Next Monday Discussion : MacOSX and the first lab : Survival Kit
It can be demonstrated on system theoretical grounds that any system which learns to perform many behavioural features with limited information handling resources is constrained within a set of bounds called the recommendation architecture by the requirement to find a compromise between the need to conserve physical information handling resources and the need to learn without severe interference with earlier learning.
Overall architecture, the definition of modules and components, and even device algorithms are all constrained, with the severity of the constraints increasing as the ratio of features to resources increases. Algorithms widely used in artificial neural networks cannot be used in some major subsystems of the recommendation architecture.
There are strong resemblances between the physical forms of a system within the recommendation architecture bounds and the physiology of the human brain including separations between and functions of the cortex, hippocampus, thalamus, basal ganglia, cerebellum, hypothalamus and amygdala; the internal organization of the cortex into layers, columns and areas; and the topology and synaptic algorithms of neurons. Detailed psychological observations of a wide range of cognitive phenomena, including semantic, episodic, working and procedural memory; processes such as arithmetic; the deficits resulting from physical damage; and sleep with dreaming can be modelled in a physiologically plausible manner by a system within the recommendation architecture bounds.
Learning can be bootstrapped from experience with minimal and plausible a priori information. Many phenomena labelled "conscious" can be modelled in terms of physiology. Electronic systems within the recommendation architecture bounds confirm the capabilities of the architecture and point the way to implementation of systems with human like cognitive capabilities.
Andrew was employed by Nortel Networks as a system designer and architect in the design of extremely complex real time control telecommunications systems from 1969 to 1999, and participated in successful projects to design and introduce commercially successful state of the art systems with up to 20 million lines of code and custom integrated circuit based hardware.
While still employed as a system architect, he wrote a book on understanding the brain as a system, introducing a novel cognitive architecture. He subsequently obtained a US patent for system architectures which can learn to manage a complex telecommunications network based on the cognitive architecture. Since 1999 he has been full time in academic research into cognitive systems, most recently as a research fellow at the Australian National University.
His new book, "A Systems Architecture Approach to the Brain: from Neurons to Consciousness", was published December 12th 2005 and will be introduced at this seminar.