MIT Reports to the President 1996-97


Learning and Memory
Tomaso Poggio's lab continues to focus on the problem of learning at the level of theory, applications and neuroscience. During the last year, work on new classification techniques called Support Vector machines led to new theoretical results that establish a close relation between SVM and sparse representations of signals and images. Sparsity seems to be a key constraint in characterizing and discovering good representations for sensory signals. The new work on sparse representations has already let to the development of an architecture that can "learn" to perform rather difficult object detection tasks in images, such as detection of a person in a static image.

Matthew Wilson's laboratory successfully adapted multiple electrode recording techniques to the mouse and, in collaboration with Susumu Tonegawa's lab, completed a recent study characterizing neuronal activity in the hippocampus of the first tissue-specific genetic knockout animal. The characterization of a genetically altered animal lacking NMDA receptors within the CA1 sub-region of the hippocampus succeeded in demonstrating the relationship between synaptic plasticity and spatial memory. This combination of genetic, neurophysiological, and behavioral approaches promises to establish a firm link between molecular mechanisms of learning and memory and the processes of higher cognition.

In the past year, Earl Miller's laboratory has made several discoveries concerning the neural basis of working memory and its role in cognition. They identified prefrontal cortex neurons that integrate information about an object's identity ("what") and its spatial location ("where"). "What" and "where" are known to be processed separately in the visual system. Thus, these neurons may comprise a crucial link that allows actions to be directed toward objects.

Guosong Liu studies mechanisms of synaptic plasticity by examining the factors that regulate synaptic strength at single synapses. In particular, modulation of the number the number and properties of glutamate receptors can play a fundamental role in the process of synaptic plasticity. He showed that the average response size at multiple synaptic sites of a single neuron is remarkably constant, while the average size at single synapses varies four-fold across different cells, decreasing with increasing synaptic innervation. Recent data suggests that the difference may be due to differential expression of glutamate receptor subtypes.

Edward Adelson's research in visual motion perception has utilized a set of novel illusions to study the rules by which information is combined across space to constrain the interpretation of moving contours. He finds that junctions -- the place where contours meet -- offer important information, and that many phenomena can be explained by assuming that the visual system is using a Bayesian approach to estimating motion. A large pool of existing psychophysical data can thus be understood in a simple framework; this approach also offers powerful tools for segmentation of video sequences by machine vision.

Mriganka Sur's laboratory has shown that even the earliest stages of visual cortex -- primary and secondary visual cortex -- process not only simple visual stimuli but also complex images such as illusory contours. Neurons in these areas respond to the orientation of illusory contours and are organized into systematic modules that respond best to a particular stimulus orientation.

Barton Anderson's lab has focused on two types of problems in visual psychophysics. One problem involves the segmentation of images along occluding contours; the other involves the segmentation of images into a layered representation of surface causes. There has been substantial empirical and theoretical progress in both of these domains in the lab within the past year, including the understanding of computational principles underlying layered image representation. This work has led to a series of experimental discoveries that reveal the generality of the theory.

Neural and Cognitive Development
Gerald Schneider's laboratory, in collaboration with Susumu Tonegawa, made an important breakthrough in research on regeneration of severed nerve fibers in the mammalian brain. Their work uncovered a key role of the proto-oncogene bcl-2 in promoting the growth and regeneration of axons. When transcription of this gene is down-regulated in the developing retinal ganglion cells, the large majority of the axons of these cells lose their ability to re-grow after injury. But, if an animal is genetically altered so the Bcl-2 protein is produced by the neurons of older mouse pups, then the loss of regenerative capacity does not occur.

Elizabeth Spelke's recent studies have focused on the development of object representation, spatial memory and number. Concerning spatial memory, she is investigating the spatial representations used by 1.5 to 2- year old children to determine their own position. Their studies reveal (a) a close homology between the representations guiding reorientation in young children and in other mammals such as rodents, and (b) a developmental change in human spatial representations, whereby the distinct representations guiding the young child's navigation become interconnected, permitting more flexible, and distinctly human, spatial performance.

