Hendrik Purwins
AI R&D, Teaching, Implementation
AI for Audio, Quality Monitoring, Text and Document Analytics (NLP), Logistics
Currently I am AI Lead for Germany, Austria, Switzerland and Russia at Accenture Emerging Technologies. I have worked for world leading companies in automotive, aerospace, pharma, entertainment, and banking. Previously I was Associate professor at Audio Analysis Lab and the Sound and Music Computing group at AAU in Copenhagen in August 2013, I have been researcher at Neurotechnology Group and the Machine Learning Group at Technische Universität Berlin where I worked on the discovery of neural correlates of music and 3D vision. Before I had been lecturer at theMusic Technology Group at the Universitat Pompeu Fabra in Barcelona. I have also been head of research and development at PMC Technologies. I have been visiting researcher at Perception and Sound Design Team, IRCAM, CCRMA, Stanford, Auditory Lab, McGill, and visiting professor at UPF. I have obtained my PhD "Profiles of Pitch Classes" on machine learning applied to audio information retrieval at the Neural Information Processing Group at the CS/EE Department at Technische Universität Berlin, receiving a scholarship from the Studienstiftung des deutschen Volkes (by then only awarded to the 0.5 % best students in Germany). Before that I studied mathematics at Universität Bonn and Universität Münster, completing a diploma in pure mathematics (all grades 'A'). Starting with playing the violin at age of 7, I have played several string quartet and orchestra concerts during my mandatory military service with the professional German head quarters chamber orchestra. I have (co-)authored 85+ scientific papers. I am first author of 5 articles with a JCR impact factor of at least 5. (Resume, Publication List).
Publications
Research & Development
In general, I am transferring cutting edge AI/Machine Learning (Deep / Online /Unsupervised /Meta / Few-Shot / Reinforcement Learning) to application domains such as audio, gesture recognition, image recognition, predictive maintenance and quality control in manufacturing, agents in video games, and robotics. In particular I have worked on:
Document reading of forms in the finance domain
Topic modelling for spoken dialogs
Prediction models for supply chain demands
Continuous quality control in automotive
Deep Learning (CNN, GAN, LSTM) for audio recognition and synthesis
An Xception Residual Recurrent Neural Network for Audio Event Detection and Tagging
Estimation of Violin Bowing Features from Audio with CNNs (ML4Audio@NIPS17)
Life-long/ one-shot /unsupervised machine learning for modeling music cognition representation/expectation/generation (paper 1, 2 , Demo Talk, incremental architecture learning)
Neural correlates of music (stimuli:'Depeche Mode') and 3D (tele)vision
Audio information retrieval
Speech processing
(Deep) Reinforcement Learning for Games and Robotics Video: Agent playing MsPackman (student project Reffstrup/Ejlertsen)
Technology for Didactics of the Performing Arts
Sound and music visualization (Ph.D. download)
Sparse coding for sound/music representation/recognition/resynthesis ( paper,code)
Failure prediction in semi conductor production lines ( Paper download)
Mobile sound interaction for children (Demo Video)
Project Management
6/2019- ongoing AI Lead at Accenture: Subject Matter Expert and Managing various teams deployed in automotive, aerospace, banking, and pharma.
3/2010-4/2012 Implementing Manufacturing science solutions to increase equiPment pROductiVity and fab pErformance (Improve): Manager of PMC Technology development team
3/2010-4/2012 Implementing Manufacturing science solutions to increase equiPment pROductiVity and fab pErformance (Improve): Manager of PMC Technology development team
8/2005-12/2007: Emergent Cognition through Active Perception (EmCAP): Co-coordinator of UPF-MTG research team
7/2006-7/2009: Closing the Loop of Sound Evaluation and Design (CLOSED): Co-coordinator of BIT-NI research team (Video)
Industry Collaborations
Other Activities
Lead guest editor for IEEE JSTSP Special Issue on Data Science: Machine Learning for Audio Signal Processing (Open Call)
I have organized the workshop Machine Learning for Audio (ML4Audio) at NIPS 2017 (Proceedings, Slides, Posters)
I am organizing the Sound and Music Computing Colloquium
Teaching
Taught far more than 1000 h on applied machine learning, artificial intelligence, statistics, signal processing, speech processing, machine listening, programming, music cognition, operating systems. Full Teaching List/Statement).
Document Analytics (NLP)
Artificial Intelligence Programming (CNN, RNN, RL)
Machine Learning for Media Technology (parametric/Bayesian/unsupervised/projection methods) and SMC MS)
Sound and Music Signal Analysis (SMC MS, promotion video )
Design and Analysis of Experiment (Media Tecchnology BS)
Ethnograpically Informed Design (Media Technology BS)
Project supervision (RL, Deep Learning, audio/gesture/image recognition) looking for students: Project/thesis proposal catalogue )
Experience with various concepts (flipped classroom: Example Video , peer assessment, problem-based learning)
Contact
Accenture, Glashaus Am Park, Anni-Albers-Straße 11, 80807 München
Email:
hendrik.purwinsATaccentureDOTcom (replace 'DOT' by '.' and 'AT' by '@')
hpurwinsATgmailDOTcom (replace 'DOT' by '.' and 'AT' by '@')
++ 49 1511 14 07 926
Former Masters Students
(at NI, TUB): Immanuel Normann (pagina GmbH),
(at MTG, UPF): Srikanth Cherla (Jukedeck), Katerina Kosta (Jukedeck), Marco Marchini (Spotify), Ahmed Nagi, Maria Panteli (Spotify), Panos Papiotis (MTG, UPF), Pratyush (Radial Inc.), Ines Salselas, Simon Scholler (Newsenselab GmbH), Felix Tutzer ( Nous Wissensmanagement Gmbh)
(at SMC, AAU): Matteo Lionello (University College London), Aleix Claramunt Molet (BMAT), Jose-Luis Diez Antich (Moodagent), Andrea Corcuera Marruffo (Oticon), Jorge Madrid Portillo, Mattia Paterna (Canecto), Nicolai Gahede (Danske Bank), Iakovos Vogiatzoglou (runtales)
(at Medialogy, AAU): Esben Knudsen, Benjamin Ejlertsen (Apple), Jevgenij Martinkevic, Sebastian Siem Bach-Nielsen, Jannik Vilhelm Reffstrup, Sevastiyan Miloslavski Tsvetkov, Krisjanis Urcs, Ernest Bofill Ylla (Sitecore)