Ideas For Domain Modeling In Bayesian Networks
Your selection of any domain must be involved enough varaibales (at least 50 variables). Students must show that their model is valid. This can be done by connecting with an expert in the domain or by having data. Once you have the Bayesian network, you should propose what kinds of experiments to conduct over it. You may want to use REES to experiment with diffrent algorithms.
Here are a few ideas for domains:
Advising a first year student: a freshman student needs to make many decisions in his first year (what is his major, what classes to take, etc.) Build a Bayesian network, based on your own expertise, that will advise a first year student in his first steps oncampus.
Recognizing agent's (terrorist) activities: Build a Bayesian network that can recognize terrorits activity and terorists. if you can have access to this domain, either based on some book on the subject or by geting in touch with an expert.
- Advising a busy agent: (example: an ICS Faculty prfessor) regarding the task they should or should not commit to. Consider a faculty prfessor that gets constantly requests to perform all sorts of activites, such as: participate in conference committees, reviews of papers, give invited talks at various parts of the world, write letters of recommendations, participate in committees in his campus and in the department, participate in students candidacy exams. Each arriving request is can be characterized by a job-type, has a deadline, and has activity nature (how much time it takes to do the job for example). There are also self-imposed tasks (e.g., submit a paper to a conference, write a journal paper, go to a conference.) In addition, there are regular periodical activities (e.g., teaching, department meetings, etc.). There could be utility and costsdescribing the rewards of performing the various activities. The Bayes network should model this situation. Given a new request it should predict a "load" variable which is a function of time. Based on this load the agent can decide whether or not to commit to the task. This problem can be a variation on job-shop scheduling. But here, the main task is to decide whether or not to accept a job and determine the load, rather than to schedule the jobs.
- Modeling a carpool driver:
The carpool agent: Assume a driver in a carpool.
The driver drives a few times a week from LA to Irvine mostly in a carpool. We want to model in a Bayes network and a constraint network the knowledge of the driver accumulated over time that allows him to make decision each week on the days and time in which he wants to go to Irvine. We want to have enough information about the freeway system, about traffic patterns during the day, during the week and during the year. The driver has constraints: his teaching time, meeting times, etc the work place. Some of these are fixed, some are changing everyweek. The main decision points each week is what days to drive, what times to tell the carpool manager they should leave from LA and depart from Irvine. The driver has knowledge about other drivers constraints and preferences which can use. Project: model a driver agents by acquiring information from the real agent and from data over time. Can we show how the agent learns? The driver want to drive as little as possible as the driver of the carpool. Can we develop a system that help makes this decisions?
Input: The driver constraints for the week, the time of the week
Output: decide what days/time to send the scheduler of the carpool
for that week. The knowledge-bases should have general information:
about time to drive LA-Irvine, traffic pattern as a function of
time (day/hour,season/holiday, etc) Information about other agents