Tools Overview

Info

ITR Grant

Members

Papers

About

The goals of the project are to build software to support Welfare to Work case managers' advising of clients.  In order to do this, we are studying the decision-making process of the case managers and of their clients;  designing software tools to facilitate the modeling of decision-making scenarios with uncertainty and constraints;  exploring algorithms and heuristics that combine decision-theoretic planning and constraint solving.

BNB:

The Bayes Net Builder (BNB) allows a user to input a Bayes net either through the GUI or through a file in the appropriate XML format.  The Bayes Net Builder can retrieve probabilities directly from the Semistructured Probabilistic Database (SPDBMS), and can perform probability fusion if multiple probability tables are retrieved.

Tools

Overview

BN BUILDER

DIET

PET

POET

SPDBMS

PET:

The Probability Elicitation Tool (PET) receives a Bayes net structure from the BNB or a file, elicits probabilities from a domain expert, and sends the data to the SPDBMS.

POET:

The Programmable Online Elicitation Tool (POET) elicits a user's preferences about a pre-set domain.  The preferences are elicited via sliders (Likert scales) for individual domain attributes, or
attributes where the preference depends on the value of other attributes. (For instance, in the movie domain, a preference for action movies (attribute: genre) might depend on the star (attribute:  actor).)

Links

AI Seminar

Logic and AI Lab

CS Colloquia

CS Department

UK

DIET:

The Domain Instance Elicitation Tool (DIET) allows an expert to describe the attributes of a domain, their values, and possibly some archetypical preference settings, all via a simple web form.  This is stored as an XML document, which POET uses to define its interfaces.  DIET also allows the domain expert to change the underlying XML document, without ever seeing XML.

SPDBMS:

The Semistructured Probabilistic Database Management System (SPDBMS) stores and manages probability tables and related information.  It can handle simple, conditional, and joint probability distributions, and can access the data using standard relational queries and probability-theoretic queries such as marginalization and conditionalizations.