• Skip to content (access key 1)
  • Skip to search (access key 7)
FWF — Austrian Science Fund
  • Go to overview page Discover

    • Research Radar
      • Research Radar Archives 1974–1994
    • Discoveries
      • Emmanuelle Charpentier
      • Adrian Constantin
      • Monika Henzinger
      • Ferenc Krausz
      • Wolfgang Lutz
      • Walter Pohl
      • Christa Schleper
      • Elly Tanaka
      • Anton Zeilinger
    • Impact Stories
      • Verena Gassner
      • Wolfgang Lechner
      • Georg Winter
    • scilog Magazine
    • Austrian Science Awards
      • FWF Wittgenstein Awards
      • FWF ASTRA Awards
      • FWF START Awards
      • Award Ceremony
    • excellent=austria
      • Clusters of Excellence
      • Emerging Fields
    • In the Spotlight
      • 40 Years of Erwin Schrödinger Fellowships
      • Quantum Austria
    • Dialogs and Talks
      • think.beyond Summit
    • Knowledge Transfer Events
    • E-Book Library
  • Go to overview page Funding

    • Portfolio
      • excellent=austria
        • Clusters of Excellence
        • Emerging Fields
      • Projects
        • Principal Investigator Projects
        • Principal Investigator Projects International
        • Clinical Research
        • 1000 Ideas
        • Arts-Based Research
        • FWF Wittgenstein Award
      • Careers
        • ESPRIT
        • FWF ASTRA Awards
        • Erwin Schrödinger
        • doc.funds
        • doc.funds.connect
      • Collaborations
        • Specialized Research Groups
        • Special Research Areas
        • Research Groups
        • International – Multilateral Initiatives
        • #ConnectingMinds
      • Communication
        • Top Citizen Science
        • Science Communication
        • Book Publications
        • Digital Publications
        • Open-Access Block Grant
      • Subject-Specific Funding
        • AI Mission Austria
        • Belmont Forum
        • ERA-NET HERA
        • ERA-NET NORFACE
        • ERA-NET QuantERA
        • ERA-NET TRANSCAN
        • Alternative Methods to Animal Testing
        • European Partnership Biodiversa+
        • European Partnership BrainHealth
        • European Partnership ERA4Health
        • European Partnership ERDERA
        • European Partnership EUPAHW
        • European Partnership FutureFoodS
        • European Partnership OHAMR
        • European Partnership PerMed
        • European Partnership Water4All
        • Gottfried and Vera Weiss Award
        • netidee SCIENCE
        • Herzfelder Foundation Projects
        • Quantum Austria
        • Rückenwind Funding Bonus
        • WE&ME Award
        • Zero Emissions Award
      • International Collaborations
        • Belgium/Flanders
        • Germany
        • France
        • Italy/South Tyrol
        • Japan
        • Korea
        • Luxembourg
        • Poland
        • Switzerland
        • Slovenia
        • Taiwan
        • Tyrol–South Tyrol–Trentino
        • Czech Republic
        • Hungary
    • Step by Step
      • Find Funding
      • Submitting Your Application
      • International Peer Review
      • Funding Decisions
      • Carrying out Your Project
      • Closing Your Project
      • Further Information
        • Integrity and Ethics
        • Inclusion
        • Applying from Abroad
        • Personnel Costs
        • PROFI
        • Final Project Reports
        • Final Project Report Survey
    • FAQ
      • Project Phase PROFI
      • Project Phase Ad Personam
      • Expiring Programs
        • Elise Richter and Elise Richter PEEK
        • FWF START Awards
  • Go to overview page About Us

    • Mission Statement
    • FWF Video
    • Values
    • Facts and Figures
    • Annual Report
    • What We Do
      • Research Funding
        • Matching Funds Initiative
      • International Collaborations
      • Studies and Publications
      • Equal Opportunities and Diversity
        • Objectives and Principles
        • Measures
        • Creating Awareness of Bias in the Review Process
        • Terms and Definitions
        • Your Career in Cutting-Edge Research
      • Open Science
        • Open-Access Policy
          • Open-Access Policy for Peer-Reviewed Publications
          • Open-Access Policy for Peer-Reviewed Book Publications
          • Open-Access Policy for Research Data
        • Research Data Management
        • Citizen Science
        • Open Science Infrastructures
        • Open Science Funding
      • Evaluations and Quality Assurance
      • Academic Integrity
      • Science Communication
      • Philanthropy
      • Sustainability
    • History
    • Legal Basis
    • Organization
      • Executive Bodies
        • Executive Board
        • Supervisory Board
        • Assembly of Delegates
        • Scientific Board
        • Juries
      • FWF Office
    • Jobs at FWF
  • Go to overview page News

    • News
    • Press
      • Logos
    • Calendar
      • Post an Event
      • FWF Informational Events
    • Job Openings
      • Enter Job Opening
    • Newsletter
  • Discovering
    what
    matters.

