Rdmr at 14:08, 10 November 2009

2009-11-10T14:08:54Z

New page

__NOTOC__
{{speakerLargeBio|
|name=Steve Young
|image=steve_young.jpg
|email= sjy@eng.cam.ac.uk
|www=http://mi.eng.cam.ac.uk/~sjy/
|bio= Steve Young received a BA in Electrical Sciences from Cambridge
University in 1973 and a PhD in Speech Processing in 1978. He held
lectureships at both Manchester and Cambridge Universities before being
elected to the Chair of Information Engineering at Cambridge University
in 1994. He was a co-founder and Technical Director of Entropic Ltd from
1995 until 1999 when the company was taken over by Microsoft. After
short period as an Architect at Microsoft, he returned full-time to the
University in January 2001 where he is now Professor and Head of
Information Engineering.

His research interests include speech recognition, language modelling,
spoken dialogue and multi-media applications. He is the inventor and
original author of the HTK Toolkit for building hidden Markov
model-based recognition systems (see http://htk.eng.cam.ac.uk), and with
Phil Woodland, he developed the HTK large vocabulary speech recognition
system which has figured strongly in DARPA/NIST evaluations since it was
first introduced in the early nineties. More recently he has developed
statistical dialogue systems and pioneered the use of Partially
Observable Markov Decision Processes for modelling them. He also has
active research in voice transformation, emotion generation and HMM
synthesis.

He has written and edited books on software engineering and speech
processing, and he has published as author and co-author, more than 200
papers in these areas. He is a Fellow of the Royal Academy of
Engineering, the Institute of Electrical Engineers and the Royal Society
of Arts. He served as the senior editor of Computer Speech and Language
from 1993 to 2004 and is now a member of the editorial board. He is a
Senior Member of the IEEE and a member of the SPS Awards Committee. He
was a member of the IEEE STC Committee from 1997 to 1999 and is
currently a co-opted member. He has served on the technical committees
of numerous workshops and conferences. He was the recipient of an IEEE
Signal Processing Society Technical Achievement Award in 2004.}}

== Date ==

* 15:00, Friday, November 13<sup>th</sup>, 2009
* Room 336

== Speaker ==

* Steve Young, Cambridge University, UK

== Abstract ==

A cognitive system is an information processing system which is able to
adapt to its environment and learn from experience. In some sense it is
"self-aware". It will typically utilize psychologically plausible
computational representations of human cognitive processes as a basis
for system designs that seek to engage the underlying mechanisms of
human cognition. The benefits of the cognitive approach are nowhere more
apparent than in the user interface itself, especially where the input
modalities are prone to error such as in speech and gesture-based
interfaces.

Most deployed user interfaces are "hard-wired" and "non-cognitive". Even
if they take account of real human behavior in their initial design,
once deployed their decision-making processes are frozen. They cannot
learn from experience and no matter how many times they interact with a
user, their ability to deal with noisy inputs, user misunderstandings
and changing tasks does not improve. As a consequence, such systems are
expensive to design, fragile in operation and difficult to maintain.

Whilst cognitive user interfaces provide a way forwards, their
development to date has mostly been focused on integrating theories of
cognitive psychology and formal linguistics into computational systems.
However, it is not clear that systems built on such explicit models of
behavior will be any less fragile than the ones they seek to replace. An
alternative engineering approach is to structure systems as
probabilistic models and then use machine learning methods to optimize
and adapt on-line. The fundamentals underlying such an approach are
firmly based on Bayes' theorem and Bellman's dynamic programming
equation, and interestingly, there is growing evidence that humans learn
using exactly the same principles.

This talk will argue that we must start designing systems which embody
elements of cognitive behavior and that probabilistic inference provides
the necessary engineering underpinning. It will begin by using a simple
example to explain how human-computer interaction can be modeled as a
partially-observable Markov decision process (POMDP). Recent results
will then be presented which demonstrate that Bayesian inference and
reinforcement learning also underpin the way that humans learn similar
tasks. Using the example of a spoken dialogue system, some of the key
issues in scaling POMDPs to real-world tasks will then be addressed. The
talk will end by describing recent results obtained using a POMDP-based
dialogue system to provide tourist information over the telephone.

[[category:Seminars]]
[[category:Seminars 2009]]
[[category:Invited Presentations]]

Cognitive User Interfaces: an Engineering Approach - Revision history

Rdmr at 14:08, 10 November 2009