Workshop
in Computer Science: Learning and Signal Processing
סדנה במדעי המחשב:
למידה ועיבוד אותות
Tuesday 13-15 Dan David 204
Prof. Nathan Intrator
The workshop provides a window to the Neural Computation and
Signal Processing (NCSP) lab at TAU. Usually, projects will be related to work
of some of the graduate students in the lab. In general, the lab applies
machine learning techniques and signal processing methods to real-world
applications.
This year, we shall concentrate on (real-time) data mining
projects with multi-dimensional representations. A typical project or a couple of projects
have the following structure:
Most of the projects will acquire online data from the web and one
or two will use offline-database. The main programming language will be Matlab™
and there will be a couple of data reading projects in Java/Python to provide an interface
between Matlab and the web for specific datasets.
Datasets to be used:
Physionet Earthquakes – Seismic Financial data of various types
Submission and Grading Reports and Presentations
Each project will have to submit the following:
1. (20%) Review and plan of work: power point presentation – this
presentation will be done in the middle of the semester and it will serve to
evaluate whether the students are going in the right direction and will be able
to finish the project on time. This will be the first students’ presentation.
2. (40%) Source code and Users manual at the end of the project. User’s manual will be comprehensive,
providing information of what each function is doing and how it is doing it,
what are the inputs, outputs, and examples of how to use the code. The code will be submitted in one ZIP file.
Please create as few files as possible, with as many functions that you like in
a file. Documentation inside the code is critical.
3. (40%) Final word document and power point presentation: In the
second and final student presentation, they will describe what they have done,
what problem they tried to address and how was it addressed before. They will
be demoing the code, how to use it and its capability – this presentation and
report will be given at the final project presentations.
If there is no presentation, then the word document report should
include the presentation items as well.
The projects will be based on the individual background of each
participant and will thus, vary in nature. Projects will have an emphasis
either on programming (Matlab, Java, C etc) or will be more emphasizing the
data mining aspect.
The projects will be done in couples or triples, but each of the
participants will be assigned a specific part and in the first presentation,
each will present their expected part in the project.
There is room to combine a project on building a presentation that
is optimized for a cell phone – iphone, android,
blackberry etc.
Projects
1. Maya Lieber, Orli
Chen: Real time synchronized commodity prices (Eddie)
2. Yair Spira Roy Russo: Analysis of
financial data (Eddie)
3. Shir Goldstein Kobi Aizer Tzach Tzabag: Real Time EEG Analysis (Vera, Alex)
4. Etgar Petrov
Lihi Mena: Real time acquiring of Seismic
Data (Vera)
5. Amir Ayalon, Maya Levil:
GUI Biomedical on Top of Guy's GUI l
(Alex)
6. Yuval Sharmi: : Prediction comparison ML
Biomedical Data (Shahar)
7. Maya Levi : Puzzle building (Daniel)
Preliminary and Partial List of Projects
1. Real time seismic data analysis for earthquake prediction
a. Real time acquiring of seismic data from a given set of stations
(GUI defined, or otherwise). Also labels such as
magnitude of seismic activity at specific locations will be added.
b. Real time processing of large amounts of data, pre-processing into
Time/Frequency representation.
c. Real time creation of Best Basis (from wavelet packets
representation) of the data (background will be provided)
d. Real Time clustering + GUI
e. Real time machine learning modeling and prediction
2. Same type of project on financial data analysis
a. Real-time acquiring of financial data (multiple sources,
extensive)
b. Prep-processing and analysis
c. Machine learning for prediction training
3. EEG Data analysis
4. Cardiac data analysis
5. fMRI data analysis and modeling
No bio background or signal processing is necessary for some of
the projects.
Those with bio or signal processing background will be able to
utilize their background.
For machine learning projects, a machine
learning, pattern recognition or neural computation course or some background
is a prerequisite.
For a signal processing project, some course in signal processing,
or a course that mentioned Fourier transform is needed. Preference is given to
wavelets and best basis background.
Students |
Student Advisor |
Title |
Pres1 Users Man
Code Pres2 |
|
Shahar |
|
|
|
Yehudit |
|
|
|
Alex |
|
|
|
Yaron |
|
|
|
Elad |
|
|
Related links to relevant literature and code
Sound analysis Ambionics Introduction to Acoustics Signification
of High Dimensional Data Auditory
display of hyperspectral colon tissue images Biomedical signals and sensors Biomedical
Signal Analysis R. M. Rangayyan Breath Sounds
Methodology N. Gavriely Digital
Signal Analysis: A Computer Science Perspective J. Stein. Robust
measurement of Carotid Heart sound delay |
Software TinyOS
operating sys for wireless applications Sensors Cheap
off-the shelf TinyOs operated robots PicoRadio: Low power wireless node with sensors Machine learning and Statistics Pattern
recognition and neural networks B. Ripley Neural
networks for pattern recognition Bishop Hardware |
Some past project titles (extensive
use of eeglab)
An important reference paper to many projects: Mirowski, LeCun et al
Signal Processing and Math Intensive Projects
Examples of past projects:
1.
Source localization for EEG
2.
Spectral Cross Correlation (few methods, 2
projects)
3.
Wavelet Denoising (Using Donoho’s method)
4.
