Following the outstanding investigations of S. Ramón y Cajal (1909) the histological architecture of the central nervous system (CNS) was envisaged as an intricate assembly of neuronal circuits. In adult specimens these circuits were considered to be organized on the basis of stable permanent connections linking a practically unvarying number of nerve cells. On the other hand, the occurrence of plastic phenomena like memory and learning remained as a serious challenge to the proposed fixity of neuronal connections. There is little doubt however, that in some systems and organs hard-wiring is essential to achieve a proper performance. A typical example of this kind of hard-wiring is represented by the organization of the so-called neural superposition eyes that evolved in muscoid dipterans. In this peculiar type of compound eye there are three functionally interdependent layers consisting of discrete units of dissimilar origin: a) the corneal lattice composed of hundreds of minute lenses, b) the photoreceptors lattice, composed of discrete photoreceptor packets or ommatidia and c) the synaptic lattice composed of hundreds of polysynaptic units termed optical cartridges.
Any attempt to understand the functional design of the neural superposition eye has to answer the following critical question: How are the topological relationships between the different kinds of units found in these eyes? Corneal lenses , photoreceptors and associated neurons constitute a very efficient system that has remained unchanged since the Cambrian period. How are these different anatomical units functionally connected? Our investigations (Trujillo-Cenóz and Melamed , 1966a-b) together with those of Kirschfeld (1967) and Braitemberg (1966) (for a complete review see Trujillo-Cenóz, 1985) have contributed to address the issue.
In this kind of eyes the wiring pattern has evolved in such a way that the synaptic terminals arising from six photoreceptors looking at the same point in space are gather together in the same synaptic unit or optical cartridge. In addition, the photoreceptor axons are twisted 180ş along their path to the synaptic layer in such a way to compensate the image inversion resulting from the passage of light through the corneal layer. It is evident that in this example hard-wiring is essential for securing the appropriate eye performance.
At the other end of the spectrum there are neuronal circuits that are built and rebuilt periodically after birth. The studies performed on the song centers of canaries by Nottebohm and his colleagues (Nottebohm et al. 1994; Rasika et al. 1994) constitute a paradigmatic example. At present, it is generally accepted that in both the CNS of adult vertebrates and invertebrates there are niches of cells that have retained the embryonic capability to divide and to originate new neurons and glial cells ( for a recent review see Kempermann et al. 2004)). We have been interested recently on this topic and we have explored the presence of mitotically active cells in the spinal cord of turtles. The CNS of this ancient tetrapod has an outstanding resistance to anoxia, a feature that greatly facilitates the preparation of living slices and the combination of anatomical, molecular and electrophysiological techniques. Our studies have revealed that dividing cells predominate in the central region of the spinal cord , particularly around the central canal (CC) (Fernández et al. 2002; Radmilovich et al. 2003) . We have also found in this region cells that express proteins found in immature neurons and are able to generate action potentials. Intermixed with these immature neurons there are non-spiking cells that express proteins usually found in ependymal and glial cells. Contrasting with the immature neurons, non-spiking cells appeared dye-coupled forming conspicuous clusters around the CC. Interestingly, the non-spiking cells that expressed glial proteins, exhibited the morphology of radial glia (RG) (Russo et al. 2004). Our studies strongly suggests that in the spinal cord, like in other regions of the vertebrate brain, RGs may represent progenitor cells that are able to divide and to generate neurons during adulthood.
