A5088945316- Topic Modeling
- Natural Language Processing Techniques
- Advanced Text Analysis Techniques
- Marine animal studies overview
- Erythropoietin and Anemia Treatment
- Animal Vocal Communication and Behavior
- Connective Tissue Growth Factor Research
- Hemoglobinopathies and Related Disorders
- Interstitial Lung Diseases and Idiopathic Pulmonary Fibrosis
- Music and Audio Processing
- Multimodal Machine Learning Applications
- Ethics and Social Impacts of AI
- Dialysis and Renal Disease Management
- Big Data and Business Intelligence
- Systemic Sclerosis and Related Diseases
- Cancer Diagnosis and Treatment
- Pancreatic and Hepatic Oncology Research
- Renal cell carcinoma treatment
- Anesthesia and Pain Management
- Underwater Acoustics Research
- Nausea and vomiting management
- Species Distribution and Climate Change
- Advanced Image and Video Retrieval Techniques
- Diverse Musicological Studies
- Advanced Graph Neural Networks
University of Illinois Urbana-Champaign
2008-2025
FibroGen (United States)
2011-2022
Microsoft Research Asia (China)
2021-2022
Microsoft (United States)
2019-2021
Fudan University
2015-2021
Microsoft Research (United Kingdom)
2021
Zhongshan Hospital
2013-2021
Université de Bretagne Occidentale
2021
Centre National de la Recherche Scientifique
2021
Harbin Institute of Technology
2018-2021
This paper creates a paradigm shift with regard to the way we build neural extractive summarization systems. Instead of following commonly used framework extracting sentences individually and modeling relationship between sentences, formulate task as semantic text matching problem, in which source document candidate summaries will be (extracted from original text) matched space. Notably, this is well-grounded our comprehensive analysis inherent gap sentence-level summary-level extractors...
Safety concerns with erythropoietin analogues and intravenous (IV) iron for treatment of anemia in CKD necessitate development safer therapies. Roxadustat (FG-4592) is an orally bioavailable hypoxia-inducible factor (HIF) prolyl hydroxylase inhibitor that promotes coordinated erythropoiesis through HIF-mediated transcription. We performed open-label, randomized hemoglobin (Hb) correction study anemic (Hb≤10.0 g/dl) patients incident to hemodialysis (HD) or peritoneal dialysis (PD). Sixty...
The recent years have seen remarkable success in the use of deep neural networks on text summarization. However, there is no clear understanding why they perform so well, or how might be improved. In this paper, we seek to better understand extractive summarization systems could benefit from different types model architectures, transferable knowledge and learning schemas. Besides, find an effective way improve current framework achieve state-of-the-art result CNN/DailyMail by a large margin...
Automated acoustic recorders can collect long-term soundscape data containing species-specific signals in remote environments. Ecologists have increasingly used them for studying diverse fauna around the globe. Deep learning methods gained recent attention automating process of species identification recordings. We present an end-to-end pipeline training a convolutional neural network (CNN) multi-species multi-label classification recordings, starting from raw, unlabeled audio. Training are...
Ming Zhong, Da Yin, Tao Yu, Ahmad Zaidi, Mutethia Mutuma, Rahul Jha, Ahmed Hassan Awadallah, Asli Celikyilmaz, Yang Liu, Xipeng Qiu, Dragomir Radev. Proceedings of the 2021 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2021.
Video-text retrieval plays an essential role in multi-modal research and has been widely used many real-world web applications. The CLIP (Contrastive Language-Image Pre-training), image-language pre-training model, demonstrated the power of visual concepts learning from collected image-text datasets. In this paper, we propose a CLIP4Clip model to transfer knowledge video-language end-to-end manner. Several questions are investigated via empirical studies: 1) Whether image feature is enough...
We present a deep learning approach towards the large-scale prediction and analysis of bird acoustics from 100 different species. use spectrograms constructed on audio recordings Cornell Bird Challenge (CBC)2020 dataset, which includes multiple potentially overlapping vocalizations with background noise. Our experiments show that hybrid modeling involves Convolutional Neural Network (CNN) for representation slice spectrogram, Recurrent (RNN) temporal component to combine across time-points...
