Extraction of visual information using maximum likelihood Hebbian learning

We explore an extension of Hebbian learning which has been called ε-insensitive Hebbian learning, and derive lateral connections from a probability density function (PDF). We use these lateral connections to move outputs towards the mode of the PDF and use the resulting outputs to train the feedforward connections. We show that ε-insensitive Hebbian learning may be considered as a special case of maximum likelihood Hebbian learning and investigate the resulting network with both real and artificial data. We finally show that the resulting network is able to identify motion in the environment.