“Probabilistic and Unsupervised Machine Learning for Auditory Data and Pattern Recognition”
Prof. Jörg Lücke
Cluster of Excellence Hearing4all and Dept for Medical Physics and Acoustics
Carl von Ossietzky University Oldenburg
I will introduce and discuss probabilistic data models and their applications to auditory data and to general pattern recognition tasks. The introduction will reflect the view that powerful data models are able to extract the true compositional nature of data, which allows for a decomposition into their structural primitives.
Probabilistic sparse coding and probabilistic version of non-negative matrix factorization (NMF) will be the first concrete models that are introduced and discussed. The state-of-the-art of these models will then be used to point to recent generalization directions. Two of these, translation invariant versions and deep generalizations, will be discussed in more detail. I will work out the benefits and challenges such novel approaches face, and I will discuss their crucial differences compared to supervised deep neural networks.
Finally, I briefly discuss semi-supervised approaches, a field where modern unsupervised and modern supervised Machine Learning algorithms come together, compete and where they are combined.
Jörg Lücke since 2013 is an Associate Professor of Machine Learning in Cluster of Excellence Hearing4all and Dept for Medical Physics and Acoustics, Carl von Ossietzky University Oldenburg, Germany and a Guest Professor and Principal Investigator in Dept of Software Engineering and Theoretical Computer Science, Technical University of Berlin, Germany.
In 2008 - 2013 was a Junior Research Group Leader (Computational Neuroscience and Machine Learning),
Frankfurt Institute for Advanced Studies and Dept of Physics,
Goethe-University Frankfurt am Main, Germany.
In 2005 - 2007 was a Senior Research Fellow,
Gatsby Computational Neuroscience Unit,
University College London (UCL), UK.
In 2001 - 2005 was a Research Associate in
Institute for Neuroinformatics,
Ruhr-University Bochum, Germany.
“Forward Error Correction in Wireless Communication Systems for Industrial Applications”
Prof. Petr Pfeifer
Prof. Heinrich Theodor Vierhaus
Computer Engineering Group
Brandenburg University of Technology Cottbus-Senftenberg
Industrial manufacturing is more and more based groups of robots in production cells. The robots consist of moving, bending and rotating arms with multiple joints. Cables that connect sections of robots undergo heavy stress from stretching and twisting, resulting in wear-out and failure. Replacing cables on robots by wireless communication therefore is an alternative that has been investigated for some time. Unfortunately, communication channels in industrial environments suffer from some adversary effect. First, standard industrial communication networks work on rigid time frames which limit allowed latencies in communication systems considerably. Second, multiple path propagation and destructive interferences make such communication channels sensitive to fading problems. Therefore forward error correction (FEC) that can compensate massive variations of signal strength becomes a must. On the other hand, forward error correction using known methods such as BCH codes, Reed-Solomon codes, turbo codes and low-density parity checks (LDPC) is not very fast by nature. Codes for single effort correction and double error detection (SEC-DED-codes) such as Hamming code and Hsiao code are fast, but they are not powerful enough to correct multiple bit errors or restore missing symbols, unless they are applied in a step-wise approximation.
PENCA (programmable encoding architecture) is a new approach in multiple error detection and correction which is, at present based on BCH codes, reasonably fast by parallel hardware. Furthermore, it allows for adaptive error correction, based on the quality of the channel, therefore providing a better overhead / performance ratio than methods that are based on a fixed number of allowed error bits in a symbol, tailored to handle worst-case conditions.
PENCA is currently becoming part of an industrial communication systems developed in the ParSeC project, which is a cooperative effort of industries, universities and research institutes, funded by the German Ministry of Research and Education (BMBF).
Petr Pfeifer has received his Ing. (Msc) in Measurement and Instrumentation from Czech Technical University in Prague in 2001. Then, he was employed in global companies like STMicroelectronics or Tyco International in senior R&D and management positions. He received an MSc in economics and management and Senior Executive MBA degree from The Nottingham Trent University. In 2015, he has received his Ph.D. in technical cybernetics, reliability of nanoscale microelectronic devices from Technical university of Liberec. He is employed by Brandenburg University of Technology (BTU Cottbus-Senftenberg) since 2015. He works in research, design and development of advanced industrial systems, and his interest and professional work cover areas from design and manufacturing of digital VLSI ASIC, advanced systems using field programmable gate arrays and complex programmable logic devices, measurement and control, signal processing, communication, industrial, automotive and safety systems, and reliability aspects of dependable systems using modern submicrometer technologies.
Heinrich Theodor Vierhaus received a diploma degree in electrical engineering from Ruhr-University Bochum (Germany) in 1975 and a doctorate in EE from the University of Siegen in 1983. From 1983 to 1996 he was a senior researcher with GMD, the German national research institute for information technology. Since 1996 he has been a full professor of computer engineering at Brandenburg University of Technology, since 2013 re-founded as BTU Cottbus-Senftenberg.
