Browse all papers
07 Dec 2015-
TL;DR: The learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks.
Abstract: We propose a simple, yet effective approach for spatiotemporal feature learning using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset. Our findings are three-fold: 1) 3D ConvNets are more suitable for spatiotemporal feature learning compared to 2D ConvNets, 2) A homogeneous architecture with small 3x3x3 convolution kernels in all layers is among the best performing architectures for 3D ConvNets, and 3) Our learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks. In addition, the features are compact: achieving 52.8% accuracy on UCF101 dataset with only 10 dimensions and also very efficient to compute due to the fast inference of ConvNets. Finally, they are conceptually very simple and easy to train and use.
4,858 citations
University of Washington1, Metropolitan Police Service2, London South Bank University3, Health Effects Institute4, Iran University of Medical Sciences5, Hong Kong Baptist University6, Norwegian Institute of Public Health7, Sapienza University of Rome8, Hacettepe University9, Cairo University10, University of Hohenheim11, College of Health Sciences, Bahrain12, Government College13, University College London14, Birzeit University15, Kwame Nkrumah University of Science and Technology16, University of Extremadura17, International Centre for Diarrhoeal Disease Research, Bangladesh18, University of the Republic19, Debre markos University20, University of Oxford21, Karolinska Institutet22, Charité23, University of Cartagena24, Albany State University25, Ahmadu Bello University26, Boston University27, École Polytechnique Fédérale de Lausanne28, St Mary's Hospital29, University of the Philippines Manila30, Emory University31, Dalarna University32, Boston Children's Hospital33, University of Manitoba34, University of Mannheim35, Public Health Agency of Canada36, Lawrence Berkeley National Laboratory37, Samsung38, University of Gothenburg39, Pontifical Xavierian University40, Madawalabu University41, University of Southampton42, Imperial College London43, King's College London44, Addis Ababa University45, Post Graduate Institute of Medical Education and Research46, University of Iowa47, University of Melbourne48, University of Sydney49, World Bank50, Anglia Ruskin University51, Danube University Krems52, University of Trnava53, University of Arizona54, German Cancer Research Center55, Technion – Israel Institute of Technology56, University of Liverpool57, University of Leicester58, Oklahoma State University Center for Health Sciences59, Harvard University60, University of Calgary61, City University of New York62, Universidad Autónoma Metropolitana63, University of Perugia64, National University of Colombia65, University of Ottawa66, University of Valencia67, National Taiwan University68, University of Queensland69, National Institutes of Health70, National University of Singapore71, International Institute of Minnesota72, University of Salerno73, Bielefeld University74, Mayo Clinic75, Johns Hopkins University76, University of California, San Diego77, Building and Road Research Institute78, Walden University79, Public Health Foundation of India80, Guy's and St Thomas' NHS Foundation Trust81, Public Health England82, Carol Davila University of Medicine and Pharmacy83, Jacobi Medical Center84, Griffith University85, University of New South Wales86, University of Auckland87, United States Department of Veterans Affairs88, Massey University89, Icahn School of Medicine at Mount Sinai90, University of Toronto91, University of Peradeniya92, Medical University of Varna93, University of Rochester94, University of Tripoli95, Arak University of Medical Sciences96, Federal University of São Paulo97, Academy of Medical Sciences, United Kingdom98, James Cook University99, Monash University100, Stanford University101, SIDI102, Howard University103, University of Burgundy104, University of Massachusetts Boston105, University of British Columbia106, University of Virginia107, Saint James School of Medicine108, University of Bristol109, New York University110, Brandeis University111, Arabian Gulf University112, University of Western Australia113, Franche Comté Électronique Mécanique Thermique et Optique Sciences et Technologies114, University of Barcelona115, Chartered Institute of Management Accountants116, University of Tehran117, Columbia University118, Albert Einstein College of Medicine119, Secretariat of the Pacific Community120, Tianjin University121, George Washington University122, Yale University123, United States Environmental Protection Agency124, Qatar University125, Aarhus University126, University of Tokyo127, American Cancer Society128, George Mason University129, Ghent University130, Washington State University131, Virginia Commonwealth University132, École normale supérieure de Lyon133, China University of Geosciences (Wuhan)134, Université catholique de Louvain135, University of Southern Denmark136, Fudan University137, All India Institute of Medical Sciences138, University of California, San Francisco139, University of Cape Town140, Hebrew University of Jerusalem141, Jordan University of Science and Technology142, Seoul National University143, National Center for Disease Control and Public Health144, Universidade Federal do Rio Grande do Sul145, Northeastern University146, Chungnam National University147, Southern University College148, Simmons College149, University of Canberra150, University of Cincinnati151, Lawrence University152, Turkish Ministry of Health153, Russian Academy154, Oregon Health & Science University155, Utrecht University156, Université de Montréal157, Erasmus University Rotterdam158, Tata Research Development and Design Centre159, University of Helsinki160, Google161, Nova