Above all, Koller believes her work "can provide a framework to use as a starting point for answering tough questions." She says she hopes to resolve "the complex network of interactions between many genes, proteins, metabolites and signals" using computational modeling, and is optimistic that she can help to identify patterns of gene expression across species as her methods are further developed.
When asked what she plans to do with the award, Koller said that the matter will require significant thought. "It's really a major decision, and they give you some time to think about it." The news always comes as a big surprise, since the selection process is confidential and no notification is given to fellows until final selections are made.
Koller's research tackles questions of how complex information with high levels of uncertainty can be approached using algorithms, probabilistic modeling and other computational methods. These tools strive to represent knowledge and reasoning at the intersection of traditional logic and subjective judgment, and have far-reaching implications in the fields of artificial intelligence and biomedical and genetic data analysis.
A significant contribution of Koller's work is the expansion of Bayesian networks -- reasoning frameworks that deal with uncertainty -- by showing how they can be organized into logical, object-oriented hierarchies. She has advanced this concept by implementing "probabilistic relational models," which blend logical and statistical representations in ways that employ standard deductive reasoning.
The strength of this approach is that it has the ability to represent models of the world with significant complexity. "So far, most models have been fairly self-contained -- but the world does not work that way," Koller said.
These models have already seen service in analyzing the yeast genome, where they helped sift through staggering amounts of gene-expression data to tease out useful information. A recent paper by her research group, published in Nature Genetics, applied a probabilistic model to identifying modules of co-regulated genes and identified regulatory roles for several proteins whose functions had been previously uncharacterized.
Koller received her bachelor's (1985) and master's (1986) degrees from the Hebrew University of Jerusalem, followed by a doctorate (1993) from Stanford. She was a postdoctoral fellow at the University of California-Berkeley from 1993 to 1995, after which she joined the Stanford Department of Computer Science as an assistant professor. In 2001, she earned her associate professorship. Her work has appeared in publications including Science, Nature Genetics, Artificial Intelligence, Bioinformatics and Games and Economic Behavior.
Koller is one of two Stanford faculty members to win a 2004 MacArthur Fellowship. The other is Julie Theriot (see separate press release). Their awards bring to 22 the number of current Stanford MacArthur winners.
The MacArthur Foundation selects individuals from varied pursuits to receive this prestigious fellowship. In addition to Koller and other scientists, this year's recipients include a farmer, a poet, a ragtime pianist and a high-school teacher. Daniel J. Socolow, director of the MacArthur Fellows program, is excited to see "such a collection of decidedly bold and risk-taking people who are changing our landscape and advancing our possibilities."
Matthew Early Wright is a science writing intern at Stanford News Service.
BY Matthew Early Wright
Contact: Dawn Levy, News Service: 650-725-1944, dawnlevy@stanford.edu
Comment: Daphne Koller, Computer Science: 650-723-6598, koller@cs.stanford.edu
Editor Note: A photo of Koller is available at http://newsphotos.stanford.edu.
Relevant Web URLs: DAPHNE KOLLER'S RESEARCH GROUP http://dags.stanford.edu/