Business Models: Temporal and Spatial Analysis

In a previous blog entry at http://notesfromworkshop.blogspot.com/2009/01/lokad-business-model.html, I discussed the use of web services for forecasting. One company that I worked as a consultant was i-cubed at http://www.i3.com/. They have a datadoors product at http://www.i3.com/products_datadoors.html that does distributed geographic data management. Notice that their architecture, http://www.i3.com/products_datadoors_architecture.html , is based on an arcGIS front end to XML to a web service tier. I think that these two business models provide good insight into learning how to run a successful service oriented architecture (SOA). Also, I like their ideas on visualization and distribution services. The combination of these two business models for spatial/temporal data would be effective for doing both analytical and predictive modeling.

IBM Technical Library

IBM has a nice technical library known as "DeveloperWorks" at http://www.ibm.com/developerworks/views/data/library.jsp which provides articles and tutorials on Java Technology as well as IBM products. One such interesting set of articles is the six part series at http://www.ibm.com/developerworks/java/library/j-rtj1/index.html?S_TACT=105AGX02&S_CMP=EDU on using Java for Real-time systems. Another good example is on writing multi-threaded applications in Java at http://www.ibm.com/developerworks/java/library/j-thread.html . Another example, Visual Editor for Eclipse is at http://www.ibm.com/developerworks/opensource/library/os-ecvisual/ and provides a framework for building GUIs with Swing and AWT. Using the articles and tutorials at DeveloperWorks can provide additional value and insight for your classroom work.

Harvard-MIT OpenCourseWare

The Harvard-MIT Division of Health Sciences and Technology has reading lists, course information, articles and lecture notes online to be able to enhance your education on medical topics. MIT OpenCourseWare provides free publication of MIT course materials that reflect the subjects taught at MIT-http://ocw.mit.edu/OcwWeb/Health-Sciences-and-Technology/ . For example, the Lecture Notes on Biomedical Information Technology at http://ocw.mit.edu/OcwWeb/Biological-Engineering/20-453JSpring-2005/LectureNotes/index.htm enables one to get a quick education based on the following from the syllabus:

"The objective of this subject is to teach the design of contemporary information systems for biological and medical data. These data are growing at a prodigious rate, and new information systems are required. This subject will cover examples from biology and medicine to illustrate complete life cycle information systems, beginning with data acquisition, following to data storage and finally to retrieval and analysis. Design of appropriate databases, client-server strategies, data interchange protocols, and computational modeling architectures will be covered. Students are expected to have some familiarity with scientific application software and a basic understanding of at least one contemporary programming language (C, C++, Java®, Lisp, Perl, Python, etc.)."

Other courses include Biomedical Signal and Image Processing,-http://ocw.mit.edu/OcwWeb/Health-Sciences-and-Technology/HST-582JSpring-2007/CourseHome/index.htm and for the purposes here Functional Magnetic Resonance Imaging:Data Acquistion and Analysis at http://ocw.mit.edu/OcwWeb/Health-Sciences-and-Technology/HST-583Fall-2006/CourseHome/index.htm . I particularly liked the labs with slides and software.

There are also graduate courses at the Sloan School of Management and the Time Series analysis courses at http://ocw.mit.edu/OcwWeb/Economics/14-384Fall-2007/CourseHome/index.htm with references to Bayesian and Markov Chain Monte Carlo. The wealth of organized information here provides an in-depth, state-of-the-art and quality education for free as well as a contact network to do high quality research.

CUDA Applications

My continue work in image processing is now connected with NVIDIA CUDA architecture for using the GPU for complex computing. You can learn more about CUDA at http://www.nvidia.com/object/cuda_learn.html . The CUDA Zone provides a classroom environment with Podcasts for

(1) Introduction to GPU Computing
(2) CUDA Programming Model
(3) CUDA Programming Basics- Part 1
(4) CUDA Programming Basics - Part 2

Articles from Dr. Dobbs Journal, seminars, tutorials as well as videos from the University of Illinois. The exercises for Visual Studio are very instructive and useful.

Furthermore, the CUDA university courses that you can access throughout the country provides a great foundation. As example, here is one from Kent State-http://www.cs.kent.edu/~zhao/gpu/lectures.htm.

fMRI Labs

The above video is from Science Friday which was broadcasted on May 2, 2008. The article on the video is at http://www.sciencefriday.com/program/archives/200805022. Columbia Unversity has an excellent program in the imaging and cognitive sciences with resources available to the student at http://www.fmri.org/. Dr. Joy Hirsch has many lecture notes on this site as well fMRI course information with recent abstracts and publications.

Neuroscience:Functional MRI

These notes are based on the examination of various syllabi available on the web to enhance the research at The Cromwell Workshop. For example, Professor Marty Sereno has a good syllabus for his cognitive science 276 class on Neuroimaging at (http://www.cogsci.ucsd.edu/~sereno/276/) with plenty of references, readings and links to the field of neuroimaging-MRI, fMRI, EEG and MEG.

Another good example is Professor Calhoon's Analyis Methods in fMRI at(http://www.ece.unm.edu/~vcalhoun/courses/fMRI_Spring07/Syllabus.pdf) with texts and readings. One of the articles that I particularly like is at

http://afni.nimh.nih.gov/sscc/staff/rwcox/ISMRM_2006/Syllabus%202006%20-%203340/files/J_06.pdf

which uses Independent component analysis (ICA) for feature set detection. The use of Bayesian techniques and state space models in this context should prove useful for some of the observations made in the conclusion of the paper.

Another good interdisciplinary course is from the group of professors Hernandez, Jonides, Nichols and Noll at the University of Michigan.

http://www.umich.edu/~fmri/course/2005/ . Dr. Noll has a very good primer on MRI and fMRI at http://www.umich.edu/~fmri/course/2005/primer2.pdf as well as their lab exercies in MatLab.

If you go through these course, readings and labs, you will have a good foundation to be able to diagnose potential problems in the field and how different spatial-temporal models can be used to solve them. Of course, good papers that should be read on fMRI time series analysis from both the frequency and Bayesian perspectives are:

(1) http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/pdfs/Ch10.pdf
(2) http://www.fil.ion.ucl.ac.uk/~wpenny/publications/vb2_2005.pdf

and the deconvolution of fMRI time series data can be found at

http://www.cogsci.ucsd.edu/~sereno/276/readings/3dDeconvolve.pdf

Journals for fMRI research are:

• NeuroImage http://www.sciencedirect.com/science/journal/10538119
• Neurocomputing http://www.sciencedirect.com/science/journal/09252312
• Journal of Magnetic Resonance Imaging http://www3.interscience.wiley.com/cgi-bin/jhome/10005199
• Magnetic Resonance Imaging http://www.sciencedirect.com/science/journal/0730725X

Yale has some interesting software for fMRI that uses MATLAB code by Pawel Skudlarski at (http://mri.med.yale.edu/individual/pawel/fMRIpackage.html). Statistical parameteric mapping software can be obtained at (http://www.fil.ion.ucl.ac.uk/spm/software/).

Of course, this is a topic for another post.