He has authored more than 100 papers in the area of test, testable design and fault tolerant computing, and he was the General Chair of the IEEE DDES 2011 symposium in Cottbus. He is also the coordinator of the East-Central European network on “Dependable Cyber Physical Systems”, linking 5 universities in 4 countries.
“Test signals used in electroacoustics and speech technology”
Prof. Andrzej Dobrucki
Prof. Stefan Brachmański
Faculty of Electronics
Wroclaw University of Technology
The presentation consists of two parts. In the first one, the typical signals applied for testing of electro-acoustical devices are presented. These signals are e.g: harmonic signal, slowly and quickly tuned sinusoid, impulses, Gaussian noise, Maximum Length Sequence, Golay’s Complementary Sequences. The statistical and spectral properties of these signals are described. The methods of analysis of stationary and nonstationary signals are presented as well. In the second part, the signals used for the evaluation of quality of coded and transmitted speech are presented. For this aim, the natural or artificial speech signals are used. The test material can consist of units having semantic meaning as well as of nonsense syllables or words. For the Polish language, the Polish Standard provides a set of logatome lists. American ANSI norms include, among others, rhyme tests. The International Telecommunication Union (ITU-T) guidelines specify the requirements for the test signal base for telecommunications applications. This recommendation presents a set of test signals of various complexity with many typical speech parameters. These signals are intended for both subjective and objective evaluation of the quality of speech transmission.
Andrzej Dobrucki received the M.Sc. degree in 1971 from the faculty of Electronics of Wroclaw University of Technology, and began his scientific career in electroacoustics at the Institute of Telecommunications and Acoustics. In 1977 he received the Ph.D. degree for a dissertation on vibration and sound radiation of conical shells. In 1993 he received the D.Sc. degree. In 2007 he received the title of full professor from hands of President of Poland. His research interests are in the construction and measurement of electroacoustical transducers, the numerical modeling of acoustic fields, vibrations in mechanical structures and digital processing of audio signals. He is consultant for several companies manufacturing electroacoustical transducers.
He published above 200 scientific works: 5 books, 49 papers in journals, 10 chapters in books, and presentations at the international and local scientific conferences. He is also an author of 6 patents. He was the supervisor in 15 PhD. projects. Prof. Dobrucki is the reviewer in many scientific journals, e.g. Physical Review, Journal of the Acoustical Society of America, Archives of Acoustics, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control.
Andrzej Dobrucki is the co-founder and active member of Polish Section of the Audio Engineering Society (AES). In years 1995-97 and 2003-2007 he held the post of Chairman of the AES Polish Section. In 2007 during the Convention in Vienna, prof. Andrzej Dobrucki was honored with the “Fellowship Award” of the AES. He is also member of the Polish Acoustical Society and he holds the post of President of the Society since 2014.
Andrzej Dobrucki is an experienced academic teacher. He gives the lectures “Physical Acoustics”, “Electroacoustics”, “Electroacoustic Transducers” and “Acoustic Measurements” for the students of Wroclaw University of Technology. For many years he gives the lectures “Modern technologies in hearing aids” for students of the University Paris 12 Val de Marne.
Stefan Brachmański is an Assistant Professor of Speech Communication at the Wroclaw University of Technology, Poland. Graduated and obtained PhD degree in the Wroclaw University of Technology. His research interests includes speech transmission quality evaluation, speech coding, speech enhancement, forensic acoustics and audio restoration. He is a co-author of Polish Standard: Methods for Measurements of Logatom Intelligibility in Analog Communication Systems. He is the author of Project Polish Standard: Methods for Measurements of Logatom Intelligibility in Digital Communication Systems. He is the author and co-author of 179 papers connected with the speech technology and asessment of speech quality He is a member of the Audio Engineering Society, Polish Acoustical Society, Polish Phonetics Association, European Acoustics Association. In the current cadency he is also the Vice-Dean in Faculty of Electronics.
“Image and Video Processing with Tensor Methods”
Prof. Bogusław Cyganek
Department of Electronics
Faculty of Computer Science, Electronics and Telecommunications
AGH University of Science and Technology
Classical methods for processing and analysis of multidimensional signals – such as color videos and hyperspectral images – do not exploit full information contained in inner their factors. On the other hand, recently developed tensor based methods allow for data representation and analysis which directly account for data multidimensionality. Examples can be found in many applications such as face recognition, image synthesis, video analysis, surveillance systems, sensor networks, data stream analysis, marketing and medical data analysis, to name a few.
This talk will be focused on presentation of the basic ideas, as well as recent achievements, in the domain of tensor based signal processing. A systematic overview of tensor data representation, tensor decompositions, as well as pattern recognition with tensors will be presented. Practical aspects and tensor implementation issues will be also discussed.
