CS 289, Fall 2004
Final projects, due Monday Dec 13th
To be done in pairs. Dec 13th is a hard deadline.
Here is a list of suggestions for final projects. These are
only suggestions. You can do a different project if you wish,
e.g., one related to your current research or to subject matter from
another class, provided that it includes knowledge representation
and reasoning.
You will need to write a short (one page) proposal, due by
Friday Nov 12th (email OK) explaining the project you have chosen and
how you are going to approach it.
Practical projects
- Implement an approximate inference algorithm (such as likelihood
weighting or MCMC) for the BLOG first-order probabilistic language
and demonstrate its operation on some example BLOG models.
- Define DBLOG, the dynamic version of BLOG specifically for
temporal models, and demonstrate a particle filtering implementation.
- Build a system based on logical or probabilistic inference for
selecting moves in Minesweeper. Scale your system up to large
instances using fast approximate inference, problem decomposition, or
other methods.
- Develop a complete formal model of FreeCell in STRIPS and use a
planning algorithm to solve it. (You might consider SATplan or a
heuristic search planner.) Investigate performance as a function of
the number of cards in each suit. You will notice that standard
FreeCell is "easy" -- one must make several mistakes in order to fail.
What happens when the rules disallow the use of the four ace-up cells
on the right? How easy is it to reach a state with four complete stacks from
king to ace?
- Develop a complete FOL ontology for information-bearing objects
(generally construed -- everything from gif files to rune stones)
and related events such as publication, printing, copying,
downloading, etc. Include as many axioms as you can.
- Exercise 10.19 in AIMA2e. Put your axioms into a knowledge
base and demonstrate its correctness and usefulness.
- Exercise 10.20 in AIMA2e.
- Read Hayes, P., ``Naive Physics I: An Ontology for Liquids,'' In
Hobbs, J. and Moore, R., Formal Theories of the Commonsense
World. Ablex, 1985. (Available in hardcopy from me.) Analyze and
correct Hayes's liquids axioms (with or without the section on
histories). Put them into a theorem prover and demonstrate the
derivations that Hayes claims his theory can support.
- Extend the AIMA Bayesian network code to include decision nodes
and utility modes, and implement 1) an optimal decision algorithm, and 2)
value-of-information calculations,
and illustrate their use in a simple consultation system.
- Extend the AIMA Bayesian network code to provide for dynamic
Bayesian networks. Implement at least two of the algorithms discussed
in the course and compare them on a nontrivial filtering problem.
- Develop a system for classifying web pages, based on relational
as well as content features.
- Implement an agent for Markov localization and map-building and
demonstrate it on simple grid worlds.
- Develop a complete OBDD package.
- Use an off-the-shelf OBDD package to reimplement the Bertoli et
al. approach to conditional planning.
- Compare a Rao-Blackwellized decayed MCMC approach for tracking
and data association to other state-of-the art methods.
- Create a full formalization of the Yu-Gi-Oh trading card game (or
other equally fiendish game) along with as many specific card rules as
you can manage. Your formalization should suffice to 1) determine the
legal actions in any state and 2) answer queries about the state
resulting from a given sequence of actions.
- Survey the application of QBF solvers to games, and demonstrate
win determination in a game of your choice. The game can be fully
observable and nontrivial or partially observable and nearly trivial.
- Develop an effective exact or approximaate state estimator
for Kriegspiel.
Theoretical projects
Each of these is fairly open-ended, but should represent a significant amount
of effort and reach as definite a set of conclusions as possible. Most
will involve additional reading, some of which I can point you to.
- Develop solutions for one or more of the problems on
Leora Morgenstern's
Commonsense Problems page.
- Investigate the representational and inferential frame problems for
the event calculus. Show in detail, with examples, how successor-state
axioms can be constructed and used to derive plans for specific goals.
- Survey the state of the art in complexity results for logical and
nonmonotonic reasoning.
- Survey concepts of independence in utility theory, and compare
the various proposals for graphically structured models of utility.
Explain how these can be combined with belief networks to represent
and solve decision problems.
- Develop a formal framework for analyzing situated replanning
agents, including suitable notions of correctness and completeness.
- Investigate the application of Poole's approach to lifted inference
in the context of BLOG.
- Define a suitable notion of intertranslatability between
first-order probabilistic languages and survey known results. (Refs
available.) Give a constructive proof that every BLP model (or
DAPER model) can be represented exactly in BLOG. Is the converse true?
- Find, describe, and critique a substantial DAML+OIL ontology.
- Survey work on state estimation in relational probability models,
and analyze the problem of convergent approximate state estimation in
an open-world first-order temporal probability model.