检索结果-内蒙古大学图书馆

SEQUENCE MODELING IN UNSUPERVISED SINGLE-CHANNEL OVERLAPPED SPEECH RECOGNITION

学校读者我要写书评

暂无评论

SEQUENCE MODELING IN UNSUPERVISED SINGLE-CHANNEL OVERLAPPED ...

IEEE International Conference on Acoustics, Speech and Signal Processing

作者： Zhehuai Chen Jasha Droppo Key Lab. of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering SpeechLab Department of Computer Science and Engineering Shanghai Jiao Tong University China Microsoft AI and Research Redmond WA USA

ISBN: (纸本)9781538646595

Unsupervised single-channel overlapped speech recognition is one of the hardest problems in automatic speech recognition (ASR). The problems can be modularized into three sub-problems: frame-wise interpreting, sequence level speaker tracing and speech recognition. Nevertheless, previous acoustic models formulate the correlation between sequential labels implicitly, which limit the modeling effect. In this work, we include explicit models for the sequential label correlation during training. This is relevant to models given by both the feature sequence and the output of the last frame. Moreover, we propose to integrate the linguistic information into the assignment decision of the permutation invariant training (PIT). Namely, a senone level neural network language model (NNLM) trained in the clean speech alignment is integrated, while the objective function is still cross-entropy. The proposed methods can be combined with an improved version of PIT and sequence discriminative training, which brings about further over 10% relative improvement of WER in the artificial overlapped Switchboard and hub5e-swb dataset.

关键词： unsupervised single channel overlapped speech recognition permutation invariant training temporal correlation modeling language model Speech recognition modelling languages linguistic data Modeling Objective function

The SRG/eROSITA All-Sky Survey: Dark Energy Survey Year 3 Weak Gravitational Lensing by eRASS1 selected Galaxy Clusters

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Grandis, S. Ghirardini, V. Bocquet, S. Garrel, C. Mohr, J.J. Liu, A. Kluge, M. Kimmig, L. Reiprich, T.H. Alarcon, A. Amon, A. Artis, E. Bahar, Y.E. Balzer, F. Bechtol, K. Becker, M.R. Bernstein, G. Bulbul, E. Campos, A. Carnero Rosell, A. Carrasco Kind, M. Cawthon, R. Chang, C. Chen, R. Chiu, I. Choi, A. Clerc, N. Comparat, J. Cordero, J. Davis, C. Derose, J. Diehl, H.T. Dodelson, S. Doux, C. Drlica-Wagner, A. Eckert, K. Elvin-Poole, J. Everett, S. Ferte, A. Gatti, M. Giannini, G. Giles, P. Gruen, D. Gruendl, R.A. Harrison, I. Hartley, W.G. Herner, K. Huff, E.M. Kleinebreil, F. Kuropatkin, N. Leget, P.F. Maccrann, N. McCullough, J. Merloni, A. Myles, J. Nandra, K. Navarro-Alsina, A. Okabe, N. Pacaud, F. Pandey, S. Prat, J. Predehl, P. Ramos, M. Raveri, M. Rollins, R.P. Roodman, A. Ross, A.J. Rykoff, E.S. Sanchez, C. Sanders, J. Schrabback, T. Secco, L.F. Seppi, R. Sevilla-Noarbe, I. Sheldon, E. Shin, T. Troxel, M. Tutusaus, I. Varga, T.N. Wu, H. Yanny, B. Yin, B. Zhang, X. Zhang, Y. Alves, O. Bhargava, S. Brooks, D. Burke, D.L. Carretero, J. Costanzi, M. da Costa, L.N. Pereira, M.E.S. De Vicente, J. Desai, S. Doel, P. Ferrero, I. Flaugher, B. Friedel, D. Frieman, J. García-Bellido, J. Gutierrez, G. Hinton, S.R. Hollowood, D.L. Honscheid, K. James, D.J. Jeffrey, N. Lahav, O. Lee, S. Marshall, J.L. Menanteau, F. Ogando, R.L.C. Pieres, A. Plazas Malagón, A.A. Romer, A.K. Sanchez, E. Schubnell, M. Smith, M. Suchyta, E. Swanson, M.E.C. Tarle, G. Weaverdyck, N. Weller, J. Universität Innsbruck Institut für Astro- und Teilchenphysik Technikerstr. 25/8 Innsbruck6020 Austria Max Planck Institute for Extraterrestrial Physics Giessenbachstrasse 1 Garching85748 Germany Faculty of Physics Ludwig-Maximilians-Universität München Scheinerstr. 1 Munich81679 Germany Universität Bonn Auf dem Hügel 71 Bonn53121 Germany Kavli Institute for Cosmology University of Cambridge Mading- ley Road CambridgeCB3 0HA United Kingdom Institute of Astronomy University of Cambridge Madingley Road CambridgeCB3 0HA United Kingdom Physics Department University of Wisconsin-Madison MadisonWI53706 United States Argonne National Laboratory 9700 South Cass Avenue LemontIL60439 United States Department of Physics and Astronomy University of Pennsylvania PhiladelphiaPA19104 United States Department of Physics Carnegie Mellon University PittsburghPA15312 United States Instituto de Astrofisica de Canarias Tenerife La LagunaE-38205 Spain Laboratório Interinstitucional de e-Astronomia - LIneA Rua Gal. José Cristino 77 RJ Rio de Janeiro20921-400 Brazil Universidad de La Laguna Dpto. Astrofísica Tenerife La LagunaE-38206 Spain Center for Astrophysical Surveys National Center for Supercom- puting Applications 1205 West Clark St. UrbanaIL61801 United States Department of Astronomy University of Illinois at Urbana-Champaign 1002 W. Green Street UrbanaIL61801 United States Physics Department William Jewell College LibertyMO64068 United States Department of Astronomy and Astrophysics University of Chicago ChicagoIL60637 United States Kavli Institute for Cosmological Physics University of Chicago ChicagoIL60637 United States Department of Physics Duke University DurhamNC27708 United States Department of Physics National Cheng Kung University Tainan70101 Taiwan NASA Goddard Space Flight Center 8800 Greenbelt Rd Green- belt MD20771 United States IRAP Université de Toulouse CNRS UPS CNES Toulouse France Jodrell Bank Center f

