🚀 go-pugleaf

        Postdoctoral position in Plan Recognition 
                  Monash University
                (Melbourne, Australia)

Project Description
-------------------
In this project we propose to design and implement a system for
performing plan recognition in realistic domains. To this effect, we
propose to consider two separate but inter-related aspects of plan
recognition: (1) modeling dynamic, complex domains; and (2) designing
and implementing algorithms for performing plan recognition using
these domain models.  In addition, we propose to de-couple the plan
recognition mechanism from the objective of the system. This will
support the construction of agents that apply their plan recognition
skills to achieve different objectives, e.g., helping or hindering a
user, or simply achieving their own goals.

The lack of a concise representation of the domain implies that it is
not feasible to hand-code such a representation. Hence, the envisioned
system must apply machine learning techniques to encode the domain
knowledge. The use of these techniques will also support adaptive
behaviour, which is required to model a dynamic domain. The system
will receive as input logs of actions performed by the users of the
system, and also partial observations of the state which include
information about the objective(s) eventually attained by these
users. The output will be a representation of sequences and/or sets of
actions which usually lead to different goals in the domain. This
representation will reflect the actions commonly performed by users
and the goals that are commonly attained, rather than a mathematical
and complete relation between actions and goals. Hence, this
representation is particularly suitable for plan recognition.

The features of the domains at hand indicate that these domains must
be modeled by means of a probabilistic mechanism that is capable of
representing hidden states, such as Hidden Markov Models (HMMs) or
Probabilistic Recursive Transition Networks (PRTNs). HMMs have been
recently used in machine translation and speech
understanding. However, they lack the expressive power of PRTNs, which
are required in order to encode Probabilistic Context Free
Grammars. In this project, we intend to investigate formalisms of this
class, and to design and implement inductive learning algorithms for
the selected formalism.

Most plan recognition mechanisms developed to date perform forward
chaining of plan operators. This inference process is not suitable for
the domains at hand, where plan operators alone are not sufficient to
model the domain adequately. Thus, plan recognition algorithms which
can use the domain models in this project will be developed. It is
envisioned that these mechanisms will take into consideration the
extent to which the observed actions match the Markov Models or PRTNs
generated in the domain modeling stage. The predictive power of these
algorithms will be tested by evaluating the accuracy of the goals they
predict at different points in the sequences of actions performed by
users of the system. At a later stage, these algorithms will be
incorporated into intelligent agents which can either assist a user,
hinder the user or execute plans of their own.

Requirements
------------
The suitable applicant should have a solid mathematical background, be
proficient in inductive learning techniques, and know C and
Lisp. Knowledge of plan recognition techniques is desirable.

Conditions
----------
Position: Research Fellow, Level A. 
Salary range: AUS$37,345-$40,087. 
Length of contract: initial appointment for 1 or 2 years, 
		renewable for up to 3 years.
Funding source: Australian Research Council
Start of contract: January 1, 1996 (or as soon as possible after that date)

The City of Melbourne
---------------------
Melbourne is a city of about 3 million people, many of whom have come
from other countries, giving the city a distinctive cosmopolitan
character. The climate is temperate, with typical daily maximum
temperatures ranging from a winter low of about 13C to a summer high
of around 35C. The metropolitan area of Melbourne is served by a
generally acceptable public transport network of trains, trams and
buses.

The Department of Computer Science at Monash University
-------------------------------------------------------
The Department of Computer Science at Monash University (Clayton
Campus) is part of the Faculty of Computing and Information
Technology, which is the largest computing education organization in
Australia, with more than 3000 students, and about 100 academic and 30
support staff in eight departments.

Monash University's Clayton campus is located on a spacious (101
hectare) landscaped site in the suburb of Clayton, 20 Km. south-east
of the central business district of Melbourne.

Enquiries
---------
Enquiries may be made by email to ingrid@cs.monash.edu.au or
annn@cs.monash.edu.au.

Applications
------------
Deadline for applications is January 10th or until position is filled.
Applicants should send a curriculum vitae, including the names of at
least two references, by e-mail to ingrid@cs.monash.edu.au or
annn@cs.monash.edu.au. If e-mail is impossible, hard-copies may be
mailed to:

Ingrid Zukerman or Ann Nicholson
Department of Computer Science
Monash University
Clayton, VICTORIA 3168
AUSTRALIA


-- 
Ann Nicholson                   _--_|\          annn@cs.monash.edu.au
Dept. of Computer Science      /      \         phone: +61 3 9905 5211
Monash University              \_.--.*/         fax:   +61 3 9905 5146
Clayton, VIC., 3168, Australia       v	    http://www.cs.monash.edu.au/~annn/