Gilles Laurent - US grants

9350374 Laurent Dr. Laurent's goal is to develop an interdisciplinary research program to study information processing and circuit design in the brain, and understand fundamental aspects of neural computation. Dr. Laurent's teaching plans are to broaden the scope of Neurobiology teaching in the Caltech Computation and Neural Systems graduate program, and to create a new interdisciplinary course on Motor Systems.

9511764 Laurent Although brains and computers are thought to be quite similar, the mechanisms which they use to encode and store information are fundamentally different. This is partly because computers were developed by engineers using tools of formal logic to solve rather limited sets of problems, while brains were developed over millions of years with selective pressure of evolution as the only engineer. Indeed, brains evolved to increase the chances of survival of their "owners", and thus primarily to facilitate the detection of prey, predators and mates. Consequently, they succeeded in solving the very complex problems of pattern recognition, problems which present computers are notoriously poor at handling. How do we recognize something as complex as a face or an odor, how do we store their infinite number of possible appearances, how does our brain represent them with neuronal activity patterns? To understand this, we must understand coding in the nervous system; we need to determine the rules that underlie stimulus representation in the brain. Conventional views of coding by the brain hold that information resides in the average number of impulses produced per unit time (rate coding); the role of relative timing of impulses produced by groups of neurons has not been carefully evaluated. The work of this young scientist focuses on circuits that process smells. Experiments are designed to determine whether the relative timing of neuronal impulses (temporal coding) is a potential carrier for information.

Humans can recognize objects and scenes using their senses. The ability of learning the appearance of a great number of objects, organizing them into categories, and quickly recognizing them later is an important skill for survival. Replicating such ability in machines would be extremely useful in a great number of scientific and industrial applications such as automatic exploration of databases of medical images, diagnostics and quality control in industrial plants, automatic classification of images and sounds on the web.

The aim of this study is to develop a theory of recognition that is applicable any type of sensory data and where no supervision is required for learning and categorization.

The approach is probabilistic: object categories are modeled by probability density functions on part appearance and object shape. Detection and recognition are formulated as statistical inference problems. Unsupervised learning of object categories is approached using maximum likelihood. In order to motivate and test the theory the investigators will engage in three applications: automatic classification and retrieval of objects from image databases, of human actions from movies, and of neuronal signals associated with perceptual tasks.

EIA 0130746-Giles J. Laurent-California Institute of Technology - Biological computation in olfactory systems: experiments, theory and analog circuit modeling

Pattern recognition (e.g., face, voice recognition) remains one of the most difficult problems for computer science. Odors are complex physical objects (blends of molecules) and yet, are perceived as singular objects (e.g., coffee, jasmine); because olfactory circuit designs appear similar across animal species (from insects to mammals), olfaction constitutes a potentially ideal system to identify key solutions to pattern recognition. The investigators are an interdisciplinary team from Biology, Physics, Computer Science and Electrical Engineering from Caltech and from the University of California-San Diego. They are investigating the basic computational principles of olfactory processing and recognition in animals, with the long-term goal to build "intelligent" pattern storage and recognition devices, designed using rules inspired by neurobiology.

We plan to study how to build simple and later, scale up, both in number of processors and in frequency, "electronic antennal lobes and mushroom bodies" so that one might, in the future, adapt and exploit the design of biological pattern recognizers and provide new computational paradigms for human uses.