Bogusław Cyganek received his M.Sc. degree in electronics in 1993, and then M.Sc. in computer science in 1996, from the AGH University of Science and Technology, Krakow, Poland. He obtained his Ph.D. degree cum laude in 2001 with a thesis on correlation of stereo images, and D.Sc. degree in 2011 with a thesis on methods and algorithms of object recognition in digital images.
During recent years dr. Bogusław Cyganek cooperated with many scientific and industrial partners such as Glasgow University Scotland UK, DLR Germany, and Surrey University UK, as well as Nisus Writer, USA, Compression Techniques, USA, Pandora Int., UK, and The Polished Group, Poland. He is an associated professor at the Department of Electronics of the AGH University of Science and Technology, Poland, as well as a visiting professor to the Wroclaw Technical University. His research interests include computer vision, pattern recognition, data mining, as well as development of embedded systems. He is an author or a co-author of over a hundred of conference and journal papers, as well as books with the latest “Object Detection and Recognition in Digital Images: Theory and Practice” published by Wiley in 2013. Dr. Cyganek is a member of the IEEE, IAPR and SPIE.
INVITED SPEAKERS - TUTORIALS GIVEN ON IEEE SPA 2016
“High efficiency video coding”
Prof. Kamisetty Ramamohan Rao
Electrical Engineering Department
The University of Texas at Arlington
In the family of video coding standards, HEVC has the promise and potential to replace/supplement all the existing standards (MPEG and H.26x series including H.264/AVC). While the complexity of the HEVC encoder is several times that of the H.264/AVC, the decoder complexity is within the range of the latter. Researchers are exploring about reducing the HEVC encoder complexity . Kim et al have shown that motion estimation (ME) occupies 77-81% of HEVC encoder implementation. Hence the focus has been in reducing the ME complexity. Several researchers have implemented performance comparison of HEVC with other standards such as H.264/AVC , MPEG-4 Part 2 visual, H.262/PEG-2 Video , H.263, and VP9 and also with image coding standards such as JPEG2000, JPEG-LS, and JPEG-XR. Several tests have shown that HEVC provides improved compression efficiency up to 50% bit rate reduction for the same subjective video quality compared to H.264/AVC . Besides addressing all current applications, HEVC is designed and developed to focus on two key issues: increased video resolution - up to 8kx4k – and increased use of parallel processing architecture. Brief description of the HEVC is provided. However for details and implementation, the reader is referred to the JCT-VC documents , overview papers , keynote speeches , tutorials , panel discussions , poster sessions , special issues , test models (TM/HM) , web/ftp site, open source software , test sequences, anchor bit streams and the latest books on HEVC . Also researchers are exploring transcoding between HEVC and other standards such as MPEG-2 and H.264. Further extensions to HEVC are scalable video coding (SVC), 3D video/multiview video coding and range extensions which include screen content coding (SCC), bit depths larger than 10 bits and color sampling of 4:2:2 and 4:4:4. SCC in general refers to computer generated objects and screen shots from computer applications (both images and videos) and may require lossless coding. Some of these extensions have been finalized by the end of 2014 (time frame for SCC is late 2016). They also provide fertile ground for R & D. Iguchi et al have already developed a hardware encoder for super hi-vision (SHV) i.e., ultra HDTV at 7680x4320 pixel resolution. Also real-time hardware implementation of HEVC encoder for 1080p HD video has been done. NHK is planning SHV experimental broadcasting in 2016. A 249-Mpixel/s HEVC video decoder chip for 4k Ultra-HD applications has already been developed . Bross et al have shown that real time software decoding of 4K (3840x2160) video with HEVC is feasible on current desktop CPUs using four CPU cores. They also state that encoding 4K video in real time on the other hand is a challenge. Multimedia research group (MRC) predicts 2 billion HEVC based devices by end of 2016.
Kamisetty Ramamohan Rao is a full professor of electrical engineering at the University of Texas at Arlington (UT Arlington). He is credited with the co-invention of discrete cosine transform (DCT), along with N. Ahmed and T. Natarajan due to their benchmark publication, “N. Ahmed, T. Natarajan, and K. R. Rao, "Discrete Cosine Transform", IEEE Trans. Computers, 90-93, Jan 1974.”
Dr. Rao received B.S. E.E from the College of Engineering, Guindy, affiliated to The University of Madras, India in 1952. In 1959, he received his M.S.E.E degree from University of Florida followed by an M.S.NuE from the University of Florida in 1960. He received the Ph. D. degree in Electrical Engineering from the University of New Mexico in 1966.
In 2011, Dr Rao reached an academic milestone, supervising his 100th Graduate Student.
