Intelligent Signal Processing Jan Larsen Informatics and - - PowerPoint PPT Presentation

Informatics and Mathematical Modelling / Intelligent Signal Processing 1 Jan Larsen

Intelligent Signal Processing

Jan Larsen Informatics and Mathematical Modelling Technical University of Denmark Web: isp.imm.dtu.dk

Jan Larsen 2 Informatics and Mathematical Modelling / Intelligent Signal Processing

DTU Facts

Independent state university since January 1st, 2001 Annual budget 1382 million DKK or 186 million Euro External research finances account for 35% 1027 faculty/academic staff members, 948 technical and administrative staff Organized in 15 departments 1500 B.Sc., 4300 M.Sc., 500 Ph.D., and 500 international students 380 courses in English, 10 international M.Sc. programs

Jan Larsen 3 Informatics and Mathematical Modelling / Intelligent Signal Processing

DTU Facts

Located in Lyngby 10km (6.3miles) north of Copenhagen More than 80 buildings Total floorage of about 375,000 m2 (3.3million ft2) Campus size 1500m from North to South, 850m from East to West (0.9 by 0.5 miles)

Jan Larsen 4 Informatics and Mathematical Modelling / Intelligent Signal Processing

IMM

86 faculty members 32 administrative staff members 60 Ph.D. students 70 M.Sc. students annually 6000 students follow an IMM course annually

Jan Larsen 5 Informatics and Mathematical Modelling / Intelligent Signal Processing

Competences and Research at IMM

image processing and graphical communication software engineering data safety knowledge-based systems networks and distributed systems embedded systems mathematical physics mathematical statistics geoinformatics

• perations research

intelligent signal processing integrated circuits and digital system numerical analysis programming languages

Jan Larsen 6 Informatics and Mathematical Modelling / Intelligent Signal Processing

ISP Group Activities

Humanitarian Demining Monitor Systems Biomedical Neuroscience Multimedia Methods Algorithms

• 3 faculty
• 3 post docs
• 11 Ph.D.

students

• 10 M.Sc.

students

• 3 faculty
• 3 post docs
• 11 Ph.D.

students

• 10 M.Sc.

students

Jan Larsen 7 Informatics and Mathematical Modelling / Intelligent Signal Processing

Research Projects

• 1991-2001 CONNECT: fundamental research in neural networks and adaptive

systems

• 1997-2001 DMM: Distributed Multi Media research program
• 1998-2001 THOR Center for Neuroinformatics
• 1996-1999 Interdisciplinary Neuroscience Initiative - neuroimaging

collaboration in Copenhagen

• 1999-2002 SITE: signal and image processing methods for telemedicine
• 2002-2006 International Center for Biomedical Research, minor contribution to

EEG/pain studies.

• 1997-2004 HBP: Human Brian Project - tools for analysis of brain scans
• 2000-2003 MAPAWARMO - EU5FP project on mapping visual cortical regions in

awake, behaving monkey using functional MRI

• 2002-2005 AE-WATT - EU5FP project on neural network and adaptive

processing for monitoring and control of large diesel engines

• 1997- Humanitarian Demining Laboratory (HDL)

Jan Larsen 8 Informatics and Mathematical Modelling / Intelligent Signal Processing

Network

IEEE Neural Networks for Signal Processing technical committee Copenhagen Signal and Image Processing Graduate school (CISP) Copenhagen Brain Research Center Nordic Demining Research Forum European partners through more 6th EU FP LoI’s

Jan Larsen 9 Informatics and Mathematical Modelling / Intelligent Signal Processing

How do we colaborate

Master thesis projects Ph.D. projects funded by DTU Industrial Ph.D. projects Research projects funded by Danish research councils / EU Commissioned research and development Special designed competence development courses

Jan Larsen 10 Informatics and Mathematical Modelling / Intelligent Signal Processing

Competences in signal processing

adaptive statistical nonlinear machine learning bayesian pattern recognition data and webmining

Learning

Jan Larsen 11 Informatics and Mathematical Modelling / Intelligent Signal Processing

Methods and Algorithms

Theory

Bayesian learning Mean field methods Unsupervised/supervised and combined learning Classification/regression Model selection Nonstationary modeling Datafusion Generalization Outlier detection Robust modeling Active learning

Models

Neural networks Relevance vector machines Independent Component Analysis Time-series models Graphical Models Gaussian Processes Mixture models

Jan Larsen 12 Informatics and Mathematical Modelling / Intelligent Signal Processing

Our modeling approach

Model Selection Validation Learning Prior knowledge and data

Jan Larsen 13 Informatics and Mathematical Modelling / Intelligent Signal Processing

Learning Based Modeling

⋅ | , )

u

p(y x w ( | )

u

p x w

Measurements Parameters Labels

Unsupervised Learning Supervised Learning Learning joint probability

Jan Larsen 14 Informatics and Mathematical Modelling / Intelligent Signal Processing

Supervised Learning

Selection of model family Preprocessing A priori knowledge constraints/probabilistic Learning (ML,MAP,Bayes) Active learning Outlier detection

– Erroneous label – Unknown label

Model structure optimization Performance issues

– Generalization – Confidence – ROC curves – Confusion matrices – Reproducibility

Visualization/interpretation

– Importance analysis

( | , )

s

p y x w

Jan Larsen 15 Informatics and Mathematical Modelling / Intelligent Signal Processing

Unsupervised Learning

Identification of structure, e.g. clusters Can be used as preprocessing in combination with supervised learning Methods

– ICA – Probabilistic clustering using RBF neural networks

Novelty detection – identification of new clusters Hierarchical clustering

( | )

u

p x w

Jan Larsen 16 Informatics and Mathematical Modelling / Intelligent Signal Processing

Neuroscience

Analysis of EEG, PET, MRI, fMRI images Lyngby Toolbox Web repository hendrix.imm.dtu.dk Programs: CBRC, HBP, EU-BIOMED II, EU-TMR, EU, Res.Counc., INC, NIH

Jan Larsen 17 Informatics and Mathematical Modelling / Intelligent Signal Processing

Multimedia

Adaptive tools for shared virtual environments Real time multi-modal communication Tools for intelligent mobile phones (CDMA) Sound processing Webmining (content,structure,usage) Programs: Center for Multimedia, DMM

Jan Larsen 18 Informatics and Mathematical Modelling / Intelligent Signal Processing

Humanitarian Demining

Humanitarian Demining Lab (Ørsted•DTU,IMM,FOFT)

– Detection and clearance of mines – Quality control and clearance validation – Socio-economical decision support systems – Resource management systems

Nordic Demining Research Forum New EU 6FP EoIs

Jan Larsen 19 Informatics and Mathematical Modelling / Intelligent Signal Processing

Monitor Systems

Supervision and fault- diagnosis of marine engines Neural optimization of pulse plating in print circuit boards Monitoring of power distribution networks Multi-channel adaptive analysis of revolving machines Adaptation of mobile base station filters

Program: AE-WATT

Jan Larsen 20 Informatics and Mathematical Modelling / Intelligent Signal Processing

Biomedical

Sun exposure modeling Skin cancer diagnosis Bioinformatics

500 1000 1500 2000 2500 3000 3500 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 A B B C D E F G Sensitivity wavenumber/cm-1

Program: Signal and Image Processing for Telemedicine