Context. Number counts of galaxy clusters across redshift are a powerful cosmological probe, if a precise and accurate reconstruction of the underlying mass distribution is performed - a challenge called mass calibration. With the advent of wide and deep photometric surveys, weak gravitational lensing by clusters has become the method of choice to perform this measurement. aims. We measure and validate the weak gravitational lensing (WL) signature in the shape of galaxies observed in the first 3 years of the Dark Energy Survey (DES Y3) caused by galaxy clusters and groups selected in the first all-sky survey performed by SRG (Spectrum Roentgen Gamma)/eROSITA (eRASS1). These data are then used to determine the scaling between X-ray photon count rate of the clusters and their halo mass and redshift. Methods. We empirically determine the degree of cluster member contamination in our background source sample. The individual cluster shear profiles are then analysed with a Bayesian population model that self-consistently accounts for the lens sample selection and contamination, and includes marginalization over a host of instrumental and astrophysical systematics. To quantify the accuracy of the mass extraction of that model, we perform mass measurements on mock cluster catalogs with realistic synthetic shear profiles. This allows us to establish that hydro-dynamical modelling uncertainties at low lens redshifts (z © 2024, CC BY.

关键词： Calibration

Opportunities and challenges for monitoring terrestrial biodiversity in the robotics age

学校读者我要写书评

暂无评论

Nature ecology & evolution 2025年 2025 May 22页

作者： Stephen Pringle Martin Dallimer Mark A Goddard Léni K Le Goff Emma Hart Simon J Langdale Jessica C Fisher Sara-Adela Abad Marc Ancrenaz Fabio Angeoletto Fernando Auat Cheein Gail E Austen Joseph J Bailey Katherine C R Baldock Lindsay F Banin Cristina Banks-Leite Aliyu S Barau Reshu Bashyal Adam J Bates Jake E Bicknell Jon Bielby Petra Bosilj Emma R Bush Simon J Butler Dan Carpenter Christopher F Clements Antoine Cully Kendi F Davies Nicolas J Deere Michael Dodd Rosie Drinkwater Don A Driscoll Guillaume Dutilleux Mads Dyrmann David P Edwards Mohammad S Farhadinia aisyah Faruk Richard Field Robert J Fletcher Chris W Foster Richard Fox Richard M Francksen Aldina M A Franco Alison M Gainsbury Charlie J Gardner Ioanna Giorgi Richard A Griffiths Salua Hamaza Marc Hanheide Matt W Hayward Marcus Hedblom Thorunn Helgason Sui P Heon Kevin A Hughes Edmund R Hunt Daniel J Ingram George Jackson-Mills Kelly Jowett Timothy H Keitt Laura N Kloepper Stephanie Kramer-Schadt Jim Labisko Frédéric Labrosse Jenna Lawson Nicolas Lecomte Ricardo F de Lima Nick A Littlewood Harry H Marshall Giovanni L Masala Lindsay C Maskell Eleni Matechou Barbara Mazzolai Alistair McConnell Brett A Melbourne Aslan Miriyev Eric Djomo Nana Alessandro Ossola Sarah Papworth Catherine L Parr Ana Payo-Payo Gad Perry Nathalie Pettorelli Rajeev Pillay Simon G Potts Miranda T Prendergast-Miller Lan Qie Persie Rolley-Parnell Stephen J Rossiter Marcus Rowcliffe Heather Rumble Jon P Sadler Christopher J Sandom Asiem Sanyal Franziska Schrodt Sarab S Sethi Adi Shabrani Robert Siddall Simón C Smith Robbert P H Snep Carl D Soulsbury Margaret C Stanley Philip A Stephens P J Stephenson Matthew J Struebig Matthew Studley Martin Svátek Gilbert Tang Nicholas K Taylor Kate D L Umbers Robert J Ward Patrick J C White Mark J Whittingham Serge Wich Christopher D Williams Ibrahim B Yakubu Natalie Yoh Syed A R Zaidi Anna Zmarz Joeri A Zwerts Zoe G Davies Durrell Institute of