Best basis formation for single electrodes (based on Coifman & Wickerhauser)
5.
Nonlinear Independence (LeCun)
Programming intensive
1. Fast Adaptive
Principal Components with Visualization (Single electrode)
2.
Fast Adaptive Principal Components with Visualization (Multiple
electrodes)
3. Fast Adaptive ICA with
Visualization
4.
Adaptive Lyapunov Exponents estimation
(more math intensive)
_______________________________________________________________________________________________________________________________
Past Projects Feb 7, 2008:
1.
Eugene Jorov: Design of a
recording and analysis microcontroller MSP430FG461x
a.
A/D for Heart Sound sensor, A/D for EKG sensor;
b.
Peak detection analysis of QRS complex and S1 sound
c.
Extraction of peak values and systolic and diastolic durations
Paramters: Window size 500-1200 samples at 400Hz, two channels. The
micro controller provides 4 channels of 16 bit sampling.
Constraints: no dynamic memory
allocation. Code will be written in C.
2.
Yuval Meir, Amit Ziv: Zigbee communication module to MSP430FG461x
with connection to a windows mobile device
a.
Microcontroller part of the communication
b.
Windows mobile part of the communication
c.
A GUI presenting results and enabling the request of data from the
microcontroller (several forms of data presentations will be discussed later)
3.
Miki Avitan: Real time
code for principal components of the S1 heart sound
4.
Shy Zimmerman, Raphi Agiv:
Real time code for clustering of S1 and S2 heart sounds, about 6
clusters, with a threshold changing to provide such number of cluseters.
5.
Liron Machluf, Nir Charny:
Real time code for detection (segmentation) of S2 and features
extraction
6.
Dionis Teshler, Igor Dofman: Production of Shimmer Mote device for EKG with Bluetooth
a.
Production of the Mote
b.
Bluetooth communication with a PC with PC GUI, Code for
Peak detection of QRS with duration information.
7.
Uri Korenstein, Nir Bitanski:
EEG Lab Utilizing Andrey’s Fisher
LDA algorithm for MEG on EEG data
8.
Ron Kimchi, Tsachi
Greemald: ICALab project with
Andrey
9.
Nadav Bendek, Uri Peltz:
Clustering of coherence measures of EEG signals for prediction of
Seizure. Use graph rep of the cluster structure
a.
Review of coherence methods used in EEG analysis; Review of
clustering methods
b.
Methodology: Implementation of different coherence methods,
implementation of different clustering methods, details about the data to be used
and the measures of comparison
c.
Results, clear graphs demonstrating which methods are best
d. Using Shos Data files
10.
Orphan project: Producing
LPC-like coefficients from EEG data and analysis of their clustering
properties.
If you are interested in this project, please email me.
_______________________________________________________________________________________________________________________________
Intended participants are encouraged to email me and describe
their background and their interest in projects.
Updated (Jan 31, 08) specific list
of projects
I’d like each group to choose one
and email back to me, on a first comes basis
1.
Code writing on a micro-controller for real-time analysis of biosignals
1.
Real Time detection of the S2 heart sound, Code in C or Java and
integration with the microcontroller
2.
Real time clustering of the S1/S2 heart sounds
3.
Real time principal components of the S1/S2 heart sounds
2.
Bluetooth 2 way communication with windows mobile device
3. Gui on a windows
mobile device
4.
Zigbee communication with a windows pc
5.
Participation The Google Online
Marketing Challenge. Two groups of two students each can join this for the
workshop. Please read the instructions and get back to me for confirmation. I
expect people jumping on this fast, so you need to respond fast to be chosen.
See me tomorrow regarding this.
General
list of some potential projects:
Analysis of fMRI/EEG data
This work is done in collaboration with Dr. Talma Hendler at the Souraski Medical Center (Ichilov)
and Prof. Eshel Ben Yaakov in Physics. See
presentation of Itai Baruchi
at my Advanced
Seminar.
TinyOS and Wireless Body Sensors Network
TinyOS is an open-source
operating system designed for wireless embedded sensor networks. It
features a component-based architecture which enables rapid development.
This OS has become a standard in the recent development of a Wireless Body
Sensors Network and tmote
sky platform. We shall develop algorithms for real time analysis of ECG
using Pluto
that is based on tmote sky. This platform can
handle up to six different body sensors at a wireless range of 125m.
We shall also develop software and algorithms to embed acoustic
sensors into this platform.
A good place to start is a recent PhD
Thesis about Tele-Cardiology Sensor Network. Currently such networks are
heavily researched at Harvard University and Hospitals in the Boston area under
the CodeBlue initiative.
There are also open projects to BioMedical
Engineering students in development of some sensors to this platform.
Analysis of Heart Sounds Via HMM
Related work by Ray
Watrous. See also presentation by Daniel Gil at
my Advanced
Seminar, as well as background and publications of Guy Amit also in the advanced
seminar page. See also, Alex Weibel TDNN. Outline:
Follow speech recognition approach of extracting automatic features, vector
quantization, HMM, Segmentation and Clustering.
Seismic Data Analysis
Related work: see presentation by Ido
Yariv and Talmor at my Advanced
Seminar.
Projects
1.
Transfer function and complimentary information from heart sounds
in different chest locations
Dima Adler and Doron
Ayad