References:
Ramón y Cajal, S. Histologie du Systčme Nerveux de lhomme et des vertébrés. (1909). Edited by the Consejo Superior de Investigaciones Científicas . Instituto Ramón y Cajal. Madrid, 1952
Trujillo-Cenóz, O., Melamed, J. (1966-a) Electron microscope observations on the peripheral and intermediate retinas of Dipterans. In: The functional organization of the Compound eye . Symp. Wenner Gren Center (1965): edit. C.G. Bernhard. pp. 359-361. pergamon Press, London
Trujillo-Cenóz, O., Melamed, J. (1966-b) Compound eyes of Dipterans: anatomical basis for integration-an electron microscope study. J. Ultrastruct. Res. 16: 395-398
Kirschfeld, K. (1967) Die projection der optische Unwelt auf das Raster der Rhabdomere des Komplexauge von Musca . Exp. Brain. Res. 3: 248-270
Braitemberg, V. (1966) Unsymmetrische Projektion der retinula zellen auf die lamina ganglionaris bei der fliege Musca domestica . Z. Vergl. Physiol. 52: 212-214
Trujillo-Cenóz, O. The eye: Development, Structure end Neural Connections. In: Comprehensive Insect Physiology Biochemistry and Pharmacology. Vol 6.(1985) Edit. G.A. Kerkurt and L.I. Gilbert. pp 171- 223.
Nottebohm, F., OLoughin, B., Gould, K. (1994) The life span of new neurons in a song control nucleus of the adult canary brain depends on time of year when these cells are born. Proc. Natl. Acad. Sci. USA 91: 7849-7853
Rasika, S., Nottebohm, F., Alvarez-Buylla, A. (1994) Testosterone increases the recruitment and/or survival of new vocal center neurons in adult female canaries. Proc. Natl. Acad. Sci. USA 91: 7854-7858
Kempermann, G., Wiskott, L. Gage, F.H. (2004) Functional significance of adult neurogenesis. Curr. Opin. Neurobiol. 14: 186-191
Fernández, A., Radmilovich, M., Trujillo-Cenóz, O.(2002) Neurogeneisi and gliogenesis in the spinal cord of turtles. J. Comp. Neurol. 453: 131-144
Radmilovich, M, Fernández, A., Trujillo-Cenóz, O.(2003) Environment temperature affects cell proliferation in the spinal cord and brain of juvenile turtles.J. Expt. Biol. 206: 3085-3093
Arbib will give two talks concerning:
Schema theory provides a framework for functional analysis of perceptual structures and distributed motor control, and is exemplified by the VISIONS system for visual scene understanding, and the relation of this system to action and working memory. The HEARSAY system exemplifies the use of schemas and blackboard architecture for speech understanding. These ideas provide the framework for placing brain systems which support language within a broader framework of action-oriented perception.
The Mirror System Hypothesis for the evolution of the language-ready brain is grounded in the homology of Broca's area with premotor area F5 of the macaque, but the functioning of F5 relies on its embedding in a larger network including parietal, temporal and prefrontal cortex. We posit the following evolutionary stages: 1) the mirror system for grasping. 2) Its extension to permit imitation. Imitation is seen as evolving via a so-called "simple" system such as that found in chimpanzees (which allows imitation of complex "object-oriented" sequences but only as the result of extensive practice) to a so-called "complex" system found in humans (which allows rapid imitation even of complex sequences, under appropriate conditions) which supports pantomime. 3) This is hypothesized to provide the substrate for the development of protosign, a combinatorially open repertoire of manual gestures, which then provides the scaffolding for the emergence of protospeech (which thus owes little to non-human vocalizations), with protosign and protospeech then developing in an expanding spiral. It is argued that these stages involve biological evolution of both brain and body. 4) By contrast, it is argued that the progression from protosign and protospeech to languages with full-blown syntax and compositional semantics was a historical phenomenon in the development of Homo sapiens, involving few if any further biological changes.
We present two explicit computational models of macaque F5 and related areas, the FARS model of the macaque "canonical system" for grasping and the MNS model of the macaque "mirror system" for grasping. These show the plasticity of the larger networks, with implications for the possible plasticity of brain mechanisms supporting language in the human brain. We close by sketching directions for an action-oriented neurolinguistics.
Slides:
El Sistema Neuronal Espejo para la Prension (Chile 2000).
From Action Recognition to Language.