Conservation and Ecology (DICE) School of Natural Sciences University of Kent Canterbury UK. s.pringle@kent.ac.uk. Centre for Environmental Policy Imperial College London London UK. Department of Geography and Environmental Sciences Northumbria University Newcastle upon Tyne UK. School of Computing Engineering and the Built Environment Edinburgh Napier University Edinburgh UK. Synthotech Ltd Milner Court Hornbeam Square Harrogate UK. Durrell Institute of Conservation and Ecology (DICE) School of Natural Sciences University of Kent Canterbury UK. Mechanical Engineering Department University College London London UK. HUTAN SWD Kota Kinabalu Malaysia. Programa de Pós-Graduação em Gestão e Tecnologia Ambiental da Universidade Federal de Rondonópolis Rondonópolis Brasil. Department of Engineering Harper Adams University Newport UK. Applied Ecology Research Group School of Life Sciences Anglia Ruskin University Cambridge UK. Centre for Ecology and Hydrology Penicuik UK. Department of Life Sciences Imperial College London Silwood Park Campus Ascot UK. Department of Urban and Regional Planning Faculty of Earth and Environmental Sciences Bayero University Kano Nigeria. Greenhood Nepal Kathmandu Nepal. Animal Rural & Environmental Sciences Nottingham Trent University Nottinghamshire UK. School of Biological and Environmental Sciences Liverpool John Moores University Liverpool UK. Lincoln Centre for Autonomous Systems University of Lincoln Lincoln UK. Royal Botanic Garden Edinburgh Edinburgh UK. School of Biological Sciences University of East Anglia Norwich Research Park Norwich UK. Digital Ecology Limited Bristol UK. School of Biological Sciences University of Bristol Bristol UK. Department of Computing Imperial College London London UK. Department of Ecology and Evolutionary Biology University of Colorado Boulder CO USA. Faculty of Science Technology Engineering and Mathematics (STEM) The Open University Walton Hall Milt

With biodiversity loss escalating globally, a step change is needed in our capacity to accurately monitor species populations across ecosystems. Robotic and autonomous systems (RAS) offer technological solutions that may substantially advance terrestrial biodiversity monitoring, but this potential is yet to be considered systematically. We used a modified Delphi technique to synthesize knowledge from 98 biodiversity experts and 31 RAS experts, who identified the major methodological barriers that currently hinder monitoring, and explored the opportunities and challenges that RAS offer in overcoming these barriers. Biodiversity experts identified four barrier categories: site access, species and individual identification, data handling and storage, and power and network availability. Robotics experts highlighted technologies that could overcome these barriers and identified the developments needed to facilitate RAS-based autonomous biodiversity monitoring. Some existing RAS could be optimized relatively easily to survey species but would require development to be suitable for monitoring of more 'difficult' taxa and robust enough to work under uncontrolled conditions within ecosystems. Other nascent technologies (for instance, new sensors and biodegradable robots) need accelerated research. Overall, it was felt that RAS could lead to major progress in monitoring of terrestrial biodiversity by supplementing rather than supplanting existing methods. Transdisciplinarity needs to be fostered between biodiversity and RAS experts so that future ideas and technologies can be codeveloped effectively.

