3/6/96

CONTACT: Stanford University News Service (650) 723-2558 COMMENT: Maureen Gassman, Alfred P. Sloan Foundation (212) 649-1649
Mary G. Baker, Computer Science & Electrical Engineering (415) 725-3711
mgbaker@cs.stanford.edu
Dale G. Drueckhammer, Chemistry, (415) 723-9541; dale@forsythe.stanford.edu
John H. Griffin, Chemistry, (415) 723-3625; jgriffin@leland.stanford.edu
Daphne Koller, Computer Science, (415) 723-6598
Jeffrey Zwiebel, Business, (415) 723-2917; fzwiebel@gsb-lira.stanford.edu

Sloan Foundation names five from Stanford for research fellowships

STANFORD -- The Alfred P. Sloan Foundation has selected five young faculty members from Stanford for its 1996 Research Fellowships. Four are assistant professors: Mary G. Baker of computer science and electrical engineering; Daphne Koller of computer science; and Dale G. Drueckhammer and John H. Griffin, both of chemistry. The fifth, Jeffrey Zwiebel, is associate professor of finance in the Graduate School of Business.

The five are investigating how enzymes work, how conflicts arise in corporations and universities, how portable computers could stay hooked into a network no matter where they go, and more. Stanford's new Sloan Fellows were among 100 researchers in physics, chemistry, computer science, mathematics, neuroscience and economics chosen for a boost at the early stages of their careers because of their "exceptional promise to contribute to the advancement of knowledge."

Each will receive a $35,000, two-year grant to pursue whatever lines of inquiry are most interesting. "This flexibility is often of great value to young scientists who are at a pivotal stage in establishing their own independent research projects," according to an announcement of the awards prepared by the Sloan Foundation.

More than 3,000 young researchers have received Sloan Research Fellowships since the program's inception in 1955. Hundreds have gone on to earn prestigious awards and honors, including 17 Nobel prizes.

Mary G. Baker

When Baker takes her portable notebook computer to the library, she unhooks it from the Ethernet connection in her office, but never loses contact with her local computer network - thanks to a system developed in her lab that keeps her logged on seamlessly as she switches from cable-based to radio-based communication.

The Baker lab's mobile computing testbed, nicknamed MosquitoNet, is aimed to do far more: She wants portable computers to be truly mobile, connected to a network wherever they go. This "ubiquitous connectivity" requires work on many levels, as the lab devises ways to allow equipment and software to adapt automatically to the vastly different characteristics of the networks encountered while traveling around. For example, Baker and her students are developing protocols to allow visitors to distant networks - say a Stanford student logging on in Japan - to interact fully with their home networks without imposing on the host for computer resources and time.

Another goal of the research is to develop self-configuring mobile local networks, simple enough so that non-experts will use them to link up on an ad hoc basis. Eventually, doctors and medical detectives tracing a fast-spreading epidemic over many remote villages will be able to use this technology to link their portable computers in a wireless data-sharing network independent of any computer server or local power source.

The Sloan grant is "substantial," Baker said, because it comes with no strings attached. She said she'll use it for projects for which it is hard to find funds, perhaps to test mobile link-ups that come with high fees, like cellphones, or to attend distant conferences and test MosquitoNet over long-range connections.

Dale G. Drueckhammer

Drueckhammer's group combines the skills and knowledge of organic chemistry with techniques from biotechnology and enzymology, to look at how enzymes work and how their working can be exploited in new chemical processes.

Two of his current projects involve mirror image isomers. These molecules have a single curvature or orientation when they are made in nature, but when they are manufactured in a test tube they form as an equal mixture of two mirror image shapes, like a left hand and a right hand. This "handedness" is important in medicine because the body recognizes only one "handed" molecule as an effective drug. The molecule's mirror image is useless and may sometimes be harmful. Thus drug companies must process out the wrong-handed molecules when they manufacture a drug.

Borrowing tricks from nature, Drueckhammer has found ways to take enantiomers, mirror-image isomers, and convert them all to the desired handedness. He also has found new ways to do synthetic chemistry with aldolases, enzymes with a more complicated mirror image symmetry.

However, he said he plans to use his Sloan grant to fund new work in a more fundamental research area. Drueckhammer was the first to learn how to manipulate the structure of CoenzymeA, or CoA, a co-factor that is required by about 4 percent of all enzymes in order to cause a reaction. That process - enzymes causing changes in biological molecules - is one of the fundamental processes of life.

