Publications

Clinical decision analysis seeks to identify the optimal management strategy by modelling the un­ certainty and risks entailed in the diagnosis, natu­ ral history, and treatment of a particular problem or disorder. Decision trees are the most frequently used model in clinical decision analysis, but can be tedious to construct, cumbersome to use, and computationally prohibitive, especially with large, complex decision problems. We present a new method for clinical decision analysis that combines the techniques of decision theory and artificial in­ telligence. Our model uses a modular represen­ tation of knowledge that simplifies model building and enables more fully automated decision mak­ ing. Moreover, the model exploits problem struc­ tures to yield better computational efficiency. As an example we apply our techniques to the problem of management of acute deep venous thrombosis.

We present a general theory of action abstraction for reducing the complexity of decision­theoretic planning. We develop projection rules for abstract actions and prove our abstraction techniques to be correct. We present a planning algorithm that uses the abstraction theory to efficiently explore the space of possible plans by eliminating suboptimal classes of plans without explicitly examining all plans in those classes. An instance of the algorithm has been implemented as the drips decision­theoretic refinement planning system. We apply the planner to the problem of selecting the optimal test/treat strategy for managing patients suspected of having deep­vein thrombosis of the lower extremities. We show that drips significantly outperforms a standard branch­and­bound decision tree evaluation algorithm on this domain. We would like to thank Charles Kahn for pointing us to the DVT application.

We describe ongoing work toward development of a decision­theoretic agent to help users choose videos based on their preferences. The DIVA (Decision­ Theoretic Interactive Video Advisor) system elicits user preferences using a case­based technique. Hard constraints are used to permit the user to communi­ cate temporary deviations from his basic preferences. If the user is not happy with the system's recommen­ dations, he can provide feedback, which is used to modify the represented preferences and generate a new set of recommendations. We describe the fundamen­ tal algorithms, the implementation, and some results from some initial experimentation.

We describe the early stages of the development and validation of a Bayesian network to assist in the detection of breast cancer. MammoNet integrates mammo­ graphic findings, demographic factors, and physical examination to determine the probability of malignancy. Conditional probabilities were obtained from the medical literature and from expert opinion. Problems (and solutions) encountered while developing the model are discussed. MammoNet is imple­ mented as a knowledge base of rules; problem­specific networks are constructed using a Bayesian network construction algorithm. The model's performance was evaluated with 77 cases drawn from a textbook and a clinical teaching file; MammoNet performed well, and achieved an area under the receiver operating curve of 0.881 (± 0.045).

We present an educational tool for bringing the in­ formation contained in a Bayesian network to the end user in an easily intelligible form. The banter shell is designed to tutor users in evaluation of hypotheses and selection of optimal diagnostic procedures. ban­ ter can be used with any Bayesian network contain­ ing nodes that can be classified into hypotheses, obser­ vations, and diagnostic procedures. We present algo­ rithms for determining optimal diagnostic procedures and for explanation generation.

This paper discusses techniques for perform­ ing efficient decision­theoretic planning. We give an overview of the drips decision­ theoretic refinement planning system, which uses abstraction to efficiently identify opti­ mal plans. We present techniques for au­ tomatically generating search control infor­ mation, which can significantly improve the planner's performance. We evaluate the effi­ ciency of drips both with and without the search control rules on a complex medical planning problem and compare its perfor­ mance to that of a branch­and­bound deci­ sion tree algorithm.

Intelligent tutoring systems assist medical faculty in training and equipping students with the required clinical reasoning skills. Plausible student solutions to a given problem are rejected by tutoring systems as being incorrect, if they do not match a specific solution accepted by the tutoring system. This leads to brittleness in evaluating student solutions. In this paper we describe a combination of knowledge base expansion and exploitation of existing knowledge structure to enhance robustness in an intelligent tutoring system for medical problem-based learning using UMLS. We present a tutoring system that enriches the solution space by collating different plausible solutions and exploiting the knowledge structure in UMLS to offer students a broader scope of reasoning.

