Fabio Cozman - Research Overview

It makes sense to classify my interests into three sections: Sets of Probabilities, Probabilistic Reasoning, and Robotics. These titles reflect most of my past and present interests.

Theory of sets of probabilities (credal sets)

First, I explore the theory of sets of probabilities. There is not a single, stable name for this theory: some people use "theory of imprecise probabilities"; others say "theory of credal sets", or "Quasi-Bayesian theory", or "theory of lower expectations", or ... several other names. I believe this theory is the right tool to model statistical uncertainty, and will ultimately be the unifying foundation for inference and decision-making. Some years ago I produced a brief tutorial on this theory; you can also look at web site of the Society for Imprecise Probability Theory and Applications. I'm a founding member of this society, and currently the editor of the society's newsletter; I also helped organized some of the International Symposium for Imprecise Probabilities and Their Applications (ISIPTA) and edited some of its proceedings (see my publications).

Credal networks

I'm interested in efficient algorithms to obtain posterior quantities, with a particular interest on multivariate models with graph-theoretic representatins (in particular, the model called "credal network"). A big part of the work can be grasped through the papers: The following paper discusses credal networks that can deal with some first-order constructs, much in the way of probabilistic logic, and also presents several algorithms for inference with credal networks: The following paper also discusses the problem of inference with credal networks:

Planning under risk and uncertainty

Another interesting piece is the work on planning, where the idea is to merge "probabilistic" and "nondeterministic" planning using the theory of sets of probabilities: Here is an older effort, looking at the problem of sequential-decision making associated with observations:

Concepts of independence

I'm also interested in concepts and properties of irrelevance/independence connected to the theory of sets of probabilities. The following papers contain a sample of basic issues in the topic:

Statistical learning of sets of probabilities

A long time ago, I explored with Lonnie Chrisman the possibility of learning convex sets of probability from data; that old has been picked up by Terry Fine and co-workers.

Probabilistic reasoning

I am quite interested in probabilistic models for uncertainty modeling, particularly Bayesian networks. Several activities in this track are generously funded by HP Labs.

JavaBayes

I develop the JavaBayes system, a general purpose inference engine for graphical models; the engine can generate posterior probabilities and expectations for probabilistic models represented as directed acyclic graphs. The system is distributed freely (under the GNU license) in the spirit of fostering teaching and research. JavaBayes is now used in many university and research labs around the world. A summary is: In the process of putting together JavaBayes, I have developed a very general, yet easy to understand, inference algorithm for Bayesian networks. The method is suited for teaching due to its simplicity. You can get it:

Classification and learning

One of the most important situations where we make decisions and use our beliefs and sensory information is when we classify data. For example, based on measurements we may classify a machine into one of several categories; for example, "broken" or "functioning".

I have been looking at classification problems where we must build ("learn") a classifier using observed data. These data may be labeled or unlabeled; that is, the data points themselves may be classified or not. I am interested in methods that can learn good probabilistic classifiers from mixtures of labeled and unlabeled data. This problem is quite complex and displays several interesting phenomena. Right now the focus of the research is on learning classifiers that have Bayesian network structures.

A description of some characteristics of the labeled/unlabeled data problem is

This is joint work with Ira Cohen, at HP Labs Palo Alto. Other people at HP Labs have contributed a lot, particularly Marsha Duro and Alex Bronstein.

Probabilistic reasoning in embedded systems

While JavaBayes is a complete system, with graphical interface, parsers, etc, I've been investigating a system that is more geared towards the needs of embedded systems. The EBayes project is an effort to produce a lightweight Bayesian network engine that is appropriate to the growing market of embedded devices. A complete algorithm for probabilistic inference under time and space constraints is presented in the following paper:

Generating Bayesian networks randomly

Still on Bayesian networks, I have worked with Jaime Shinsuke Ide on the problem of testing algorithms; we have produced interesting methods for generating random Bayesian networks:

Applying Bayesian networks

Another interest of mine is the application of Bayesian networks in practical problems. I was involved, between 2001 and 2003, in a project with the University Hospital, where we try to encode medical knowledge about cardiac problems in the form of Bayesian networks.

Sensitivity analysis in Bayesian networks

Finally, I am interest in applying sensitivity analysis techniques (from the realm of robust Statistics) to Bayesian networks. I have been developing techniques that use the theory of sets of probabilities as a tool for the assessment of sensitivity in graphical statistical models:

Robotics: Teleoperation, mobile robots, automated orthosis...

Third, I was involved for a long time, in one way or another, with robotic devices, mostly with mobile robots.

During 2000-2002, I participated in an effort to develop devices that can help the disabled walk with less effort and discomfort. The project started from interactions with doctors and engineers at the Associação de Assistência a Criança Defeituosa and is supported by FAPESP.

A student involved with this project, Marco Ackermann, received the prize of Best Master Thesis in Mechanical Engineering in Brazil 2003, granted by the Brazilian Association for the Mechanical Sciences (ABCM), for this work.

My involvement with robotics started quite a while ago.

Right after my undergraduate course, I took a Master of Engineering in Brazil, and worked in the first Brazilian mobile robot, called Ariel. We produced a complete system, from the mechanical structure to the planning software; the result was very impressive and we ended up showing it off in the Jornal da Globo (Brazil's second most important TV news source at the time). Unfortunately, that material is not online. Here are two significant papers, perhaps of historic value:

I worked, for two years, in the Lunar Rover project during my PhD years at Carnegie Mellon. My main contribution to the Lunar Rover project was the Viper system, a piece of technology that was used in the Atacama mission. The Viper system, estimates position from a stream of images, by matching images to a previously constructed map of the environment. The estimator builds an occupancy map for the position of the robot; the catch is that the occupancy maps actually represents a full density ratio familiy of distributions which generate both the estimates and the confidence on the estimates. The system is described at

There is also a description of an old version of the Viper system at

The vision algorithms developed for the Viper system reported in the following papers.

During a few years at CMU I worked with the Ratler robot. We actually had it rolling for some fifty kilometers in our outdoor tests; you can take a look at the following paper.

I also worked on a few other problems.

Some years ago I produced a line linker based on the Akaike Information Criterion (AIC), which was distributed in the net.

Another aspect of my work was the investigation of celestial data as a source of position estimates for mobile robots:

And finally, another twist in this work was the study of atmospheric scattering as a clue for depth in outdoor environments; as far as I know, the first study of scattering in the context of image understanding.

I was interested for some time in the problem of calculating bounds for dynamical systems; there is a huge literature in this area. I have published some work on the specific topic of manipulating ellipsoidal models of error in Robotics:

fgcozman@usp.br