Dustin Wehr

I'm a PhD student in computer science at the University of Toronto. My main research interests are applications of formal logic. As an MSc student I worked in Computational Complexity Theory. My supervisors are Steve Cook and Alasdair Urquhart. Send email to [my last name] followed by "@cs.toronto.edu".

Research stuff

Paper proving the easier of the two main conjectures posed by Anna Gál, Michal Koucký, and Pierre McKenzie in their 2008 paper "Incremental branching programs":
Lower bound for deterministic semantic-incremental branching programs solving GEN
The harder problem, proving superpolynomial lower bounds for nondeterministic semantic-incremental branching programs, is still open.

Publications

The first publication listed below is a journal paper that contains all the results from the two conference papers that follow it in the list.

Fractional Pebbling and Thrifty Branching Programs, FSTTCS 2009, Mark Braverman, Stephen Cook, Pierre McKenzie, Rahul Santhanam, Dustin Wehr

Also my masters paper (arXiv) contains some extensions of results from that work.

Here is a writeup of an unpublished result that we mentioned in the TCT journal paper:
Exact size of smallest minimum-depth deterministic BPs solving the tree evaluation problem

Some school projects

Using Restricted Boltzmann Machines for recommendations (i.e. collaborative filtering): a description and some small improvements on the influential work of R Salakhutdinov, A Mnih, G Hinton (one of the key algorithms used by the winners of the Netflix Prize). This is a report on a course project with my friend Wesley George for Geoff Hinton's graduate course Introduction to Machine Learning.

A detailed statement and proof of Büchi's Theorem, which gives a relationship between Monadic Second Order Logic and finite automata on infinite words. This is a report I did during my undergrad for Prakash Panangaden's graduate course Formal Verification.

Hobby stuff

Firefox extension mingaling. Google is shutting down their Translate API, so I'm now considering replacements. Microsoft's Bing API looks best, so far. Still using Google Translate. :-)