexibility and efficiency. These GCNs capabilities have spawned a new wave of research in medical imaging analysis with the overarching goal of improving quantitative disease understanding, monitoring, and diagnosis. Yet daunting challenges remain for designing the important image-to-graph transformation for multi-modality medical imaging and gaining insights into model interpretation and enhanced clinical decision support. In this review, we present recent GCNs developments in the context of medical image analysis including imaging data from radiology and histopathology. We discuss the fast-growing use of graph network architectures in medical image analysis to improve disease diagnosis and patient outcomes in clinical practice. To foster cross-disciplinary research, we present GCNs technical advancements, emerging medical applications, identify common challenges in the use of image-based GCNs and their extensions in model interpretation, large-scale benchmarks that promise to transform the scope of medical image studies and related graph-driven medical research.

Published
2022

35. Analysis of Two-Tier Question Scoring Methods: A Case Study on the Lawson's Classroom Test of Scientific Reasoning

Author

Zhou, Shao-Na, Liu, Qiao-Yi, Koenig, Kathleen, Xiao, Qiu-ye Li-Yang, and Bao, Lei

Abstract

The Lawson's Classroom Test of Scientific Reasoning (LCTSR) is a popular instrument that measures the development of students' scientific reasoning skills. The instrument has a two-tier question design, which has led to multiple ways of scoring and interpretation. In this research, a method of pattern analysis was proposed and applied to analyze two-tier item pairs on the subskill of Control-of-Variable (COV) of LCTSR. The data were collected from students in grade 4 through college in both the United States and China. Students' response patterns to two combined item pairs were analyzed and compared against students at different grade levels and reasoning development levels. Six performance levels were established based on students' response patterns, serving as indicators of COV reasoning development levels. With the new method, a relation was obtained between students' level of COV skills and grade level, as well as their level of overall reasoning development. It can provide useful information on the possible developmental levels of students' reasoning skills.

Published
2021

36. Infant BCG vaccination and risk of pulmonary and extrapulmonary tuberculosis throughout the life course: a systematic review and individual participant data meta-analysis.

Author

Huerga, Helena, Villalba, Julian, Grandjean, Louis, Sotgiu, Giovanni, Egere, Uzochukwu, Singh, Sarman, Zhu, Limei, Lienhardt, Christian, Denholm, Justin, Seddon, James, Whalen, Christopher, García-Basteiro, Alberto, Triasih, Rina, Chen, Cheng, Singh, Jitendra, Huang, Li-Min, Sharma, Surendra, Hannoun, Djohar, Del Corral, Helena, Mandalakas, Anna, Malone, LaShaunda, Ling, Du-Lin, Kritski, Afrânio, Stein, Catherine, Vashishtha, Richa, Boulahbal, Fadila, Fang, Chi-Tai, Boom, W, Netto, Eduardo, Lemos, Antonio, Hesseling, Anneke, Kay, Alexander, Jones-López, Edward, Horsburgh, C, Lange, Christoph, Andrews, Jason, Martinez, Leonardo, Cords, Olivia, Liu, Qiao, Acuna-Villaorduna, Carlos, Bonnet, Maryline, Fox, Greg, Carvalho, Anna, Chan, Pei-Chun, Croda, Julio, Hill, Philip, Lopez-Varela, Elisa, Donkor, Simon, Fielding, Katherine, Graham, Stephen, Espinal, Marcos, Kampmann, Beate, and Reingold, Arthur

Subjects
Adolescent, Adult, Aged, BCG Vaccine, Child, Child, Preschool, Humans, Infant, Infant, Newborn, Retrospective Studies, Tuberculosis, Tuberculosis, Pulmonary, Vaccination
Abstract

