.po 1i
.ll 6.5i
.nr ps 11
.nr pp 11
.ce 10
\fB\s16Semester Project Suggestions\fP\s0
.sp .5
.sz 14
CS 395T: Machine Learning
.sz 12
.sp
One Page Proposal Due March 30
Final Project Due May 15
.ce 0
.sp
.uh "Inductive Learning From Examples"
.np
A large dataset in the domain of Audiology used for the PROTOS exemplar-based
learning system is available for experiments comparing this system with a
standard learning from example system(s). The dataset contains missing
features, so ID3 and any other system involved in the comparison would need to
be augmented to handle this.
.np
Implement a structural description language for the Version Space system and
experiment, may require altering VS to do limited search.
.np
A version of the INDUCE structural learning system is available for applying
to some domain and conducting experiments.
.np
Implement noise and/or missing feature capabilities of ID3 and experiment.
.np
Determine, implement, and test a way to handle noise and/or missing
feature values in VS or AQ.
.np
Implement one of the incremental versions of ID3 (ID4 or ID5) or AQ (GEM,
AQ15) and compare it to the batch version.
.np
Implement and test the perceptron-tree learning algorithm which combines
symbolic and neural net learning methods.
.np
Experiment with "fuzzy" or probabilistic interpretations of inductively
learned rules and compare to a purely logical interpretation.
.np
Add the ability to shift bias to an existing system and experiment. Such a
system would start with a strong bias (e.g. conjunctive only) and then if this
doesn't work shift to a more weakly biased language.
.np
Implement and test a system that learns by questioning an oracle about the
class of an example (like MARVIN and ALVIN).
.np
Implement and test the probilistic version of ID3 (gives probability that
instance is in a clas instead of a binary answer)
.np
Run experiments with modified versions of existing learning systems which
test the hypothesis that simpler rules do perform better on novel data.
.np
Implement and test a method for grammar induction.
.uh "Learning by Observation and Discovery"
.lp
.np
Implement and test a small version of the CLUSTER system.
.np
Build a numerical taxonomy or standard dynamic clustering system to cluster
examples and then use one of the standard learning from examples
systems to learn concept descriptions for the classes.  Compare results
with CLUSTER and UNIMEM.
.np
Design and test a method for learning parameter settings for UNIMEM.
.np
Implement and test a version of the COBWEB conceptual clustering system.
.np
Implement a small version of a scientific discovery system such as
BACON, GLAUBER, STAHL, or DALTON.
.np
Make the EXPLORER system  more like the real AM and test it.
.uh "Explanation-Based Learning"
.lp
.np
Develop rules for the EGGS system in some reasonably difficult domain
and conduct experiments on learning macro-rules in this domain.
.np
Build an EBL system for learning plans using STRIPS operators
by explaining and generalizing observed sequences of operators.
.np
Develop and experiment with methods for eliminating little used macro-rules or
selectively learning macro-rules.
.np
Develop and experiment with methods for learning heuristics for when to
apply rules compared to learning macro-rules.
.np
Develop and experiment with methods for limiting the use of learned rules
to insure performance improvement.
.np
Implement and test a forward-chaining rewrite-rule system which learns
new rules and improves its problem solving performance.
.np
Using EGGS, implement a version of the BAGGER system for learning
iterative  macro-rules.
.np
Implement and test Induction Over Explantions as a learning system which
combines explanation-based and empirical techniques.
.uh "Connectionism"
.lp
.np
Experiment with the delta rule or back-propagation systems by applying them
to a specific learning problem and compare to other approaches.
.np
Develop and test various ways of handling missing data in connectionist
learning methods.
.np 
Test various ways of using neural net learning algorithms effectively in
an incremental setting.
.np 
Imlement and test the Boltzman machine learning algorithm.
.np
Develop and test a genetic learning system.
.uh "Analogy and Exemplar-Based Learning"
.lp
.np
Implement and test a version of the Structure Mapping Engine for analogy.
.np
Implement and experiment with a simple exemplar-based learning system,
comparing it to other approaches for learning from examples.


