Last
modified: 22 July 2012
Dr. Bryson
supervises two sorts of projects: AI projects, primarily those
to do with making the design of AI systems easier, and NI
projects, primarily those modelling the evolution of social
and cognitive behaviour. NI projects normally (but not always)
include a systems AI component, since the tools we use for
modelling natural intelligence are all AI. Improving the AI or
tools is one way to make the computer science component of the
project clear. AI projects similarly will always have some
interesting AI product to be produced. Being your own
user is also the best way to make sure your tools are actually
useful.See the AmonI software page for details of some of our existing projects, and related papers on the designing intelligent systems page.
Dr. Bryson meets with all her project students once every other week in Autumn term, and weekly in Spring. This is a group meeting, allowing students to experience peer supervising and supervision, and can generally work together on understanding common problems such as new technologies. There are of course a few longer individual meetings at critical periods in both terms. She sometimes accepts novel project proposals from strong students, and is happy to negotiate about and adjust her own proposals to the interests and capabilities of a particular student. However, please note that not every project below can be adjusted to suit every student's capabilities.
Previous projects and students.
Examples of possible AI projects for this year:
Evaluating and Improving an AI IDE
This summer we are working to
substantially improve our core piece of AI development
technology, ABODE (current
/ old version is here). But as I said above,
only use makes tools good. Taking this project requires
strong java skills and an interest in real-time AI systems
such as robots and/or game characters (see e.g. the next two
projects), since, as stated above, you will need to
demonstrate that you can build something with the tool.
A Behaviour Library for LEGO Robots
The department has recently
acquired a set of LEGO robots to be used by final-year
students (mostly in the Spring). We would like to make them
compatible with Behavior-Oriented
Design (BOD) and to work with ABODE (see above).
Possible tasks include RoboCup
Football, RoboCup
Rescue a Robot
Companion, or overcoming the Dalek problem in the East
Building (the stairs.)
Improving the Action Selection Mechanism for an Artificial Football Team
Behavior-Oriented
Design (BOD) is a way of making artificial
intelligence based on combining object oriented design and
proactive planning, normally using POSH action selection.
Last year, Tom Hyde produced a good
team for the Robocup
Football
Simulator League (that's VR, not robots), and
in so doing identified some problems in the real-time
scheduling for POSH. This project will probably be in Java,
but other technologies are possible through an API. Note:
this project requires a very competent programmer. Interest
in sport is not particularly useful & could even get in
the way!
Creating a Bath Implementation of Behaviour Trees
Over the years, a lot of game & robot technology has
been produced at Bath (& a few other universities) using
POSH action selection. However, there are other
action-selection mechanisms (ASM) capable of describing
dynamic plans. One of the most widely cited presently
is the Behaviour Tree (BT), however there is no canonical
downloadable, open-source version of BTs. This project
would involve building a BT action selection
mechanism. Ideally this would be used to replace the
POSH action selection on one of our existing behaviour
libraries (e.g. UT Capture the Flag, Robocup Footbal,
StarCraft: Brood Wars) so that the student would a) be
easily able to compare the two technologies and b) not have
to implement a substantial AI as well as building the ASM.
Visualization and Maintenance of an Emotional Virtual
Agent
Previous PhD student Emmanuel
Tanguy built the Emotionally
Expressive Facial Animation System (EE-FAS). The
basic system is very modular, having been built on top of psyclone.
However, understanding the architecture is not trivial for
new users. This project would involve making it easier to
both develop and understand emotional AI systems. This
might very well involve reimplementing and simplifying
EE-FAS system.
Extending & Generalising BOD Star Craft to more
species and map.
Behavior-Oriented
Design (BOD) is intended as an iterative methodology
to make building AI easier and faster. In 2012, Simon Davies
created a single BOD StarCraft
or VR social spaces, such as Second Life. Can Davies' solution
be refactored to make extending the AI to other maps and /
or species easier? This project would ideally create a
number of additional extensions to.
Examples of possible NI projects this year include:
Modelling matriarchical dominance structures non-human contexts
This project requires
modelling the social structure of lemurs
and/or spotted
hyenas. This would be interesting in itself, but it is
particularly interesting if it helps us understand why these
species have evolved to have female dominant social
structures (most other similar species have male-dominated
structures). This project requires spatial agent-based
modelling, and developing a good understanding of some
concepts from contemporary biology, particularly inclusive
fitness.
Modelling fission / fusion social behaviour and its impact
on information dissemination
Modelling the fission/fusion
society and inter-troop relations of chimpanzees, spider
monkeys or Mongolian asses. There are two existing
models that could be extended for this project, or a new one
could be written from scratch. This project also requires
reading some biology, but would probably not require the
level of theoretical sophistication of the female dominance
project.
Modelling the impact of in-group & out-group
assessments on regional economics
This project
extends from a recent grant run in our group on Understanding
Cultural Variation in Anti-Social Punishment. It would
however require a new model, probably a spatial agent-based
model, but a strong independent student could possibly look
at this with game theory.
Deriving Individual Semantics from Small Text Corpra
This project extends from an
undergraduate dissertation by Avri Bilovich on using simple
vector-based automatic text analysis to track the change of
meaning in texts over time. We would now like to see if we
can track individual differences in word use and shifts in
semantics over time by looking at corpra in the public
domain, such as Enron's email corpus, government records or
social media. See Detecting
the Evolution of Semantics and Individual Beliefs Through
Statistical Analysis of Language Use. Bilovich's
software is written in common lisp; it could be reused
or a new version could be written in another language.
See also Eugene Bann's 2012 dissertation on deriving
information about emotions from twitter corpra.