Manish Saggar

In this project we developed a computational model of how motivation influences learning, elaborating on the empirical study of Markman, Baldwin and Maddox (2005). In a decision criterion learning task with unequal payoffs, the subjects were more likely to maximize the reward when their motivation was in line with the reward structure (i.e. when they were in a regulatory fit), whereas they were more likely to maximize accuracy when their motivation did not match the reward structure (i.e. when they were in a regulatory mismatch). The model accurately replicates this pattern of results, and also accounts for the individual subject's behavior. In addition, the model makes the novel prediction that regulatory-fit subjects who are near the reward threshold will shift their strategy toward maximizing accuracy, whereas regulatory-mismatch subjects who are far from the reward threshold will shift their strategy toward maximizing reward. When the original data was reanalyzed, this model-driven prediction was confirmed. These results constitute a first computational step towards understanding how motivation influences learning and cognition.

Being able to detect collisions and recover quickly during a 4-legged robot soccer game is crucial in order to improve the fluency of the games and avoid penalties due to pushing other robots. Previous work mostly used the difference between actual motion and intended motion as an indicator of collision, and only specific walking directions and speeds were considered during training. Accelerometer readings combined with the motion command are good indicators of regularities and novel situations, which in that case are collisions. This paper proposes a neural networks-based solution to the problem of detecting collisions during omnidirectional motion of Sony Aibo ERS-7 with the motivation of considering it as an instance of general novelty detection problem.

A single biological neuron is able to perform complex computations that are highly nonlinear in nature, adaptive, and superior to the perceptron model. A neuron is essentially a nonlinear dynamical system. Its state depends on the interactions among its previous states, its intrinsic properties, and the synaptic input it receives. These factors are included in Hodgkin-Huxley (HH) model, which describes the ionic mechanisms involved in the generation of an action potential. This paper proposes training of an artificial neural network to identify and model the physiological properties of a biological neuron, and mimic its input-output mapping. An HH simulator was implemented to generate the training data. The proposed model was able to mimic and predict the dynamic behavior of the HH simulator under novel stimulation conditions; hence, it can be used to extract the dynamics (in vivo or in vitro) of a neuron without any prior knowledge of its physiology. Such a model can in turn be used as a tool for controlling a neuron in order to study its dynamics for further analysis.