Erasmus+ Info Veranstaltung für Studierende an der TF

Technische Fakultät organisiert erneut Siegerehrung des Landeswettbewerbs „Jugend forscht – Schüler experimentieren“

Sonderkolloquium, Jun.-Prof. Dr. Mattias Heinrich, Uni Lübeck / am 07.07.2017

07.07.2017 von 14:15 bis 15:45

Institut für Informatik,Christian-Albrechts-Platz 4, R.715, 24114 Kiel

Titel: Learning Sparse Binary Features for Medical Image Segmentation of the Abdomen

Abstract: In this talk, we explore the capabilities of sparse binary features for medical image segmentation. Due to insufficient contrast and anatomical shape variations local image patches rarely provide sufficient information for accurate segmentation of abdominal structures. Based on our two recent MICCAI papers, we propose to use long-range binary features to robustly capture the image context. Two different classification strategies are subsequently developed. 

First, a very fast approximate nearest neighbour search based on vantage point forests and Hamming distances between feature strings is presented. The classifier can be learned and applied to new data in few seconds. The approach reaches state-of-the-art performance for larger organs on the VISCERAL3 benchmark.

Second, we develop a deep neural network architecture that combines a local CNN path with a new contextual path that encodes the sparse binary features. Following the ideas from Network-in-Network, 1x1 convolutions are employed to learn the best combination of different binary offset locations. We demonstrate experimentally that this restricted feature extraction in the first layer enables to regularise the network with a huge receptive field and leads to short training times of less than 10 minutes. Using only 1 million trainable parameters, the model achieves a accuracy of 64.5% Dice, which is comparable to the best performing, much more complex deep CNN approach for pancreas segmentation.

Finally, the potential use of learned binary features for other tasks in medical image analysis, such as image registration and disease classification will be discussed.

Prof. Carsten Meyer

Diesen Termin meinem iCal-Kalender hinzufügen

zurück

Kolloquiumsvortrag (ET&IT), Prof. Elisabetta Chicca, Uni Bielefeld / am 13.11.2017

13.11.2017 von 17:15 bis 18:15

Institute Ostufer, Geb. D, "Aquarium", Kaiserstr. 2, 24143 Kiel

Titel: Learning in silico beyond STDP

Abstract: Synaptic plasticity empowers biological nervous systems with the ability to learn from experience and adjust to environmental changes. Such abilities are a must for artificial autonomous systems and therefore researchers have been devoting significant efforts to the understanding and modelling of plasticity mechanisms. In particular, the field of neuromorphic engineering focuses on the development of full-custom hybrid analog/digital electronic systems for the implementation of models of biological computation and learning in hardware. I will give a short historical overview of the most important plasticity circuits developed following the approach originally proposed by Carver Mead in the late eighties. Afterwards, I will present recent advancements in this field.

zur Person: Elisabetta Chicca hat einen "Laurea"-Abschluss (M.Sc.) in Physik von der Universität Rom 1 ("La Sapienza", Italien 1999) einen Ph.D. in Naturwissenschaften von der Eidgenössischen Technischen Hochschule Zürich (ETHZ, Abteilung Physik) und einen Ph.D. in Neurowissenschaften von dem neurowissenschaftlichen Zentrum Zürich (beides 2006).

 Prof. Chicca forschte als Postdoktorandin (2006-2010) und als Gruppenleiterin (2010-2011) am Institut für Neuroinformatik (Universität Zürich und ETH Zürich) an der Entwicklung von Neuromorphen Signalverarbeitungs- und sensorischen Systemen. Seit 2011 leitet sie die Gruppe Neuromorphic Behaving Systems an der Universität Bielefeld (Faculty of Technology and Cognitive Interaction Technology Center of Excellence, CITEC). Ihre aktuellen Interessen liegen in der Entwicklung von VLSI-Modellen kortikaler Schaltkreise für Gehirn-inspirierte Schaltungen, Lernen in spiking VLSI-neuronalen Netzen und Memristor basierten Systemen, bioinspirierter Wahrnehmung (Olfaktion, aktive Elektrolokalisierung, Hörvermögen) und Motorsteuerung. Die von ihr verwendeten Verfahren zur Herstellung kundenspezifischer Schaltungen (ASICs), sind solche, die auch von unseren neuen Kollegen Andreas Bahr und Robert Rieger genutzt werden.

 In ihrem Vortrag wird Elisabetta Chicca einen kurzen historischen Überblick über die wichtigsten neuromorphen Systeme geben, die nach dem Ansatz von Carver Mead in den späten achtziger Jahren entwickelt wurden. Danach wird sie die neuesten Fortschritte in diesem Bereich vorstellen.

PD Martin Ziegler

Diesen Termin meinem iCal-Kalender hinzufügen

zurück

Kolloquiumsvortrag (ET&IT), M.Sc. Jonas Sauter, Nuance Communications / am 20.11.2017

20.11.2017 von 17:00 bis 18:00

Institute Ostufer, Geb. D, "Aquarium", Kaiserstr. 2, 24143 Kiel

Titel: Artificial Bandwidth Extension for Speech Signals Using Deep Neural Networks

Abstract: In mobile communication, the bandwidth of transferred speech signals is either narrow-band (300Hz – 3.4kHz) or wide-band (50Hz – 7kHz or higher). As the limitation to 3.4kHz degrades the speech quality and intelligibility, it is of great interest to artificially extend narrow-band speech signals to wide-band speech.

