|
| |
DR. SHANE LEGG
Shane Legg, BCMS (hons.), M.Sc., Ph.D. is currently a post doctoral
researcher at the
Swiss Finance Institute, University of Lugano. He is studying the
effect of cognitive bias on models of asset pricing and optimal
portfolio choice. In 2009 he will take up a two year post doctoral
position at the Gatsby Computational Neuroscience Unit, University
College London.
Shane believes that the deep abstraction
and sequence
prediction that occurs in the cortex lies at the heart of human
intelligence. As such, his research at the Gatsby Unit will focus on
parameter free temporal difference learning and hierarchical temporal
generative models. Besides hopefully forming the basis for an
artificial general intelligence (AGI), Shane is also interested in the
application of powerful prediction algorithms to problems in finance.
He began his artificial intelligence research in the mid '90s
working with Professor John Cleary on the WEKA Machine Learning project
at
the University of Waikato, New Zealand. Upon the completion of his
undergraduate degree, he shifted to the University of Auckland for an
MSc in Kolmogorov complexity and Solomonoff induction with Professor
Cris
Calude. In the late '90s he worked with
Dr. Ben Goertzel in New York
on the Webmind AGI project, and then in 2002 with
Peter Voss in Los
Angeles on the Adaptive Intelligence AGI system.
In 2003, Shane moved to Switzerland to pursue his PhD under the
supervision of
Professor Marcus Hutter at the Dalle Molle Institute for
Artificial Intelligence (IDSIA). His PhD thesis, titled
Machine
Super Intelligence studies aspects of Hutter's AIXI model of
universal artificial intelligence. In particular it looks at: the
classes of environments for which universal agents such as AIXI learn
to behave optimally; the nature of intelligence and intelligence
testing; a proposed mathematical definition of machine intelligence;
and the constraints placed on powerful intelligent machines by Goedel
incompleteness. Based on this work, Shane won the SIAI-Canada
Academic Prize for 2008.
He authored
Friendly AI is Bunk,
Solomonoff Induction, and
Is there an Elegant Universal Theory of Prediction?,
and coauthored
Universal Intelligence: A Definition of Machine Intelligence,
Fitness Uniform Optimization,
Algorithmic Probability Theory,
Solving Problems with Finite Test Sets,
Temporal Difference Updating without a Learning Rate,
Tests of Machine Intelligence,
A Collection of Definitions of Intelligence,
A Formal Measure of Machine Intelligence,
Fitness Uniform Deletion: A simple way to preserve diversity,
An MDL Estimate of the Significance of Rules,
Objective Evaluation of Inferred Context-Free Grammars,
Ergodic MDPs Admit Self-Optimizing Policies, and
A Taxonomy for Abstract Environments.
Read his
blog!
Print bio!
|
|