All vertebrate brains have a cerebellum, and most of them have one or more additional structures that are histologically similar to the cerebellum. The cerebellum-like structures include the medial octavolateral nucleus in most aquatic vertebrates; the dorsal octavolateral nucleus in many aquatic vertebrates with an electrosensory system; the marginal layer of the optic tectum in rayfinned fishes; electrosensory lobes in the few groups of advanced bony fish with an electrosensory system; the rostrolateral nucleus of the thalamus in a few widely scattered groups of bony fish; and the dorsal cochlear nucleus in all mammals except monotremes. All of these structures receive topographically organized sensory input in their deep layers. Purkinje-like cells receive the sensory input near their cell bodies. These cells extend apical dendrites up into the molecular layer where they receive synaptic input from parallel fibers. The cerebellum itself can be included within this characterization by considering the climbing fiber as at least in part a conveyor of sensory information and by recalling that climbing fibers in more basal vertebrates terminate on smooth dendrites close to the soma. Physiological findings from three different systems suggest the hypothesis that cerebellum- like structures remove predictable features from the sensory inflow. Phylogenetic homology can explain the similarities across different taxa for some types of cerebellum-like structures, but similarities within other types cannot be explained in this way. Moreover, phylogenetic homology cannot explain the similarities among different types of cerebellum-like structures. Evolutionary convergence provides the best explanation for all these similarities that cannot be explained by homology. The convergence is almost surely constrained by the availability of a genetic-developmental program for creating cerebellum-like circuitry and by the need within many different systems for the type of information processing that cerebellum-like circuitry can provide.
Adaptive processing of electrosensory information occurs in the cerebellum-like structures of three distinct groups of fish. Associations within each of these structures result in the generation of negative images of predictable features of the sensory inflow. Addition of these negative images to the actual inflow removes the predictable features, allowing the unpredictable, information-rich sensory signals to stand out. Evidence from all three groups of fish indicates that the negative images are mediated by plasticity at parallel fiber synapses.
Slides:
The Cerebellums of Mormyrid Fish and What They Might Tell us About Cerebellar Function.
Building a direct, artificial, connection between the brain and the world requires answers to the following questions:
1. What "signals" can we measure from the brain? From what regions? With what technology?
2. How is information represented (or encoded) in the brain?
3. What algorithms can we use to infer (or decode) the internal "state" of the brain?
4. How can we build practical interfaces that exploit the available technology?
This talk will summarize work at Brown on developing neural prostheses and will provide preliminary answers to the above questions with a focus on the problem of modeling and decoding motor cortical activity. Recent work has shown that linear models can be used to approximate the firing rates of a population of cells in primary motor cortex as a function of the position, velocity, and acceleration of the hand. I will describe a real-time Kalman filter for inferring (or decoding) hand motion from the firing rates of a population of cells recorded with a chronically implanted microelectrode array. will show recent results with direct neural control of smooth 2D cursor motion and will suggest future applications for brain machine interfaces and neural robot control.
References:
Connecting brains with machines: The neural control of 2D cursor movement, Black, M. J., Bienenstock, E., Donoghue, J. P., Serruya, M., Wu, W., Gao, Y., 1st International IEEE/EMBS Conference on Neural Engineering, pp. 580-583, Capri, Italy, March 20-22, 2003.
Slides:
The detection and tracking of people in video is challenging due the variability of human appearance, the high dimensionality of articulated body models, self occlusion, the loss of 3D information in the projection to 2D images, and the complexity of human motion. Recent progress on this problem is due in large part to the judicious use of machine learning methods. This talk will introduce the problem of human motion tracking and explore how machine learning techniques can be employed in a variety of sub-problems. In particular the talk will focus on methods for learning probabilistic models of how people appear and move over time.