关键词：

Invocation-driven neural approximate computing with a multiclass-classifier and multiple approximators

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Song, Haiyue Xu, Chengwen Xu, Qiang Song, Zhuoran Jing, Naifeng Liang, Xiaoyao Jiang, Li MoE Key Lab of Artificial Intelligence AI Institute School of Electronic Information and Electrical Engineering Shanghai Jiao Tong University China Department of Computer Science and Engineering Chinese University of Hong Kong Hong Kong

Neural approximate computing gains enormous energy-efficiency at the cost of tolerable quality-loss. A neural approximator can map the input data to output while a classifier determines whether the input data are safe to approximate with quality guarantee. However, existing works cannot maximize the invocation of the approximator, resulting in limited speedup and energy saving. By exploring the mapping space of those target functions, in this paper, we observe a nonuniform distribution of the approximation error incurred by the same approximator. We thus propose a novel approximate computing architecture with a Multiclass-Classifier and Multiple Approximators (MCMA). These approximators have identica network topologies, and thus can share the same hardware resource in an neural processing unit(NPU) clip. In the runtime, MCMA can swap in the invoked approximator by merely shipping the synapse weights from the on-chip memory to the buffers near MAC within a cycle. We also propose efficient co-training methods for such MCMA architecture. Experimental results show a more substantial invocation of MCMA as well as the gain of energy-efficiency. Copyright © 2018, The Authors. All rights reserved.

关键词： Input output programs

Varying control intensity of synchronized chaotic system with time delay

学校读者我要写书评

暂无评论

Journal of Physics: Conference Series 2021年第1期1828卷

作者： Babatunde O. Onasanya Shiping Wen Yuming Feng Wei Zhang Ning Tang Adeleke T. Ademola Key Laboratory of Intelligent Information Processing and Control Chongqing Three Gorges University Wanzhou Chongqing 404100 China Department of Mathematics University of Ibadan Ibadan Nigeria Australian AI Institute University of Technology Sydney Ultimo NSW 2007 Australia School of Computer Science and Engineering Chongqing Three Gorges University Wanzhou Chongqing 404100 China Chongqing Engineering Research Center of Internet of Things and Intelligent Control Technology Chongqing Three Gorges University Wanzhou Chongqing 404100 China School of Three Gorges Artificial Intelligence Chongqing Three Gorges University Wanzhou Chongqing 404100 China Department of Mathematics Obafemi Awolowo University Post Code 220005 Ile-Ife Nigeria

Hyperchaotic system is a very useful tool in secure and encrypted communications. But situations arise when engineers and scientists seek to synchronize two hyperchaotic systems. This gives another (error) system. The goal is to minimize the error as much as can be in order to make one system look like the other by synchronization. This is a particularly challenging situation. In this paper, two hyperchaotic systems are synchronized by impulsive control. Also, the condition for uniform asymptotic stability of the synchronized error system was given. Finally, the simulation results to justify the reliability of this method is also presented.

关键词：

Legible normativity for ai alignment: The value of silly rules

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Hadfield-Menell, Dylan Andrus, McKane Hadfield, Gillian K. Department of Electrical Engineering and Computer Science University of California Berkeley Center for Human Compatible AI Faculty of Law and Rotman School of Management University of Toronto Vector Institute for Artificial Intelligence OpenAI

It has become commonplace to assert that autonomous agents will have to be built to follow human rules of behavior-social norms and laws. But human laws and norms are complex and culturally varied systems;in many cases agents will have to learn the rules. This requires autonomous agents to have models of how human rule systems work so that they can make reliable predictions about rules. In this paper we contribute to the building of such models by analyzing an overlooked distinction between important rules and what we call silly rules -rules with no discernible direct impact on welfare. We show that silly rules render a normative system both more robust and more adaptable in response to shocks to perceived stability. They make normativity more legible for humans, and can increase legibility for ai systems as well. For ai systems to integrate into human normative systems, we suggest, it may be important for them to have models that include representations of silly rules. Copyright © 2018, The Authors. All rights reserved.