    FWF-Newsletter Press-Newsletter Calendar-Newsletter Job-Newsletter scilog-Newsletter

    SOCIAL MEDIA

    • LinkedIn, external URL, opens in a new window
    • , external URL, opens in a new window
    • Facebook, external URL, opens in a new window
    • Instagram, external URL, opens in a new window
    • YouTube, external URL, opens in a new window

    SCILOG

    • Scilog — The science magazine of the Austrian Science Fund (FWF)
  • elane login, external URL, opens in a new window
  • Scilog external URL, opens in a new window
  • de Wechsle zu Deutsch

  

Resource-Efficient Deep Models for Embedded Systems

Resource-Efficient Deep Models for Embedded Systems

Franz Pernkopf (ORCID: 0000-0002-6356-3367)
  • Grant DOI 10.55776/I2706
  • Funding program Principal Investigator Projects International
  • Status ended
  • Start October 1, 2016
  • End December 31, 2020
  • Funding amount € 214,116
  • Project website

DACH: Österreich - Deutschland - Schweiz

Disciplines

Electrical Engineering, Electronics, Information Engineering (45%); Computer Sciences (55%)

Keywords

    Deep learning, Embedded systems, Resource-efficient machine learning, Machine Learning, Reconfigurable computing, Hardware/software co-design

Abstract Final report

Deep representation learning is one of the main factors for the recent performance boost in many image, signal and speech processing problems. This is particularly true when having big amounts of data and almost unlimited computing resources available as demonstrated in competitions such as for example ImageNet. However, in real-world scenarios the computing infrastructure is often restricted and the computational requirements are not fulfilled. In this research project we suggest several directions for reducing the computational burden, i.e. the number of arithmetic operations, while maintaining the level of recognition performance. This enables to use deep models in mobile devices and embedded systems with limited power-consumption and computational resources.

Deep learning is one of the main ingredients for the recent performance boost in many applications such as image and speech processing. This is particularly true when having big amounts of data and almost unlimited computing resources available. However, in real-world scenarios the computing infrastructure is often restricted and the computational requirements are not fulfilled. In this research project we worked on several directions for reducing the computational burden, i.e. the number of arithmetic operations, while maintaining the level of recognition performance. This enables to use deep models in mobile devices and embedded systems with limited power-consumption and computational resources.

Research institution(s)
  • Technische Universität Graz - 100%
International project participants
  • Paul Chow, University of Toronto - Canada
  • Yoshua Bengio, Université de Montréal - Canada
  • Holger Fröning, Ruprecht-Karls-Universität Heidelberg - Germany
  • Sudhakar Yalamanchili, Georgia Institute of Technology - USA

