Theory and Practice in ML 2013 : ACML'2013 workshop on Theory and Practice in Machine Learning
Call For Papers
Several different forms of theoretical foundations for machine
learning have been developed over the years for different settings.
Examples include PAC and PAC-Bayesian learning, online learning,
statistical learning theory, Bayesian learning theory, model selection
using MML/MDL and Algorithmic Information Theory. These frameworks
provide performance guarantees for certain categories of algorithms
when applied to classes of problems. Although strong theoretical
guarantees do not necessarily imply good empirical results, theory can
arguably still provide principles and insight when developing new
algorithms or applying them to new problems.
Conversely, there has been a recent and dramatic uptake in the
application of machine learning techniques to an increasingly diverse
range of problems. Arguably, some of the most successful algorithms on
practical problems are not completely understood theoretically (e.g.,
random forests) or tend to be very ad hoc (e.g., the collections of
techniques that won the Netflix prize). Extracting general principles
and theories from these successes could help improve how quickly and
easily new, practical problems are solved.
In this workshop we intend to discuss what makes a theory relevant
for practical development of algorithms and how the gap between theory
and practice can be decreased. We invite submission of 500-1000 words
to firstname.lastname@example.org by the 4:th of October on the following
(non-exhaustive) list of topics:
⁃ Tightening the relationship between theory and practice
⁃ Applications of theory to explain empirical successes or failures
⁃ Gaps where theory don’t explain practical successes or failures
⁃ New, empirically inspired directions for theory
⁃ Case studies where theory helped improve application of ML
or with general well argued positions on
⁃ What kind of foundations contribute to practical success?
⁃ The role of theory for interpreting empirical findings; Can an
experiment be trusted without a theory?
⁃ What is a (un)principled algorithm?
Notification of acceptance for short talk or poster is sent by the
11:th of October. Please state if you have a preference between oral or
Webpage : https://sites.google.com/site/mltheoryandpractice/
Tiberio Caetano, NICTA Sydney and Ambiata
Fang Chen, NICTA Sydney
⁃ Peter Sunehag, the Australian National University email@example.com
⁃ Marcus Hutter, the Australian National University firstname.lastname@example.org
⁃ Mark Reid, the Australian National University email@example.com