In well-defined domains such as Physics, Mathematics, and Chemistry, solutions to a posed problem can objectively be classified as correct or incorrect. In ill-defined domains such as medicine, the classification of solutions to a patient problem as correct or incorrect is much more complex. Typical tutoring systems accept only a small set of approved solutions for each problem scenario fed to the system. Plausible student solutions that fall outside the scope of this small set of approved solutions are rejected as being incorrect, even though these solutions may be acceptable or close to acceptable. This leads to brittleness in the evaluation of student solutions. This paper describes a tutoring system for medical problem-based learning (PBL), which can accept a wide variety of plausible solutions without placing an extensive burden on knowledge acquisition. A widely available medical knowledge source is deployed as a domain ontology, and concept relationships in the ontology are used to make inferences and expand the space of plausible solutions beyond the scope of solutions explicitly provided to the system. Parent-child relationships are used to infer generalized solutions, whereas relationships of synonymy are used to infer alternate solutions. Evaluations of the system after expanding the solution space indicate accuracy close to that of human experts, who agreed among themselves with Pearson Correlation Coefficient of 0.48 and p < 0.05. The system precision dropped by 32%, while the recall increased by five times. The geometric mean of sensitivity and specificity was increased by 0.33.

Any anytime algorithm used for decision­making should have the property that it considers the most important aspects of the decision problem first. In this way, the algorithm can first eliminate disas­ trous decisions and recognize particularly advan­ tageous decisions, considering finer details if time permits. We view planning as a decision­making process and discuss the design of anytime algo­ rithms for decision­theoretic planning. In particu­ lar, we present an anytime decision­theoretic plan­ ning algorithm that uses abstraction to focus atten­ tion first on those aspects of a planning problem that have the highest impact on expected utility. We discuss control schemes for refining this behav­ ior and methods for automatically creating good abstractions.

Any anytime algorithm used for decision­making should have the property that it considers the most important aspects of the decision problem first. In this way, the algorithm can first eliminate disastrous decisions and recognize particularly ad­ vantageous decisions, considering finer details if time permits. We view planning as a decision­making process and discuss the design of anytime algorithms for decision­ theoretic planning. In particular, we present an anytime decision­theoretic planning algorithm that uses abstraction to focus attention first on those aspects of a planning problem that have the highest impact on expected utility. We discuss control schemes for refining this behavior and methods for automatically creating good abstractions.

The need to reason with imprecise probabil­ ities arises in a wealth of situations rang­ ing from pooling of knowledge from multi­ ple experts to abstraction­based probabilistic planning. Researchers have typically repre­ sented imprecise probabilities using intervals and have developed a wide array of differ­ ent techniques to suit their particular require­ ments. In this paper we provide an analysis of some of the central issues in representing and reasoning with interval probabilities. At the focus of our analysis is the probability cross­product operator and its interval gen­ eralization, the cc­operator. We perform an extensive study of these operators relative to manipulation of sets of probability distribut­ tions. This study provides insight into the sources of the strengths and weaknesses of various approaches to handling probability intervals. We demonstrate the application of our results to the problems of inference in in­ terval Bayesian networks and projection and evaluation of abstract probabilistic plans.

In this paper we propose to model a negotiator’s decision-making behavior, expressed as preferences between an offer/counter-offer gamble and a certain offer, by learning from implicit choices that can be inferred from observed negotiation actions. The agent’s actions in a negotiation sequence provide information about his preferences and risk-taking behavior. We show how offers and counter-offers in negotiation can be transformed into gamble questions providing a basis for inferring implicit preferences. Finally, we present the results of experiments and evaluation we have undertaken.

Effectively selling products online is a challenging task. Today's product domains often contain a dizzying variety of brands and models with highly complex sets of characteristics. This paper addresses the problem of supporting product search and selection in domains containing large numbers of alternatives with complex sets of features. A number of online shopping websites provide product choice assistance by making direct use of Multi-Attribute Utility Theory (MAUT). While the MAUT approach is appealing due to its solid theoretical foundations, there are several reasons that it does not fit well with people's decision making behavior.This paper presents an approach designed to better fit with people's natural decision making process. The system is called VMAP for Visualizing Multi-Attribute Preferences. VMAP provides on one screen both a multi-attribute preference tool (MAP-tool) and a product visualization tool (V-tool). The product visualization tool displays the set of available products, with each product displayed as a point in a 3D attribute space. By viewing the product space, users can gain an overview of the range of available products, as well as an understanding of the relationships between their attributes. The MAP-tool integrates expression of preferences and filter conditions, which are then immediately reflected in the V-tool display. In this way, the user can immediately see the consequences of his expressed preferences on the product space.The VMAP system is evaluated on a number of factors by comparing users' subjective ratings of the system to those of a more traditional MAUT product selection tool. The results show that while VMAP is somewhat more difficult to use than a traditional MAUT product selection tool, it provides better flexibility, provides the ability to more effectively explore the product domain, and produces more confidence in the selected product.