Speaker: Dr. Brendan Juba(Harvard Univ.)
Time and date: 4:00pm - 5:30pm, May 8 
Place: CELC seminar room (404, 4F)
Title: Knowledge Farming

Abstract: A variety of problems can be cast as "data mining with a
goal." I propose a framework for solving such problems by
integrating machine learning into logical reasoning problems. 
In machine learning, we may obtain a new formula satisfied by data. 
In logical reasoning, we seek a proof that derives a query proposition
from background knowledge, given as additional input propositions. In 
the combined approach, we seek a proof that derives a query from the 
background knowledge and any propositions that can be learned from 
the input data. The crucial advantage of this integrated approach is 
that the query and background knowledge serve as context for 
learning. This context guides an algorithm to identify the relevant 
propositions satisfied by the data. I will show how these algorithms 
can possess desirable properties, such as tolerance to adversarial 
noise. Moreover, for some applications, such integrated algorithms 
provide the first algorithms known to be efficient.