Kenneth Wexler's research on early inflectional/grammatical development has been expanded to include many new phenomena, including Specific Language Impairment (SLI). He has found evidence for the heretibility of tensing behavior, perhaps the first evidence in normal children for the heretibility of a particular aspect of grammatical variation. A new development concerns evoked potentials: the lab has shown for the first time that a distinct module of grammar, binding theory, elicits a P600 "syntactic" violation signature, lining it up with more simple P600 violations. Moreover, and strikingly, a syntactic binding violation yields a P600, but a non-syntactic "pragmatic violation does not.

Steven Pinker published the results of a project on neural dissociations between grammatical computation and lexical lookup. Regular suffixation of "-ed", is rule-governed and freely extended to new verbs (e.g., "fax-faxed"), so it must be computed by a mental rule of grammar; irregular inflection is unpredictable (compare "sink-sank" with "think-thought") so the forms must be memorized in the mental dictionary. Examination of patients with distinct neurological syndromes points to a substrate for grammatical processing in a procedural system located in basal-ganglia/frontal-cortex circuits, and to a substrate for lexical memory in a declarative memory system located in limbic/posterior-cortex circuits.

Edward Gibson has proposed a major new theory of the relationship between working memory and human language comprehension: the Syntactic Prediction Locality Theory (SPLT). The SPLT starts with the hypothesis that the human language processor is a predicting machine: It predicts the categories that are necessary to complete the current input string as a grammatical sentence. There is a memory cost associated with each of these predictions, so the more predictions there are at a particular point, the harder it is to process at that point. Although the structure of languages is highly variable cross-linguistically, current evidence suggests that the SPLT applies uniformly across all languages.


Of the 15 Ph.D.s granted in the last year, 11 have postdoctoral positions, and the remainder will be doing research in industrial settings. Applications to the program have been consistently increasing in quality and number and 12 new students were accepted from a pool of 196.

Several new graduate courses have been created, including: Neurology, Neuropsychology and Neurobiology of Aging, Molecular and Cellular Mechanisms of Neuronal Communication, Mechanisms of Synaptic Transmission and Plasticity, Evolutionary Psychology, and Cognitive Artifacts and Architectures. In addition, the department's graduate students organized a mandatory course on ethics. Topics covered included: obligatons and responsibilities of the publicly funded scientist, ethical issues involved in mentoring, use and abuse of animals and humans in research, intellectual and patent rights, ethics of the peer review and grant processes, ethical issues in the gathering of data, and ethical issues in the publication of data.

The number of majors has increased, as has the number of students who elected to concentrate or minor in Psychology. A marked increase was also noted in the number of UROPS who were being paid or working for credit. In addition, a new undergraduate course in the neuroscience track was implemented.


Barton Anderson

NIH Shannon Award

Peter Dayan:

Surdna Junior Faculty Award

Earl Miller:

NIH First Award

Whitehall Foundation Fellowship

Steven Pinker:

Linguistics, Language and the Public Service Award from the Linguistics Society of America

Tomaso Poggio:

Member, Daimler-Benz Circle Member Group

Elected to the American Academy of Arts and Sciences (1997)

Elizabeth Spelke

Elected to the American Academy of Arts and Sciences (1997)

Hermann Steller

Appointed Investigator, Howard Hughes Medical Institute

Mriganka Sur:

Teuber Scholar in the Brain Sciences

McDonnell-Pew Visiting Professor, Oxford University

Kenneth Wexler

Editor's Award, Best Paper on Language, JOURNAL OF SPEECH & HEARING

Matthew Wilson:

Middleton Neurosciences Award

Richard J. Wurtman:

32nd Annual Waldo E. Nelson Lecturer, Philadelphia, PA

Emilio Bizzi, M.D.

MIT Reports to the President 1996-97