K. R. Rao received the Ph. D. degree in electrical engineering from The University of New Mexico, Albuquerque in 1966. He is now working as a professor of electrical engineering in the University of Texas at Arlington, Texas. He has published (coauthored) 16 books, some of which have been translated into Chinese, Japanese, Korean and Russian. Also as ebooks and paper back (Asian) editions. He has supervised 87 Masters and 31 doctoral students. He has published extensively and conducted tutorials/workshops worldwide. He has been a consultant to academia, industry and research institutes.
“Computational models for predicting sound quality”
Prof. Brian C.J. Moore
Department of Experimental Psychology
University of Cambridge
Downing Street, Cambridge CB2 3EB, England
The quality of an audio device, such as a microphone, amplifier, or headphone, depends on how accurately the device transmits the properties of the sound source to the ear(s) of the listener. Two types of “distortion” can occur in this transmission: (1) “Linear” distortion, which may be described as a deviation of the frequency response from the “target” response; (2) Nonlinear distortion, which is characterised by frequency components in the output of the device that were not present in the input. These two forms of distortion have different perceptual effects. Their effects on sound quality can be predicted using a model of auditory processing with the following stages: (1) A filter to take into account the transmission of sound from the device to the ear of the listener; (2) A filter to simulate the effects of transmission through the middle ear; (3) An array of bandpass filters to simulate the auditory filters that exist in the cochlea of the inner ear. For predicting the perceptual effects of linear distortion, a model operating in the frequency domain can be used. For predicting the perceptual effects of nonlinear distortion, a model operating in the time domain is required, since the detailed waveforms at the outputs of the auditory filters need to be considered. The models described have been shown to give accurate predictions for a wide range of “artificial” and “real” linear and nonlinear distortions.
Brian Moore is Emeritus Professor of Auditory Perception in the University of Cambridge. His research interests are: the perception of sound in normal and impaired hearing; design of signal processing hearing aids for sensorineural hearing loss; methods for fitting hearing aids to the individual; perception of music and of musical instruments. He is a Fellow of the Royal Society, the Academy of Medical Sciences, the Acoustical Society of America, The Audio Engineering Society, and the Association for Psychological Science, and an Honorary Fellow of the Belgian Society of Audiology and the British Society of Hearing Aid Audiologists. He is President of the Association of Independent Hearing Healthcare Professionals (UK). He has written or edited 20 books and over 640 scientific papers and book chapters. He has been awarded the Littler Prize and the Littler Lecture of the British Society of Audiology, the Silver and Gold medals of the Acoustical Society of America, the first International Award in Hearing from the American Academy of Audiology, the Award of Merit from the Association for Research in Otolaryngology, the Hugh Knowles Prize for Distinguished Achievement from Northwestern University and an honorary doctorate from Adam Mickiewicz University, Poland. He is wine steward of Wolfson College, Cambridge.
“The crucial role of mathematics in circuits, systems and signal processing research and education”
Prof. Joos Vandewalle
Katholieke Universiteit Leuven
Electrical Engineering Department - ESAT, Stadius Division
Leuven, Flanders, Belgium
Over the recent years the role of mathematics in innovations for Circuits, Systems and Signal processing has increased considerably. The talk will overview the dynamical forces and their impact on the research and education. Examples will be given of mathematical methodologies for signal and image classification, data fusion, biomedical diagnostics with support vector machines, matrix and tensor decompositions. Also cryptographic algorithms are crucial in our modern society. Important lessons can be learned for research planning, dissemination and reproducibility as well as for teaching in engineering.
Joos Vandewalle obtained the electrical engineering degree and doctorate in applied sciences from KU Leuven, Belgium in 1971 and 1976. Until October 2013 he was a full professor at the Department Electrical Engineering (ESAT), Katholieke Universiteit Leuven, Belgium; head of the SCD division at ESAT, with more than 150 researchers. Since October 2013 he is a professor emeritus with assignments at KU Leuven. His present tasks include chairing the positioning test for engineering in Flanders, board member of the Flemish Academy in Brussels, chairing PhD defenses, …
He held visiting positions University of California, Berkeley and I3S CNRS Sophia Antipolis, France.
He taught courses in linear algebra, linear and nonlinear system and circuit theory, signal processing and neural networks. His research interests are in mathematical system theory and its applications in circuit theory, control, signal processing, cryptography and neural networks. He (co-)authored more than 300 international journal papers and obtained several best paper awards and research awards and in 2016 the IEEE CAS Desoer Technical Achievement Award. His publications received over 30 000 googlescholar citations. He is a Fellow of IEEE, IET, and EURASIP and member of the Academia Europaea and of the Belgian Academy of Sciences. From 2009 till 2013 he was a member of the Board of Governors of the IEEE Circuits and Systems Society. He is a member of the Fetzer Advisory Council on Engineering and Chair of the IEEE CAS Circuits and Systems Education and Outreach TC, and Chairman of the class of Technical sciences of the Belgian Academy KVAB.