Matt Fredrikson

Student

Dept. of Mathematics and Computer Science

Duquesne University, Pittsburgh

Email: my last name + the letter 'm' [at] duq.edu

 

Update: I have moved to the University of Wisconsin-Madison, where I will begin graduate school in the fall of 2007. My new page can be found at http://cs.wisc.edu/~mfredrik.

I am currently a senior Computer Science (B.S.) and Mathematics (B.A.) student at Duquesne University. I am originally from Sioux City, IA and now reside in Squirrel Hill, a neighborhood in the east end of Pittsburgh.

 

For the past two years, I have worked as an undergraduate research assistant to Dr. Eric Rawdon, a Math professor in my department. His work is in Physical and Computational Knot Theory. To date, I have worked on three significant problems for Dr. Rawdon

  • Crossing Conversion: Many researchers find themselves working in knot theory for one reason or another. However, certain types of analysis are difficult given only raw vertex data describing a knot. It is useful to store this data in a more convenient form, e.g. as a graph which describes the crossings in a knot and their relation to one another. There is not a single, robust software solution to this problem which is up-to-date and easy to use. I have written a complete software package that performs this task reliably not only for arbitrarily complex knots and links but also for links with an arbitrary number of components. The output is both compact and verbose, making this a very useful tool. A public release of the package will be made available soon.

  • Knot/Link Complexity Reduction: Oftentimes knot or link data comes from noisy or random sources. In these cases, much of the time the link contains superfluous complexities that can be removed while maintaining knot type. It is desirable to remove these complexities for many reasons, including polynomial invariant computations and similar processing. I have written the first piece of software which implements Reidemeister I and II as well as generalized Reidemeister I and II reductions reliably for both knots and links. On average, this removes 80% of the crossings on two-component links generated by random walk. Reduction performance for links of arbitrary component number has not yet been extensively evaluated, but will be in the future. At the moment, source code for this project is not yet available but it can be obtained in binary format for both Win32 and Unix platforms by email.

  • Improving HOMFLY-PT Calculation: The HOMFLY polynomial for knots and links is a very powerful invariant that is practical for common use. However, software to calculate the polynomial is either very outdated (cannot compute on complex knots) or simply unreliable. I am currently working on a solution that can compute the HOMFLY polynomial for arbitrarily large/complex knots and links and does so quickly enough to remain a practical solution for such knots. This is a very difficult problem; calculation is NP-Hard. Hopefully a beta release of the software can be made public soon.

 

As well as my work for Dr. Rawdon, I have recently started working with Dr. Patrick Juola (with plenty of help from Dr. Don Simon!) on a computer security problem. We are currently examining the possibility of applying certain machine learning techniques to the problem of discovering the execution of malicious code on Windows NT-based systems. I will provide more details on this project in the future.