Forward Thinking, MOSAIC model and Mirror Neurons

MOSAIC model has been shown to be effective as a adaptive control system for non-linear and non-stationary control tasks (e.g. Haruno et al. 2001; Wolpert and Kawato 1998). Central to MOSAIC is the notion of multiple forward-inverse pair of models working together to accomplish the control task. The model selection is based on forward models prediction performance. The inverse corresponding to good predictors are engaged in the control task. This makes MOSAIC very effective in non-stationary control problems where most learning systems fail.

The Mirror Neurons (Dipellegrino et al. 1992; Gallese et al. 1996; Rizzolatti et al. 1996) are located in monkey area F5 (part of ventral premotor cortex) and become activate when the monkey observes meaningful actions performed by a demonstrator human or another monkey. The mirror neurons look like the usual motor-related F5 neurons with respect to their motor properties. A mirror neuron that responds for a particular action becomes active also when the monkey observes the demonstrator do the same action (Rizzolatti et al.1996; Gallese et al. 1996). The actions studied so far include grasping, breaking peanuts, placing and tearing (paper). The mirror neurons construct a link between self-actions and observed actions by representing the observed action in terms of internal motor encoding. Therefore, it has been suggested that the mirror neurons can be the basis for higher-level cognitive skills such as theory of mind and language in the evolution scale (Rizzolatti and Arbib 1998; Rizzolatti et al. 2001; Fadiga et al. 2000; Rizzolatti et al. 2000). Recently a computational model of Mirror Neurons has been developed based on the assumption that during infancy the visual stimuli produced by self generated grasping actions adapt the ventral premotor cortex (area F5) neurons that receive projections area 7b into Mirror Neurons (Oztop and Arbib 2002).

The current project is aimed at exploring possible brain mechanism for forward thinking and theory of mind and their relation to MOSAIC architecture. For advancing our understanding about forward thinking and MOSAIC, we work on multiple projects as follows.

1)	Extension of MOSAIC: Incremental MOSAIC and theoretical studies
2)	Exploring the relation between Mirror Neurons and MOSAIC: Reach and Grasp trajectory generation with MOSAIC => emergence of Mirror Neurons?
3)	Neuroscience studies: Localizing brain regions involved in theory of mind and forward thinking. Our initial attempt is to see how well MOSAIC model can explain F5 mirror neuron activity (2)
4)	MOSAIC in highly complex control task: Implementing MOSAIC learning and control in a humanoid robot

Currently the projects (1) and (2) constitute our main study. The incremental MOSAIC (item 1) is essential for high dimensional control problems, as the number of model required in advance cannot be determined.The theoretical studies are necessary to guarantee safe and successful operation in a real robot (as devising learning mechanisms that can avoid local minima). One may ask the questions, why do you need to implement MOSAIC on a robot; aren?t you interested in the real brain. The answer to this question is a functional one. We would like to also see MOSAIC being able to control a physical mechanism that is close to a human motor system so that our claim that MOSAIC can be a model for human motor control is functionally justified.

The ventral premotor cortex model of MOSAIC (item 2), instead is aimed at testing whether MOSAIC structure can be a model of monkey reach and grasp circuit leading to mirror neurons in a natural way. Note that, now we are one step higher in the cognition scale. Here MOSAIC structure is used as a reach and grasp (kinematics) trajectory generator, which exhibits (without our explicit intension of) action recognition functionality that may be the basis of forward thinking and theory of mind. The results of this project will lead us into (3) where we will try to climb up the cognition scale with new biologically realistic models of theory of mind.

Related Reading

Dipellegrino G, Fadiga L, Fogassi L, Gallese V, Rizzolatti G (1992) Understanding motor events - a neurophysiological study. Exp Brain Res 91: 176-180

Fadiga L, Fogassi L, Gallese V, Rizzolatti G (2000) Visuomotor neurons: Ambiguity of the discharge or 'motor' perception? Int J Psychophysiol 35: 165-177

Gallese V, Fadiga L, Fogassi L, Rizzolatti G (1996) Action recognition in the premotor cortex. Brain 119: 593-60

Haruno M, Wolpert DM, Kawato M (2001) Mosaic model for sensorimotor learning and control. Neural Comput 13: 2201-222

Oztop E, Arbib MA (2002) Schema design and implementation of the grasp-related mirror neuron system. Biological Cybernetics 87: 116-140

Rizzolatti G, Arbib MA (1998) Language within our grasp. Trends Neurosci 21: 188-194

Rizzolatti G, Fadiga L, Gallese V, Fogassi L (1996) Premotor cortex and the recognition of motor actions. Cognitive Brain Res 3: 131-141

Rizzolatti G, Fogassi L, Gallese V (2000) Mirror neurons: Intentionality detectors? Int J Psychol 35: 205-205

Rizzolatti G, Fogassi L, Gallese V (2001) Neurophysiological mechanisms underlying the understanding and imitation of action. Nat Rev Neurosci 2: 661-670

Wolpert DM, Kawato M (1998) Multiple paired forward and inverse models for motor control. Neural Networks 11: 1317-1329