This talk presents a deep neural network (DNN) approach to artificial bandwidth extension with a focus on robustness in practical applications.

It is based on the source-filter model which decomposes the signal into two parts: an excitation signal and a spectral envelope. The excitation (source part) describes the fine spectral structure which consists of white noise for unvoiced speech and an impulse train for voiced speech. The spectral envelope (filter part) describes the coarse spectral structure, i.e. the formants or resonance frequencies that make up different phonemes.

While the extension of the excitation signal can be done with simple mathematical methods that do not introduce strong artifacts, the envelope is much more relevant for the quality of the reconstructed wide-band signal. That is why the wide-band envelope is estimated with DNNs in this approach, which are trained on a large speech corpus.

Short biography

Jonas Sautter studied Electrical Engineering, Information Technology and Computer Engineering at RWTH Aachen University, Germany. He received his Master of Science degree in 2016. The Master’s thesis with the title “Digital Robust Control for Active Noise Cancellation in Headphones and Hearing Aids” was composed at the Institute of Communication Systems at RWTH Aachen. Since November 2016, he is a PhD student at Nuance Communications in Ulm, supervised by Professor Gerhard Schmidt, Head of the Digital Signal Processing and System Theory group at Christian-Albrechts-Universität, Kiel.

Prof. Schmidt

Diesen Termin meinem iCal-Kalender hinzufügen

zurück

Kolloquiumsvortrag (ET&IT), Dr.-Ing. Vasudev Kanade Rajan / am 27.11.2017

27.11.2017 von 17:15 bis 18:00

Institute Ostufer, Geb. D, "Aquarium", Kaiserstr. 2, 24143 Kiel

Titel: Digital Road Noise Cancellation System Through Active Noise Control

Abstract: The application of active noise cancellation in real-world has not been fully realized yet. From reducing environment noise through the usage of headphones, to engine noise on commercial jets there are a number of use cases. Each of these use case brings its own set of challenges which can be understood only through multi-disciplinary work. One such use case the the reduction of road noise in vehicles. Structure-borne road noise dominates the cabin of modern vehicles. Several road noise cancellation (RNC) prototype systems have been implemented and demonstrated. These systems are based mainly on analog sensors. The placement of these sensors has been so far been based on random optimization methods. In this talk I will talk about the challenges in developing a generic digital RNC system which includes problem analysis, sensor placement, and performance. An adaptive algorithm process the acceleration signals with high convergence and reaction time for various speed and surface ranges, in order to maintain high audible effects for the passengers. Several modern vehicle platforms are integrated with the digital RNC system with ANC microphone at the headliners and the standard audio loudspeaker setup in order to integrate the technology with the existing audio layout of the vehicle.

Short biography

Vasudev Kandade Rajan received Bachelors degree in Electronics and Communication from Visvesvaraya Technological University, Bangalore, India. He joined as Project Research Assistant in July 2008 in the Electrical Communication Engineering Dept, Indian Institute of Science, Bangalore. There he worked on performance management of IEEE 802.11 WLANs until Sept 2009. He then went to obtain his Masters degree (MSc.) in Digital Communications, 2011 and PhD degree in Signal Processing, 2017 from Universtiy of Kiel, Germany. Currently he is working in the R&D department of Harman Becker Automotive Systems GmbH, Straubing, Germany.

Prof. Schmidt

Diesen Termin meinem iCal-Kalender hinzufügen

zurück

Aktuelle Meldungen

Forschung & Projekte

  • Logo Erneuerbare Energien und Klimaschutz Schleswig-Holstein Biomagnetic Sensing Nina Logo
    Software Challenge Germany Logo
    Kosse SH Logo

     

    FOR2093
    Power Girls and Power Boys Online Mathematik Brückenkurs (OMB+)  Jufo Landeslogo
      Materials for Brain  

     

Kalender

« November 2017 »
Mo Di Mi Do Fr Sa So
30 31 1 2 3 4 5
6 7 8 9 10 11 12
13
  • 17:15: Kolloquiumsvortrag (ET&IT), Prof. Elisabetta Chicca, Uni Bielefeld / am 13.11.2017
  • Klicken Sie, um Details zu allen 1 Terminen zu sehen.
14 15 16 17 18 19
20
  • 17:00: Kolloquiumsvortrag (ET&IT), M.Sc. Jonas Sauter, Nuance Communications / am 20.11.2017
  • Klicken Sie, um Details zu allen 1 Terminen zu sehen.
21 22 23 24 25 26
27
  • 17:15: Kolloquiumsvortrag (ET&IT), Dr.-Ing. Vasudev Kanade Rajan / am 27.11.2017
  • Klicken Sie, um Details zu allen 1 Terminen zu sehen.
28 29 30 1 2 3