To illustrate the ideas I will explore the use of a human body model in which the limbs are loosely connected. Conditional probabilities relating the 3D pose of connected limbs are learned from motion-captured training data. Similarly, we learn probabilistic models for the temporal evolution of each limb (backwards and forwards in time). Maximum entropy learning techniques are used to learn image likelihood models relating human limbs to a rich set of image measurements. Human pose and motion estimation is then formulated as a problem of inference in a graphical model and solved with non-parametric belief propagation using a variation of particle filtering that can be applied over a general loopy graph.
Joint work with:
Leonid Sigal (Brown University), Michael Isard (Microsoft Research), Stefan Roth (Brown University), Sidharth Bhatia (Brown University)
References:
Tracking loose-limbed people, Sigal, L., Bhatia, S., Roth, S., Black, M. J., Isard, M.,IEEE Conf. on Computer Vision and Pattern Recognition , vol. I, pp. 421-428, June 2004.
Slides:
See other recent Black's talksWhile much of the attention in neuroscience research has been directed to the electrophysiological investigations of neural dynamics, the interplay between neural shape and function has been relatively overlooked.
This talk presents how concepts of statistical mechanics and shape analysis have been applied in order to address this interesting paradigm.
More specifically, we show that the critical point in growing morphological networks provide a natural and direct measurement of the connectivity potential of the neuronal cells. We also investigate the recall potential of Hopfield neuronal networks whose connection matrix is determined by a Barabasi-Albert preferential attachment scheme and as a function of the morphology of the involved cells.
Slides:
A Complex Network Approach to Integrating Neural Shape and Function.
References:
Series P, Lorenceau J, Fregnac Y. The "silent" surround of V1 receptive fields: theory and experiments. J Physiol Paris. 2003 Jul-Nov;97(4-6):453-74. Review.
Monier C, Chavane F, Baudot P, Graham LJ, Fregnac Y. Orientation and direction selectivity of synaptic inputs in visual cortical neurons: a diversity of combinations produces spike tuning. Neuron. 2003 Feb 20;37(4):663-80.
Slides:
Coming soon.
References:
Gomez L, Budelli R, Grant K, Caputi AA. Pre-receptor profile of sensory images and primary afferent neuronal representation in the mormyrid electrosensory system. J Exp Biol. 2004 Jun;207(Pt 14):2443-53
Meek J, Grant K. The role of motor command feedback in electrosensory processing. Eur J Morphol. 1994 Aug;32(2-4):225-34. Review.
Slides:
Coming soon.
According to Helmholtz principle, human vision achieves the detection of geometrical structures in images as non-accidental coincidences. In other words, the grouping process required for the perception of gestalts (usual geometric configurations like alignement, parallelism, convexity, clusters, etc.) is based on a low-probability common arrangement of small image elements.
In the first part of this talk, we present a systematic formalization of Helmholtz principle, introduced a few years ago to detect partial gestalts (that is, structures arising from a single geometric property) in images. Such an approach is based on "a contrario" models, that derive decision rules to control the rate of false detections for a given model of noise. This "noise model" may be pure white noise, or may use some statistics of the considered image. We illustrate this point by considering several applications like the the detection of straight parts, contrasted edges, stereo correspondences, etc.
In the second part of the talk, we show that each detection of a partial gestalt using an a contrario model can be interpreted as a reduction of the image entropy. This link with information theory suggests a definition of a "visual information", including only perceptual descriptors of the image contents. Such an approach for gestalt cooperation leads to interesting issues in terms of computational complexity, and we present a simple approximation using a notion of short-range dependecy between random variables.
The first part of the talk is a joint work with A. Desolneux and J.-M. Morel. The second part is a joint work in progress with S. Vamech.
References:
A. Desolneux, L. Moisan, J.-M. Morel, "Meaningful Alignments", International Journal of Computer Vision, vol 40:1, pp. 7-23, 2000.
A. Desolneux, L. Moisan, J.-M. Morel, "Edge detection by Helmholtz Principle", Journal of Mathematical Imaging and Vision, vol 14:3, pp. 271-284, 2001.
A. Desolneux, L. Moisan, J.-M. Morel, "A grouping principle and four applications", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol 25:4, 2003.