BACKGROUND: BCG vaccines are given to more than 100 million children every year, but there is considerable debate regarding the effectiveness of BCG vaccination in preventing tuberculosis and death, particularly among older children and adults. We therefore aimed to investigate the age-specific impact of infant BCG vaccination on tuberculosis (pulmonary and extrapulmonary) development and mortality. METHODS: In this systematic review and individual participant data meta-analysis, we searched MEDLINE, Web of Science, BIOSIS, and Embase without language restrictions for case-contact cohort studies of tuberculosis contacts published between Jan 1, 1998, and April 7, 2018. Search terms included mycobacterium tuberculosis, TB, tuberculosis, and contact. We excluded cohort studies that did not provide information on BCG vaccination or were done in countries that did not recommend BCG vaccination at birth. Individual-level participant data for a prespecified list of variables, including the characteristics of the exposed participant (contact), the index case, and the environment, were requested from authors of all eligible studies. Our primary outcome was a composite of prevalent (diagnosed at or within 90 days of baseline) and incident (diagnosed more than 90 days after baseline) tuberculosis in contacts exposed to tuberculosis. Secondary outcomes were pulmonary tuberculosis, extrapulmonary tuberculosis, and mortality. We derived adjusted odds ratios (aORs) using mixed-effects, binary, multivariable logistic regression analyses with study-level random effects, adjusting for the variable of interest, baseline age, sex, previous tuberculosis, and whether data were collected prospectively or retrospectively. We stratified our results by contact age and Mycobacterium tuberculosis infection status. This study is registered with PROSPERO, CRD42020180512. FINDINGS: We identified 14 927 original records from our database searches. We included participant-level data from 26 cohort studies done in 17 countries in our meta-analysis. Among 68 552 participants, 1782 (2·6%) developed tuberculosis (1309 [2·6%] of 49 686 BCG-vaccinated participants vs 473 [2·5%] of 18 866 unvaccinated participants). The overall effectiveness of BCG vaccination against all tuberculosis was 18% (aOR 0·82, 95% CI 0·74-0·91). When stratified by age, BCG vaccination only significantly protected against all tuberculosis in children younger than 5 years (aOR 0·63, 95% CI 0·49-0·81). Among contacts with a positive tuberculin skin test or IFNγ release assay, BCG vaccination significantly protected against tuberculosis among all participants (aOR 0·81, 95% CI 0·69-0·96), participants younger than 5 years (0·68, 0·47-0·97), and participants aged 5-9 years (0·62, 0·38-0·99). There was no protective effect among those with negative tests, unless they were younger than 5 years (0·54, 0·32-0·90). 14 cohorts reported on whether tuberculosis was pulmonary or extrapulmonary (n=57 421). BCG vaccination significantly protected against pulmonary tuberculosis among all participants (916 [2·2%] in 41 119 vaccinated participants vs 334 [2·1%] in 16 161 unvaccinated participants; aOR 0·81, 0·70-0·94) but not against extrapulmonary tuberculosis (106 [0·3%] in 40 318 vaccinated participants vs 38 [0·2%] in 15 865 unvaccinated participants; 0·96, 0·65-1·41). In the four studies with mortality data, BCG vaccination was significantly protective against death (0·25, 0·13-0·49). INTERPRETATION: Our results suggest that BCG vaccination at birth is effective at preventing tuberculosis in young children but is ineffective in adolescents and adults. Immunoprotection therefore needs to be boosted in older populations. FUNDING: National Institutes of Health.

Published
2022

49. An Informative Tracking Benchmark

Author

Li, Xin, Liu, Qiao, Pei, Wenjie, Shen, Qiuhong, Wang, Yaowei, Lu, Huchuan, and Yang, Ming-Hsuan

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Along with the rapid progress of visual tracking, existing benchmarks become less informative due to redundancy of samples and weak discrimination between current trackers, making evaluations on all datasets extremely time-consuming. Thus, a small and informative benchmark, which covers all typical challenging scenarios to facilitate assessing the tracker performance, is of great interest. In this work, we develop a principled way to construct a small and informative tracking benchmark (ITB) with 7% out of 1.2 M frames of existing and newly collected datasets, which enables efficient evaluation while ensuring effectiveness. Specifically, we first design a quality assessment mechanism to select the most informative sequences from existing benchmarks taking into account 1) challenging level, 2) discriminative strength, 3) and density of appearance variations. Furthermore, we collect additional sequences to ensure the diversity and balance of tracking scenarios, leading to a total of 20 sequences for each scenario. By analyzing the results of 15 state-of-the-art trackers re-trained on the same data, we determine the effective methods for robust tracking under each scenario and demonstrate new challenges for future research direction in this field., Comment: 10 pages, 6 figures

Published
2021

Catalog

Books, media, physical & digital resources
See catalog results