Southeastern University162, Korea University163, University at Albany, SUNY164, Novartis165, Chinese Center for Disease Control and Prevention166, Sichuan University167, Centre national de la recherche scientifique168, Heart and Stroke Foundation of Canada169, Wayne State University170, Box Hill Hospital171, University of Bari172, United States Department of Health and Human Services173, University of São Paulo174, Aintree University Hospitals NHS Foundation Trust175, Shiraz University of Medical Sciences176, University of Zambia177, Baylor University178, Queen Mary University of London179, Paris School of Economics180, University of Paris181, Woolcock Institute of Medical Research182, École Normale Supérieure183, University of the East184, Democratic Republic of the Congo Ministry of Health185, University of York186, University of Pennsylvania187, Mekelle University188, Thomas Jefferson University189, United Nations190, Ifakara Health Institute191, Pacific Institute192, Curtin University193, National University of Malaysia194, University of Papua New Guinea195, Jet Propulsion Laboratory196, Queensland University of Technology197, University of Crete198, Egerton University199, American University of Beirut200, University of Oslo201, New York Medical College202, Ministry of Health and Social Welfare203, Pompeu Fabra University204, Aga Khan University205, Deakin University206, University of Bergen207, Lund University208, Makerere University209, Kyung Hee University210, Teikyo University211, Christian Medical College & Hospital212, Case Western Reserve University213, Kosin University214, University of Pittsburgh215, World Health Organization216, University of London217, University of Porto218, Washington State Department of Health219, Flinders University220, Aalborg University221, James Hutton Institute222, Singapore Ministry of Health223, Environment Agency224, University of Newcastle225, University of Illinois at Chicago226, Tehran University of Medical Sciences227, University of Missouri228, Brown University229, Suez Canal University230, Sao Paulo State University231, University of Lincoln232, International Agency for Research on Cancer233, Muhimbili University of Health and Allied Sciences234, Polytechnic University of Milan235, Symantec236, Marshall University237, University of Oviedo238, University of Zurich239, Microsoft240, University of Maryland, College Park241, Dartmouth College242, University of Alabama at Birmingham243, Stellenbosch University244, University of Colorado Denver245, University of Bath246, Tufts University247, Health Canada248, Finnish Institute of Occupational Health249, Washington State University Spokane250, University of Edinburgh251, Reykjavík University252, Stavanger University Hospital253, University of Western Ontario254, Universiti Tunku Abdul Rahman255, National and Kapodistrian University of Athens256, Ludwig Maximilian University of Munich257, Northwestern University258, University of California, Irvine259, University of Illinois at Urbana–Champaign260, University of Occupational and Environmental Health Japan261, GlaxoSmithKline262, Westchester Medical Center263, Auckland University of Technology264, Federal University of Pernambuco265, Outcomes Research Consortium266, Aristotle University of Thessaloniki267, University of Tennessee Health Science Center268, University of Kinshasa269, Cleveland Clinic270, Royal Institute of Technology271, National Institute for Occupational Safety and Health272, NICTA273, Peking Union Medical College274, Capital Medical University275, BioWare276, Royal Children's Hospital277, University of California, Davis278, University of Nottingham279, University of Gondar280, Royal Cornwall Hospital281, Children's Hospital of Philadelphia282, Duke University283, KAIST284, Abdou Moumouni University285, Mansoura University286, Chongqing University287, Siemens288, Zhejiang University289
Abstract: The Global Burden of Disease, Injuries, and Risk Factor study 2013 (GBD 2013) is the first of a series of annual updates of the GBD. Risk factor quantification, particularly of modifiable risk factors, can help to identify emerging threats to population health and opportunities for prevention. The GBD 2013 provides a timely opportunity to update the comparative risk assessment with new data for exposure, relative risks, and evidence on the appropriate counterfactual risk distribution. Attributable deaths, years of life lost, years lived with disability, and disability-adjusted life-years (DALYs) have been estimated for 79 risks or clusters of risks using the GBD 2010 methods. Risk-outcome pairs meeting explicit evidence criteria were assessed for 188 countries for the period 1990-2013 by age and sex using three inputs: risk exposure, relative risks, and the theoretical minimum risk exposure level (TMREL). Risks are organised into a hierarchy with blocks of behavioural, environmental and occupational, and metabolic risks at the first level of the hierarchy. The next level in the hierarchy includes nine clusters of related risks and two individual risks, with more detail provided at levels 3 and 4 of the hierarchy. Compared with GBD 2010, six new risk factors have been added: handwashing practices, occupational exposure to trichloroethylene, childhood wasting, childhood stunting, unsafe sex, and low glomerular filtration rate. For most risks, data for exposure were synthesised with a Bayesian meta-regression method, DisMod-MR 2.0, or spatial-temporal Gaussian process regression. Relative risks were based on meta-regressions of published cohort and intervention studies. Attributable burden for clusters of risks and all risks combined took into account evidence on the mediation of some risks such as high body-mass index (BMI) through other risks such as high systolic blood