Slides:
Finding structures in images with A Contrario model.
Inpainting is the art of modifying and image in a form that is not detectable to an ordinary observer. The applications of this are numerous, from special effects in movies to wireless image transmission. In this talk we will describe novel algorithms for image inpainting that we have been developing in the last few years. The algorithms are based on partial differential equation such as those used to model fluids. We will show how we can simultaneously reconstruct structured and textured regions, show extensions to 3D surface, and present numerous examples in special effects, image reconstruction (from family photos to the Venus mission), image compression and transmission, color reproduction, and 3D art reconstruction.
We will also talk about the connections of our algorithms with biological processes.
References:
Slides:
Image/surface inpainting and camouflage: Do not believe what you see.
We propose to model the functional geometry of the primary visual cortex as a principal fiber bundle where the 2-dimensional retinal plane is the base manifold and the secundary variables of orientation and scale constitute the vertical fibers over each point as a rotation-dilation group. The total space is equipped with the natural symplectic structure neurally implemented by long range horizontal connections.
This model naturally integrates several features of the visual cortex observed in experiments of neurophysiology, psychophysics and neuroimaging.
We start from the well known observation that the set of simple cell receptive profiles (RPs) is generated by the action of the affine group on a Gabor mother function.
Each hypercolumn of simple cells RPs, defined on a (x,y) retinal point, forms a 2-dimensional subgroup of rotation-dilation. This structure is identically repeated for every point of the retina, then can be considered as a fiber of the 4-dimensional principal bundle.
Then we show that the neural process of maxima selection due to intracortical short range connections selects the section of the fibration lifting boundaries to curves and whole figures to surfaces.
From the neurophysiological point of view there is evidence of connections between simple cells belonging to different hypercolumns. These neural connection are the so called long range horizontal connections, that are a physical implementation of the geometric horizontal connections in principal bundles, dedicated to parallel transport between fibers. The connection can be expressed either by the exponential map or in terms of the symplectic form, introducing a $C2$ quasi complex structure.
The integral curves of the natural contact structure is a mathematical representation of the association field of Field, Hayes and Hess, while the complex integral curves of the symplectic structure model the connectivity pattern between simple cells in the visual cortex V1, as observed by electrophysiological experiments. In other words the symplectic structure introduces a system of natural coordinates in 4-dimensional space $\C2$ that is implemented by neural connectivity.
Slides:
On the subriemannian geometry of the visual cortex.
We describe some of the successful modeling techniques that have been used in information-theoretic approaches to image processing tasks such as lossless image compression, denoising, and simulation. We discuss the relations between various commonly used modeling tools, show how they are used in practice, and explain their effectiveness in terms of a hierarchichal organization of "learning" tasks. We draw parallels to natural phenomena, including some of the modern theories of language learning, and speculate that our techniques are just primitive analogues of sophisticated mechanisms that Nature has built through the evolutionary process.
Slides:
Information-theoretic models in image processing.
It this talk, we give a formulation of a stochastic snake model based the theory of interacting particle systems and hydrodynamic limits. Curvature flows have been extensively considered from a deterministic point of view. They have been shown to be useful for a number of applications including crystal growth, flame propagation, and computer vision. We have described a random particle system, evolving on the discretized unit circle, whose profile converges toward the Gauss-Minkowsky transformation of solutions of curve shortening flows initiated by convex curves. The present note shows that this theory may be implemented as a new way of evolving curves as a possible alternative to level set methods. This will be used to derive a new stochastic snake model. We will illustrate the method on a wide variety of medical imagery.
Slides:
Optimal Transport, Conformal Mappings, and Stochastic Methods for Registration and Surface Warping.
November 10, 15:00 hs.
November 11, 15:00 hs.
"Representation of reality by brain and machines;
crossed views from neurosciences and computer vision."
home |
aims |
talks |
posters |
registered |
sponsors |
about uruguay