"By manipulating the co-factor's structure, we can alter specific interactions between an enzyme and the CoA molecule," Drueckhammer said. He plans to use this technique to analyze the interactions that occur as an enzyme causes a reaction.

John H. Griffin

"This is flexible funding," Griffin said of the Sloan grant, "so I'll use it to push new frontiers." Griffin was also honored last month as a recipient of one of six Terman Fellowships (see Stanford Report, Feb. 28, p. 7), for his research on catalytic antibiotics - drugs that destroy foreign bacteria without self-destructing - and for work to identify the DNA and amino acid sequence of cyclase enzymes.

Griffin said there are frontiers to explore in both directions. Because the Sloan grant encourages trials of risky new ideas, he said his lab may attempt to do something no one else has ever done, using their new knowledge about cyclase enzymes.

Starting with one oxidosqualene cyclase enzyme, they will attempt to change its product specificity (i.e. what the enzyme makes) through mutation. "No one has ever done this without simultaneously altering the substrate specificity (i.e. what the enzyme uses)," Griffin said. "We want to see whether we can alter the cyclase so that it makes a different product using the same starting material.

"For example, the human cyclase forms lanosterol, which is subsequently converted to cholesterol, which is an essential membrane component," Griffin said. "On the other hand, plants start with cyclase and make cycloartenol, then make their membrane sterols from cycloartenol rather than lanosterol. We believe that rather simple changes in structure can convert a lanosterol-forming cyclase into a cycloartenol-forming cyclase, and vice versa.

"In this way, we hope to recapitulate some of the evolutionary processes that cyclase enzymes have undergone in bacterial, fungal, plant and vertebrate cells," he said.

Daphne Koller

At the core of Koller's work in artificial intelligence are two competing factors: rational decision making and uncertain knowledge. Koller works to create computer systems that reason and act under uncertainty. Among other applications, she has applied her work to a freeway traffic surveillance system, a management decision support system, and a poker coach that tells when to fold 'em and when to hold 'em - or, more specifically, when to bluff with a low hand.

Koller draws on mathematical tools from decision theory and probability theory to help her apply fundamental notions of rationality to problems arising in computer science. She adds calculations from a discipline seldom linked with computers - economics. Traditionally, these theories have been used to create abstract models. Economists use models to predict how groups will act, on average, in an economic system. Game theorists predict optimal overall strategy, but usually cannot tell the optimum move at any point in the game.

"If we want the models to work for us, we have to program a computer," Koller said. Until recently it was impractical to adapt an economic model to a real-life decision-making problem like automated freeway traffic control because it required too much computational power and time to deal with all the complexities. Koller designs faster, more efficient algorithms to make the principles of game theory, decision theory and economics usable.

"As tools to do this, I look at how people in real life solve complex problems with uncertain knowledge," Koller says. For example, Koller has worked on a problem called "dynamic re-focus of attention." On a freeway, where the driver must make rapid decisions in uncertain conditions with a lot of unknown factors at play, he is constantly re-focusing attention on the more important factors involved - like the car that just cut in front of him. In a joint project with the University of California-Berkeley, Koller is working on a program that could prompt drivers to make the right decisions - the ones that unclog traffic instead of making it worse.

In a test of a program that may aid corporate decision makers, Koller and her students taught the computer to play poker. There are many computer poker games, but those that use rational game theory to deduce optimal strategy for each of four players typically use so much computer power that they can only handle a deck of about four cards. With algorithms developed in her lab, Koller said, "we can solve games where you have a full deck of cards, each person gets one card, and the game continues through a couple of rounds of bidding."

In the process, she's been able to prove something that game theorists predicted, but never had the computer power to demonstrate: that bluffing in poker is not only good psychological strategy, it turns out to be economically rational.

Computer scientists have been included among Sloan Fellows for only a few years. Koller said, "It's a great honor to be named as one of the early fellows."

Jeffrey Zwiebel

Economists traditionally have looked at a business firm or corporation as a "profit-maximizing black box," Zwiebel said. While organizational behavior researchers took the box apart to see what made the people inside tick, economic theory assumed that the firm as a whole would take the production opportunities it was given and use them to maximize profits. Zwiebel is one of a new school of economists using the tools of economic analysis to examine actions inside the corporate structure that can influence these decisions.

For example, he has examined how economic incentives may lead to herd behavior among managers - the tendency to copy the behavior of others even when good information suggests that's not the best action to take. He has looked at how corporate debt forces managers to spend resources wisely and how extra money within the firm tends to be spent inef