pressure and high cholesterol. All risks combined account for 57·2% (95% uncertainty interval [UI] 55·8-58·5) of deaths and 41·6% (40·1-43·0) of DALYs. Risks quantified account for 87·9% (86·5-89·3) of cardiovascular disease DALYs, ranging to a low of 0% for neonatal disorders and neglected tropical diseases and malaria. In terms of global DALYs in 2013, six risks or clusters of risks each caused more than 5% of DALYs: dietary risks accounting for 11·3 million deaths and 241·4 million DALYs, high systolic blood pressure for 10·4 million deaths and 208·1 million DALYs, child and maternal malnutrition for 1·7 million deaths and 176·9 million DALYs, tobacco smoke for 6·1 million deaths and 143·5 million DALYs, air pollution for 5·5 million deaths and 141·5 million DALYs, and high BMI for 4·4 million deaths and 134·0 million DALYs. Risk factor patterns vary across regions and countries and with time. In sub-Saharan Africa, the leading risk factors are child and maternal malnutrition, unsafe sex, and unsafe water, sanitation, and handwashing. In women, in nearly all countries in the Americas, north Africa, and the Middle East, and in many other high-income countries, high BMI is the leading risk factor, with high systolic blood pressure as the leading risk in most of Central and Eastern Europe and south and east Asia. For men, high systolic blood pressure or tobacco use are the leading risks in nearly all high-income countries, in north Africa and the Middle East, Europe, and Asia. For men and women, unsafe sex is the leading risk in a corridor from Kenya to South Africa. Behavioural, environmental and occupational, and metabolic risks can explain half of global mortality and more than one-third of global DALYs providing many opportunities for prevention. Of the larger risks, the attributable burden of high BMI has increased in the past 23 years. In view of the prominence of behavioural risk factors, behavioural and social science research on interventions for these risks should be strengthened. Many prevention and primary care policy options are available now to act on key risks. Bill & Melinda Gates Foundation.
4,851 citations
Abstract: Reliable computer systems must handle malfunctioning components that give conflicting information to different parts of the system. This situation can be expressed abstractly in terms of a group of generals of the Byzantine army camped with their troops around an enemy city. Communicating only by messenger, the generals must agree upon a common battle plan. However, one or more of them may be traitors who will try to confuse the others. The problem is to find an algorithm to ensure that the loyal generals will reach agreement. It is shown that, using only oral messages, this problem is solvable if and only if more than two-thirds of the generals are loyal; so a single traitor can confound two loyal generals. With unforgeable written messages, the problem is solvable for any number of generals and possible traitors. Applications of the solutions to reliable computer systems are then discussed.
4,834 citations
TL;DR: The creation, maintenance, information content and availability of the Cambridge Structural Database (CSD), the world’s repository of small molecule crystal structures, are described.
Abstract: The Cambridge Structural Database (CSD) contains a complete record of all published organic and metal–organic small-molecule crystal structures. The database has been in operation for over 50 years and continues to be the primary means of sharing structural chemistry data and knowledge across disciplines. As well as structures that are made public to support scientific articles, it includes many structures published directly as CSD Communications. All structures are processed both computationally and by expert structural chemistry editors prior to entering the database. A key component of this processing is the reliable association of the chemical identity of the structure studied with the experimental data. This important step helps ensure that data is widely discoverable and readily reusable. Content is further enriched through selective inclusion of additional experimental data. Entries are available to anyone through free CSD community web services. Linking services developed and maintained by the CCDC, combined with the use of standard identifiers, facilitate discovery from other resources. Data can also be accessed through CCDC and third party software applications and through an application programming interface.
4,784 citations
Abstract: We introduce a new representation learning approach for domain adaptation, in which data at training and test time come from similar but different distributions. Our approach is directly inspired by the theory on domain adaptation suggesting that, for effective domain transfer to be achieved, predictions must be made based on features that cannot discriminate between the training (source) and test (target) domains.
The approach implements this idea in the context of neural network architectures that are trained on labeled data from the source domain and unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of features that are (i) discriminative for the main learning task on the source domain and (ii) indiscriminate with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation and stochastic gradient descent, and can thus be implemented with little effort using any of the deep learning packages.
We demonstrate the success of our approach for two distinct classification problems (document sentiment analysis and image classification), where state-of-the-art domain adaptation performance on standard benchmarks is achieved. We also validate the approach for descriptor learning task in the context of person re-identification application.
4,760 citations