CIS 410/510: Computational Data Science

Spring 2008

Location: Room 200, Deschutes Hall
Time: 10:00-11:20AM, Tu/Th
Prerequisites: Calculus.

Instructor: Matthew Sottile (matt@cs.uoregon.edu)
Office: 203 Deschutes Hall
Office Hours: 2:30-4:00pm, Mon/Wed.

Course Overview

Data is everywhere, and the volume of data is increasing rapidly as improvements in simulation and experimental science progress. The focus of this course is on how one sifts through this vast volume of data to gain some knowledge from it. Very often the information that one seeks, particularly from numerical data, requires a coupling of clever algorithms with mathematical techniques in order to tease out the needle from this numerical haystack. Very often the algorithms become simpler if one leverages the mathematical properties of the data in how the problem is approached. This course will look at a combination of numerical techniques and algorithms to solve interesting problems in data analysis. We will explore how computation drives the science of understanding data: hence, computational data science. This name is also chosen to reflect the connection of the topic to computational science. In computational science, we are concerned with the modeling and implementataion of scientific probles. In this course we will look at the computational science issues specifically related to analysis of data produced by computational scientists from their models and simulations.

Students will learn about the various forms that data can take on, including images, volumetric data, time series, and graphs. We will look at how one can infer meaningful structure and properties from these representations using algorithms and mathematics. We will explore how one defines relationships such as a quantitative measure of ``similarity'' between data elements. We will discuss questions such as:

• What can basic statistics tell us about data? What can they not tell us?
• What is the Fast Fourier Transform, and why is it such a useful algorithm?
• How does one quantitatively define the concept of similarity? How do you go about building algorithmic measures and metrics that take into account expert opinions? What exactly are measures and metrics?
• How can we take a question from a scientific domain (such as neuroscience, physics, or biology), and determine the data processing methods that can be applied to aid in answering it?

The course work will be composed of a mixture of exercises to demonstrate an understanding of the material, one or more whole-class, hands on projects to get a feel for working with data, and a few basic programming exercises. Students will be exposed to tools such as Mathematica, Matlab, and GNU Octave as implementation vehicles for working with data. Prior knowledge of these tools is not required, as the necessary information on how to use them will be part of the course itself. The tools will be made available to the students in the CIS department lab and on departmental servers.

Syllabus

Rough outline of topics (subject to change):

• Overview of data representations: images, volumes, time series, graphs
• Mathematica, Matlab, Maple, Octave: What are they, and how do you use them?
• Basic statistics
• Frequency domain techniques: the Fourier transform and convolution theorem.
• Clustering algorithms
• Image processing algorithms: segmentation, edge detection, registration.
• Optimization algorithms
• Error analysis and uncertainty quantification

Policies

• Attendance: E-mail / phone notification required if you will not be present. Missing more than 20% of the lectures is unacceptable without a prior explanation as to why, and a plan that makes up for the missed material.
• Academic honesty: The usual University academic honesty policies and standards apply. Be sure that you diligently cite any references that you use to complete writing and programming assignments. This includes borrowing code, which is discouraged in the first place, but if you're going to do it, you should cite it instead of claiming it as your own.

Materials

Notes, news, and details can be found here.