Twenty years ago, R. Brooks revealed to the A.I. community that elephants don't play chess [1] and ten years later, A. Clark explained in [2] that "we ignore the fact that the biological mind is, first and foremost, an organ for controlling the biological body. Minds make motions, and they must make them fast ? before the predator catches you, or before your prey gets away from you. Minds are not disembodied logical reasoning devices." This lecture proposes to look back at (almost) 60 years of Artificial Intelligence researches in order to address the question of what has been accomplished so far towards our understanding of intelligence and cognition. In this context, we'll introduce the action-perception loop, the embodied cognition paradigm and the symbol grounding problem as it has been identified by S. Harnad [3]. This problem has became prominent in the cognitive science society and the idea that a symbol is much more than a mere meaningless token that can be processed through some algorithm sheds a new light on higher brain functions [4]. More specifically, we'll explain how those theories can impact modeling on computer vision.
[1] Brooks, R., 1990. Elephants don't play chess. Robotics and Autonomous Systems 6, 3-15. [2] Clark, A., 1998. Being There: Putting Brain, Body, and World Together Again. MIT Press. [3] Harnad, S. 1990, The Symbol Grounding Problem. Physica D 42, 335?346. [4] Rougier,N. and Noelle, D. and Cohen, J. and Braver, T. and O'Reilly, R., 2005. Prefrontal Cortex and the Flexibility of Cognitive Control: Rules Without Symbols. Proceedings of the National Academy of Science, 102, pp. 7338-7343.
"On ne voit que ce qu'on regarde" Maurice Merleau Ponty, ("We only see what we're watching")
This lecture proposes to review current psychological and physiological data as and classical experiments related to visual attention as well as anatomical and physiological data related to the oculomotor control in the primate. We will introduce the two main forms of visual attention, namely exogeneous (bottom up) and endogeneous (top down) visual attention that are known to play a critical role in the perception and processing of a visual scene. Facilitating and inhibitory effects of visual attention will be presented in light of Posner experiments (1980) related to the concept of inhibition of return that play a major role in a number of computational models of visual attention. Finally, integrative theories related to visual attention will be introduced, namely the premotor theory of attention [1], the active perception paradigm [2] and the deictic codes for the embodiment of cognition [3].
[1] Rizzolatti, G., Riggio, L., Dascola, I. and Umilta C., 1987. Reorienting attention across the horizontal and vertical meridians: evidence in favor of a premotor theory of attention. Neuropsychologia 25: 31-40. [2] O'Regan, K. and Noe, A., 2001. A sensorimotor account of vision and visual consciousness, Behavioral and Brain Sciences 24, 939-10031. [3] Ballard, D., Hayhoe, M., Pook K., Rao, R., 1997. Deictic codes for the embodiment of cognition.
This lecture introduces main concepts related to classical artificial neural networks as well as computational neuroscience. Standard artificial neural network models related to supervised, unsupervised and reinforcment learning will be briefly introduced as well as key concepts from neuro-anatomy and neuro-physiology. This lecture will also focus on the dynamic neural field (DNF) Theory as it has been originally introduced by Wilson and Cowan [1] in the early seventies and later formalized by S.I. Amari [2] and J.G. Taylor [3]. These theories explain the dynamic of pattern formation for lateral-inhibition type homogeneous neural fields with general connections. They show that, in some conditions, continuous attractor neural networks are able to maintain a localised bubble of activity in direct relation with the excitation provided by a stimulation. We will investigate further these theories in order to explain how their functional properties can be linked to visual attention defined as the capacity to attend to one stimulus in spite of noise, distractors or saliency effects [4].
[1] Wilson, H., Cowan, J., 1972. Excitatory and inhibitory interactions in localized populations of model neurons. Biophysical Journal 12, 1-24. [2] Amari, S., 1977. Dynamic of pattern formation in lateral-inhibition type neuralfields. Biological Cybernetics 27, 77-88. [3] Taylor, J., 1999. Neural bubble dynamics in two dimensions: foundations. Biological Cybernetics 80, 5167-5174. [4] Rougier, N., Vitay, J. 2006, Emergence of Attention within a Neural Population, Neural Networks 19, 573-581.
The visual exploration of a scene involves the interplay of several competing processes (for example to select the next saccade or to keep fixation) and the integration of bottom-up (e.g. contrast) and top-down information (the target of a visual search task). Identifying the neural mechanisms involved in these processes and the integration of these information remain a challenging question. Visual attention refers to all these processes, both when the eyes remain fixed (covert attention) and when they are moving (overt attention). Popular computation models of visual attention consider that the visual information remains fixed when attention is deployed [1] while the primate are executing around three saccadic eye movements per second, abruplty changing the whole visual information. We'll introduce in this lecture a model relying on dynamic neural fields [2] and show that covert and overt attention can emerge from such a substratum. We'll identify and propose a possible interaction of four elementary mechanisms for selecting the next locus of attention, memorizing the previously attended locations, anticipating the consequences of eye movements and integrating bottom-up and top-down information in order to perform a visual search task with saccadic eye movements.
[1] Itti, L., Koch, C., 2001. Computational modeling of visual attention, Nature Review Neuroscience 2 194-203. [2] Fix, J., Rougier, N., Alexandre, F., 2007. From physiological principles to computational models of the cortex, Journal of Physiology 101 32-39.