关键词： Autonomous agents

Information geometry of orthogonal initializations and training

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Sokól, Piotr Aleksander Park, Il Memming Department of Neurobiology and Behavior Departments of Applied Mathematics and Statistics Electrical and Computer Engineering Institutes for Advanced Computing Science and AI-driven Discovery and Innovation Stony Brook University Stony BrookNY11733

Recently mean field theory has been successfully used to analyze properties of wide, random neural networks. It gave rise to a prescriptive theory for initializing feed-forward neural networks with orthogonal weights, which ensures that both the forward propagated activations and the backpropagated gradients are near `2 isometries and as a consequence training is orders of magnitude faster. Despite strong empirical performance, the mechanisms by which critical initializations confer an advantage in the optimization of deep neural networks are poorly understood. Here we show a novel connection between the maximum curvature of the optimization landscape (gradient smoothness) as measured by the Fisher information matrix (FIM) and the spectral radius of the input-output Jacobian, which partially explains why more isometric networks can train much faster. Furthermore, given that orthogonal weights are necessary to ensure that gradient norms are approximately preserved at initialization, we experimentally investigate the benefits of maintaining orthogonality throughout training, from which we conclude that manifold optimization of weights performs well regardless of the smoothness of the gradients. Moreover, motivated by experimental results we show that a low condition number of the FIM is not predictive of faster learning. Copyright © 2018, The Authors. All rights reserved.

关键词： Fisher information matrix

Deep RNN framework for visual sequential applications

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Pang, Bo Zha, Kaiwen Cao, Hanwen Shi, Chen Lu, Cewu Shanghai Jiao Tong University China Department of Computer Science and Engineering Shanghai Jiao Tong University MoE Key Lab of Artificial Intelligence AI Institute Shanghai Jiao Tong University China

Extracting temporal and representation features efficiently plays a pivotal role in understanding visual sequence information. To deal with this, we propose a new recurrent neural framework that can be stacked deep effectively. There are mainly two novel designs in our deep RNN framework: one is a new RNN module called Context Bridge Module (CBM) which splits the information flowing along the sequence (temporal direction) and along depth (spatial representation direction), making it easier to train when building deep by balancing these two directions;the other is the Overlap Coherence Training Scheme that reduces the training complexity for long visual sequential tasks on account of the limitation of computing resources. We provide empirical evidence to show that our deep RNN framework is easy to optimize and can gain accuracy from the increased depth on several visual sequence problems. On these tasks, we evaluate our deep RNN framework with 15 layers, 7× than conventional RNN networks, but it is still easy to train. Our deep framework achieves more than 11% relative improvements over shallow RNN models on Kinetics, UCF-101, and HMDB-51 for video classification. For auxiliary annotation, after replacing the shallow RNN part of Polygon-RNN with our 15-layer deep CBM, the performance improves by 14.7%. For video future prediction, our deep RNN improves the state-of-the-art shallow model’s performance by 2.4% on PSNR and SSIM. The code and trained models are published accompanied by this paper: https://***/BoPang1996/Deep-RNN-Framework. Copyright © 2018, The Authors. All rights reserved.

关键词： Machine learning

Predictive Analytics on Facebook Data 2

学校读者我要写书评

暂无评论

Predictive Analytics on Facebook Data

2nd World Congress on Computing and Communication Technologies, WCCCT 2017

作者： Adarsh Samuel Raj, J. Lenin Fernando, J. Sunil Raj, Y. Department of Computer Science and Engineering Sastra University Thanjavur India Department of Information Technology AI Musanna College of Technology Sultanate of Oman Oman Oman Tiruchirappalli India

ISBN: (纸本)9781509055739

Data in every organization are increasing at a rapid speed along with the volume of it increases at a large extend through various domains such as education, social network, meteorology, government and much more. Accordingly, Big Data refers to data as traditional data, Machine-generated Sensor data and Social data which are both structured and unstructured. Apache's Hadoop has proven to provide salient properties such as scalability, ease of use, and most notably robustness to node failures. Processing K-Nearest Neighbor queries in high dimensional data has received a lot of attention by researchers in recent years, which provides a way of predicting in a better way as done in this analysis. This research analysis is to find the means to enhance the classification accuracy, utilizing logistic regression and K-Nearest Neighbour a fast process technique in a machine learning technique for classification. The work aims at enhancing the rate of accuracy, true positive, false positive, precision, recall, sensitivity, specificity rate of classification using the HADOOP platform. As a source for analysis, political values are collected from social network users. The Prediction with K-NN neighbor approach is helpful to predict the patterns using opinion. © 2017 IEEE.

关键词： Big data