Over a decade after the Cook Inlet beluga (Delphinapterus leucas) was listed as endangered in 2008, population has shown no sign of recovery. Lack ecological knowledge limits understanding of, and ability to manage, potential threats impeding recovery this declining population. National Oceanic Atmospheric Administration Fisheries, partnership with Alaska Department Fish Game, initiated passive acoustics monitoring program 2017 investigate seasonal occurrence by deploying series acoustic...
Dialogue is an essential part of human communication and cooperation. Existing research mainly focuses on short dialogue scenarios in a one-on-one fashion. However, multi-person interactions the real world, such as meetings or interviews, are frequently over few thousand words. There still lack corresponding powerful tools to understand process long dialogues. Therefore, this work, we present pre-training framework for understanding summarization. Considering nature conversations, propose...
Microfluidic technology allows the manipulation of mass-limited samples and when used with cultured cells, enables control extracellular microenvironment, making it well suited for studying neurons their response to environmental perturbations. While matrix-assisted laser desorption/ionization (MALDI) mass spectrometry (MS) provides off-line coupling microfluidic devices characterizing small-volume releasates, performing quantitative studies MALDI is challenging. Here we describe a...
In this paper, we take stock of the current state summarization datasets and explore how different factors influence generalization behaviour neural extractive models. Specifically, first propose several properties datasets, which matter for Then build connection between priors residing in model designs, analyzing choices structure design training methods. Finally, by taking a typical dataset as an example, rethink process based on experience above analysis. We demonstrate that when have...
Previous work for text summarization in scientific domain mainly focused on the content of input document, but seldom considering its citation network. However, papers are full uncommon domain-specific terms, making it almost impossible model to understand true meaning without help relevant research community. In this paper, we redefine task by utilizing their graph and propose a graph-based CGSum which can incorporate information both source paper references. addition, construct novel...
Although domain shift has been well explored in many NLP applications, it still received little attention the of extractive text summarization. As a result, model is under-utilizing nature training data due to ignoring difference distribution sets and shows poor generalization on unseen domain. With above limitation mind, this paper, we first extend conventional definition from categories into sources for summarization task. Then re-purpose multi-domain dataset verify how gap between...
IntroductionErythropoiesis-stimulating agents are associated with increased cardiovascular risk when higher doses used toward hematocrit targets. Patients new to dialysis at for morbidity and mortality. Systematic evaluation of this population was predefined in the roxadustat clinical development program. Roxadustat is a hypoxia-inducible prolyl hydroxylase inhibitor.MethodsData were pooled from 3 phase 3, randomized, open-label, active-controlled trials. Eligible adults had kidney failure...
The goal of this project is to use acoustic signatures detect, classify, and count the calls four populations blue whales so that, ultimately, conservation status each population can be better assessed. We used manual annotations from 350 h audio recordings underwater hydrophones in Indian Ocean build a deep learning model song types. method we was Siamese neural networks (SNN), class network architectures that are find similarity inputs by comparing their feature vectors, finding they...
Neural network-based models augmented with unsupervised pre-trained knowledge have achieved impressive performance on text summarization. However, most existing evaluation methods are limited to an in-domain setting, where summarizers trained and evaluated the same dataset. We argue that this approach can narrow our understanding of generalization ability for different summarization systems. In paper, we perform in-depth analysis characteristics datasets investigate under a cross-dataset in...
Meetings are a key component of human collaboration. As increasing numbers meetings recorded and transcribed, meeting summaries have become essential to remind those who may or not attended the about decisions made tasks be completed. However, it is hard create single short summary that covers all content long involving multiple people topics. In order satisfy needs different types users, we define new query-based multi-domain summarization task, where models select summarize relevant spans...
Runtime and scalability of large neural networks can be significantly affected by the placement operations in their dataflow graphs on suitable devices. With increasingly complex network architectures heterogeneous device characteristics, finding a reasonable is extremely challenging even for domain experts. Most existing automated approaches are impractical due to significant amount compute required inability generalize new, previously held-out graphs. To address both limitations, we...
Background: Early use of enteral nutrition (EN) is indicated following surgical resection esophageal cancer. However, early EN support does not always meet the optimal calorie or protein requirements, and benefits supplementary parenteral (PN) remain unclear. We aimed to evaluate efficacy safety PN esophagectomy. Materials Methods: enrolled 80 consecutive patients who underwent Resting energy expenditure body composition measurements were performed in all preoperatively postoperatively. was...