Global organization of the lexicon

(preprint available in publication list)

with M. Sigman

The lexicon consists of a set of word meanings and their semantic relationships. A systematic representation of the English lexicon based in psycholinguistic considerations has been put together in the database Wordnet in a long-term collaborative effort1. We present here a quantitative study of the graph structure of Wordnet in order to understand the global organization of the lexicon. We find that semantic links follow power-law, scale-invariant behaviors typical of self-organizing networks. Polysemy, the ambiguity of an individual word, can act as a link in the semantic network, relating the different meanings of a common word. Inclusion of polysemous links has a profound impact in the organization of the semantic graph, converting it into a small world, with clusters of high traffic (hubs) representing abstract concepts. Our results show that polysemy organizes the semantic graph in a compact and categorical representation, and thus may explain the ubiquity of polysemy across languages.

Simple motor gestures for birdsong

(preprint available in publication list)

with T. Gardner, R. Laje, M. Magnasco and G.B. Mindlin

Vocal learning is relatively rare in the animal kingdom, and the complexity of birdsong combined with the feasibility of studying bird vocalizations in a controlled environment makes birdsong a unique system of general interest. The structure of song in all of its developing forms holds significant potential as a window into neural processes of learning and memory. However, inferring neurla organization from the acoustic waveform is complicated by the fact that neural control produces song through a vocal apparatus which is not well understood. We present here a model of sound production in the songbird's vocal organ, and find that the transformation from muscular control to acoustic waveform can be particularly simple for a wide class of vocalizations typified here by canary song. This observation is an explanation for the relationship between different syllables in simple physical terms. If the core of the motor program for song is as simple as we propose, the prospects are improved for understanding many of the intriguing questions about the neural basis of song learning. Accepted, Phys. Rev. Lett. (2001).

On a common circle: natural scenes and Gestalt rules

with M. Sigman, C.D. Gilbert and M. Magnasco

To understand how the human visual system analyzes images, it is essential to know the structure of the visual environment. In particular, natural images display consistent statistical properties that distinguish them from random luminance distributions. We have studied the geometric regularities of oriented elements (edges or line segments) present in an ensemble of visual scenes, asking how much information the presence of a segment in a particular location of the visual scene carries about the presence of a second segment at different relative positions and orientations. We observed strong long-range correlations in the distribution of oriented segments that extend over the whole visual field. We further show that a very simple geometric rule, cocircularity, predicts the arrangement of segments in natural scenes, and that different geometrical arrangements show relevant differences in their scaling properties. Our results show similarities to geometric features of previous physiological and psychophysical studies. We discuss the implications of these findings for theories of early vision. Proc. Natl. Acad. Sci. 98: 1935-1940 (2001).

Unsupervised learning and adaptation in a model of adult neurogenesis

with L.T. Petreanu, A. Alvarez-Buylla and M.O. Magnasco

(preprint available in publication list)

The olfactory system provides an attractive system in which to model the possible role of adult
neuronal replacement. The adult olfactory bulb (OB) receives a constant flow of neuronal precursors from the subventricular zone (SVZ), which differentiate into inhibitory interneurons (granule cells, GR). There is evidence that cell birth and migration are not dependent on olfaction or activity  within the OB. However, activity dependent mechanisms seem to play an important role in the recruitment or
survival of postnatally generated GR. Mitral and tufted cells (M-T), the OB output neurons, receive
their synaptic input from the olfactory receptors within single glomeruli. Odor information seems to
be encoded in the spatial and temporal responses of M-T cells. GR cells are thought to play an important role in lateral inhibition and in the tuning of M-T responses to different odors. The discrimination of odors is importantly dependent on the progressive tuning of the combinatorial response of M-T. Inspired by the  organization of the olfactory bulb, we present here a simple computational model - that through replacement of GR - predicts a progressive fine tuning of olfactory
discrimination. The model is self organizing, suggesting that by a random integration of newly formed GR and their survival conditioned by correlative activity, the representation of each odorant is dramatically segregated. The model predicts a constant adaptation to changes in the olfactory environment. These properties may be important in the case of olfaction, given the multidimensionality of the stimuli and the constant change in the olfactory environment (Soc. Neurosci. Abst. 2000).

Noise in Neurons is Message-Dependent

with M. Sigman, J-M. Alonso, L. Martinez, D. Chialvo and M. Magnasco

Go to the Stochastic Dynamics Page.

Variability in their response is a conspicuous property of neurons. It is shown how such variability (i.e., the precision of action potential timing) depends on the particulars of the input that drives them. The results suggest that a simplification usually made in the application of information theory to neural processing is violated: noise is not always independent of the message. To ensure the contrary neurons should work at a regime of fast threshold crossings otherwise, real neurons could change its voice with each tune ( PNAS, May 9, 2000; vol. 97, no. 10; 5557-5561).

Same number of spikes but different reliability in cat's LGN.

Noise-induced memory in extended excitable media

with D. Chialvo and M. Magnasco

Go to the Stochastic Dynamics Page.

Towards a Song Code: Syllabic Representation in the Canary Brain

with Sidarta Ribeiro, Claudio Mello and Marcelo Magnasco

We show that presentation of individual syllables derived from a whole canary song results in distinct expression patterns of the immediate early gene ZENK in the caudomedial neostriatum (NCM) of adult canaries. Information on the spatial distribution and staining levels of labeled cells is enough to provide for a classification of ZENK expression patterns that: 1) accords to the intra and interfamily relationships between the corresponding stimuli and, 2) confers salience to natural stimuli over artificial ones. Moreover, complex syllable maps cannot be reduced to any linear combinations of simple syllable maps. These properties arise from the collective response of NCM's neurons to auditory stimuli, rather than from the behavior of single neurons in NCM. We believe the syllabic representation described here constitutes an important step towards deciphering the rules underlying the auditory representation of birdsong.

-Neuron, Vol. 21, 359-371, August, 1998.

The panel shows an artistic rendering of an ensemble of neurons expressing the immediate-early gene ZENK in NCM, different colors representing different levels of expression.

(More figures to be here shortly)

An automated system for the mapping and quantitative analysis of immunocytochemistry of an inducible nuclear protein

with Sidarta Ribeiro, Claudio Mello and Marcelo Magnasco

We describe here an automated system that accurately maps tissue sections stained by immunocytochemistry for an inducible nuclear protein. The sections are scanned with a computer-controlled microscope setup hooked to a CCD camera. Raw images captured at high resolution are filtered using highly selective criteria for the recognition of labeled cell nuclei. The total population of recognized labeled nuclei is then divided into separate bins, according to their labeling intensities. Finally, information about both the position and labeling intensity of labeled nuclei is represented in average density maps. The system was optimized for the quantitative mapping of neuronal cells expressing the inducible gene ZENK in the brain of songbirds, in response to stimulation with song, but should be of general applicability for the mapping of inducible nuclear proteins. - J. Neurosci. Methods 87, 147-158 (1999).

The panel shows the "navigator", the program that scans the sections that will later be analyzed to detect labeled cell nuclei.