Jeff Jackson's Research Interests

The bulk of my research has been in Computational Learning Theory (COLT) and has involved developing mathematical models formally describing various senses of learning as well as proving within these models that certain types of learning problems can or cannot be solved efficiently. My best-known work in this area has been the development of the Harmonic Sieve, an algorithm that efficiently solves the problem of learning certain types of functions (DNF expressions) in a certain relatively natural setting (using membership queries with respect to the uniform distribution).

Recently, I have been attempting to answer a fundamental question that should be of interest to machine learning practitioners as well as to theoreticians: what assumptions must we make in order to reasonably be convinced that a learning algorithm has successfully solved a given learning problem? There are widely accepted answers to this question based on the well-known No Free Lunch theorems for learning; my answer (unpublished joint work with Tino Tamon) is radically different. This research has led me to do some related work in the philosophy of mathematics.

One other major "research" area has involved writing my textbook, Web Technologies: A Computer Science Perspective. While this book has very little relationship with my more theoretical research mentioned above, it represents a tremendous amount of time spent researching (in the college term paper sense), explaining, and illustrating various web technologies.

Finally, I've dabbled in research in a number of other areas, including pure math, empirical machine learning, software engineering, human-computer interaction, and astrophysics (as an undergraduate assistant).

A listing of my research publications appears below. Some links to other information about me and some of what I'm involved in:



Recent non-COLT Papers

COLT Papers

Other Work

E-mail: Please search for my name by visiting the Duquesne University home page and selecting a directory search from the menu.