Research Output

  • 315 Citations
  • 28 Publications
Publications
  • 2022
    Title Lung Sound Classification Using Co-Tuning and Stochastic Normalization
    DOI 10.1109/tbme.2022.3156293
    Type Journal Article
    Author Nguyen T
    Journal IEEE Transactions on Biomedical Engineering
    Pages 2872-2882
    Link Publication
  • 2021
    Title Compressing and Mapping Deep Neural Networks on Edge Computing Systems
    DOI 10.11588/heidok.00030166
    Type Other
    Author Schindler G
    Link Publication
  • 2021
    Title Crackle Detection In Lung Sounds Using Transfer Learning And Multi-Input Convolitional Neural Networks
    DOI 10.48550/arxiv.2104.14921
    Type Preprint
    Author Nguyen T
  • 2021
    Title On Resource-Efficient Bayesian Network Classifiers and Deep Neural Networks
    DOI 10.1109/icpr48806.2021.9413156
    Type Conference Proceeding Abstract
    Author Roth W
    Pages 10297-10304
    Link Publication
  • 2021
    Title Resource-Efficient DNNs for Keyword Spotting using Neural Architecture Search and Quantization
    DOI 10.1109/icpr48806.2021.9413191
    Type Conference Proceeding Abstract
    Author Peter D
    Pages 9273-9279
    Link Publication
  • 2021
    Title End-to-end Keyword Spotting using Neural Architecture Search and Quantization
    DOI 10.48550/arxiv.2104.06666
    Type Preprint
    Author Peter D
  • 2020
    Title Resource-Efficient Speech Mask Estimation for Multi-Channel Speech Enhancement
    DOI 10.48550/arxiv.2007.11477
    Type Preprint
    Author Pfeifenberger L
  • 2020
    Title Differentiable TAN Structure Learning for Bayesian Network Classifiers
    DOI 10.48550/arxiv.2008.09566
    Type Preprint
    Author Roth W
  • 2021
    Title Crackle Detection In Lung Sounds Using Transfer Learning And Multi-Input Convolutional Neural Networks
    DOI 10.1109/embc46164.2021.9630577
    Type Conference Proceeding Abstract
    Author Nguyen T
    Pages 80-83
  • 2021
    Title Lung Sound Classification Using Co-tuning and Stochastic Normalization
    DOI 10.48550/arxiv.2108.01991
    Type Preprint
    Author Nguyen T
  • 2019
    Title Towards Efficient Forward Propagation on Resource-Constrained Systems
    DOI 10.1007/978-3-030-10925-7_26
    Type Book Chapter
    Author Schindler G
    Publisher Springer Nature
    Pages 426-442
  • 2019
    Title Bayesian Learning of Sum-Product Networks
    DOI 10.48550/arxiv.1905.10884
    Type Preprint
    Author Trapp M
  • 2019
    Title Parameterized Structured Pruning for Deep Neural Networks
    DOI 10.48550/arxiv.1906.05180
    Type Preprint
    Author Schindler G
  • 2019
    Title Optimisation of Overparametrized Sum-Product Networks
    DOI 10.48550/arxiv.1905.08196
    Type Preprint
    Author Trapp M
  • 2019
    Title Deep Structured Mixtures of Gaussian Processes
    DOI 10.48550/arxiv.1910.04536
    Type Preprint
    Author Trapp M
  • 2019
    Title Acoustic Scene Classification with Mismatched Devices Using CliqueNets and Mixup Data Augmentation
    DOI 10.21437/interspeech.2019-3002
    Type Conference Proceeding Abstract
    Author Nguyen T
    Pages 2330-2334
  • 2019
    Title Deep Complex-valued Neural Beamformers
    DOI 10.1109/icassp.2019.8683517
    Type Conference Proceeding Abstract
    Author Pfeifenberger L
    Pages 2902-2906
  • 2018
    Title Resource Efficient Deep Eigenvector Beamforming
    DOI 10.1109/icassp.2018.8462503
    Type Conference Proceeding Abstract
    Author Zohrer M
    Pages 3354-3358
  • 2018
    Title Bayesian Neural Networks with Weight Sharing Using Dirichlet Processes
    DOI 10.1109/tpami.2018.2884905
    Type Journal Article
    Author Roth W
    Journal IEEE Transactions on Pattern Analysis and Machine Intelligence
    Pages 246-252
  • 2018
    Title Heart Sound SegmentationAn Event Detection Approach Using Deep Recurrent Neural Networks
    DOI 10.1109/tbme.2018.2843258
    Type Journal Article
    Author Messner E
    Journal IEEE Transactions on Biomedical Engineering
    Pages 1964-1974
  • 2020
    Title Resource-efficient DNNs for Keyword Spotting using Neural Architecture Search and Quantization
    DOI 10.48550/arxiv.2012.10138
    Type Preprint
    Author Peter D
  • 2020
    Title Resource-Efficient Neural Networks for Embedded Systems
    DOI 10.48550/arxiv.2001.03048
    Type Preprint
    Author Roth W
  • 2020
    Title Training Discrete-Valued Neural Networks with Sign Activations Using Weight Distributions
    DOI 10.1007/978-3-030-46147-8_23
    Type Book Chapter
    Author Roth W
    Publisher Springer Nature
    Pages 382-398
  • 2020
    Title On Resource-Efficient Bayesian Network Classifiers and Deep Neural Networks
    DOI 10.48550/arxiv.2010.11773
    Type Preprint
    Author Roth W
  • 2020
    Title Towards Real-Time Single-Channel Singing-Voice Separation with Pruned Multi-Scaled Densenets
    DOI 10.1109/icassp40776.2020.9053542
    Type Conference Proceeding Abstract
    Author Huber M
    Pages 806-810
  • 2018
    Title Hybrid generative-discriminative training of Gaussian mixture models
    DOI 10.1016/j.patrec.2018.06.014
    Type Journal Article
    Author Roth W
    Journal Pattern Recognition Letters
    Pages 131-137
    Link Publication
  • 2018
    Title Learning Deep Mixtures of Gaussian Process Experts Using Sum-Product Networks
    DOI 10.48550/arxiv.1809.04400
    Type Preprint
    Author Trapp M
  • 2018
    Title Efficient and Robust Machine Learning for Real-World Systems
    DOI 10.48550/arxiv.1812.02240
    Type Preprint
    Author Pernkopf F

Discovering
what
matters.

Newsletter

FWF-Newsletter Press-Newsletter Calendar-Newsletter Job-Newsletter scilog-Newsletter

Contact

Austrian Science Fund (FWF)
Georg-Coch-Platz 2
(Entrance Wiesingerstraße 4)
1010 Vienna

office(at)fwf.ac.at
+43 1 505 67 40

General information

  • Job Openings
  • Jobs at FWF
  • Press
  • Philanthropy
  • scilog
  • FWF Office
  • Social Media Directory
  • LinkedIn, external URL, opens in a new window
  • , external URL, opens in a new window
  • Facebook, external URL, opens in a new window
  • Instagram, external URL, opens in a new window
  • YouTube, external URL, opens in a new window
  • Cookies
  • Whistleblowing/Complaints Management
  • Accessibility Statement
  • Data Protection
  • Acknowledgements
  • IFG-Form
  • Social Media Directory
  • © Österreichischer Wissenschaftsfonds FWF
© Österreichischer Wissenschaftsfonds FWF