Statement of Purpose

Timmer's Statement of Purpose - October 1994

A critical behavior in all creatures is the orientation response to novel or ``interesting'' stimuli. In the struggle to find food and to avoid predators, this behavior must be fast and reliable. In most animals the neural pathway from sensor to muscle is short and typically initiates an orienting response to improve sensory conditions. This orientation behavior is usually quick (e.g. saccades) and often requires specialized neural controllers which do not use proprioceptive or visual feedback during the movement. Such systems use past performance to calibrate the movement to maintain accuracy throughout the lifetime of the organism. In vertebrates, visual information from the retina is known to project directly to a brainstem structure called the superior colliculus (or optic tectum) which is involved in orientation movements. Activation of the superior colliculus by indirect visual stimulus or direct electrical stimulation generates head movements in owls, head and eye movements in cats, and eye movements in primates. In addition to visual input, acoustic localization information and somatosensory information also converge in the superior colliculus. The superior colliculus is an important convergence point for multimodal spatial information involved in orientation behaviors. In all these behaviors, learning is incorporated to achieve rapid, accurate, and stable eye movements and to keep the various sensory modalities aligned to each other. While it is not known specifically where the sites of learning are, models which incorporate data from both adaptation experiments and neuroanatomical studies are being investigated.

My research interest lies in understanding the neurobiological mechanisms of orientation responses and how these systems learn to improve their performance over time. I am building a saccadic eye movement system with analog VLSI implementations of neural models proposed for this sensorimotor system. In particular, I am developing a hardware model of the superior colliculus and brainstem burst generator system which will participate in the generation of saccades and will be capable of continually improving its own performance.

The investigation of neural information processing architectures in analog VLSI can provide insight into the issues that real biological nervous systems face. Analog VLSI architectures share many of the advantageous properties with neural systems such as speed, space efficiency, and low power consumption. In addition, analog VLSI must face similar constraints such as real-world noise, component variability or failure, and interconnection limits. Not only can adaptive neural systems which self-calibrate overcome many of these drawbacks, they will be useful in the exploration of architectures which control non-linear devices.