[Compstats] Talk on principles of machine learning (Wed 4pm DC2306C)

Pascal Poupart ppoupart at cs.uwaterloo.ca
Mon Jan 18 17:57:49 EST 2010


Hi everyone,

Shai will break the ice on Wednesday with a first talk on the principles 
of machine learning.  The talk will be at 4pm in the AI lab (DC2306C).

Speaker: Shai Ben-David
Title: No Free Lunch in Machine Learning

Abstract:
 One of the most fundamental insights of machine learning is the No 
Free Lunch (NFL) principle, stating that no learning is possible without 
the application of prior domain knowledge.
In spite of the crisp and basic nature of this principle, many seem to 
be ignorant of it and of its implications.
On the application side, we come across doctors that  expect machine 
learning to deliver medical insights based on just raw
archives of patient files, data base people that expect clustering to 
detect duplicate records from input consisting of just a collection of 
records and the list goes on and on. Even more bothersome is the recent 
emergence of CS professionals that
promise (either implicitly or explicitly) "universal" machine learning 
tools that can "learn any needed the prior knowledge autonomously". Two 
upcoming distinguished talks here in Waterloo (Hinton's talk at the CS 
club on Jan 26th, and Lipson's Perimeter Institute February Public Talk 
http://www.perimeterinstitute.ca/en/Outreach/Public_Lectures/Public_Lectures/) 
are relevant in this context.

In my talk, I will explain the NFL principles and discuss its 
applications in several established as well as emerging ML paradigms 
(including why it rules out the existence of universal learning tools).



-- 
------------------------
Pascal Poupart
Associate Professor
David R. Cheriton School of Computer Science
University of Waterloo
200 University Avenue West
Waterloo, Ontario
Canada N2L 3G1
------------------------
Web: http://www.cs.uwaterloo.ca/~ppoupart
Email: ppoupart at cs.uwaterloo.ca 
Telephone: 1-519-888-4567x36239 
Fax: 1-519-885-1208
------------------------



More information about the Compstats mailing list