2026-09-20T04:37:29Zhttps://api.repository.cam.ac.uk/server/oai/request

oai:www.repository.cam.ac.uk:1810/3922042025-11-12T01:42:10Zcom_1810_221811com_1810_256062col_1810_221812

Models of sensory coding https://doi.org/10.17863/CAM.122988 Foldiak, Peter This work is aimed at understanding and modelling some of the general computational principles of sensory information processing in the brain. The sensory system detects physical quantities of the environment and transforms them into internal representations on which behavioural decisions are based. The properties that make a representation useful are discussed, and ways in which the sensory system may form such representations from a complex array of receptor signals are considered. The brain needs extensive knowledge about the statistical structure of the sensory environment for the interpretation of sensory signals. The acquisition and use of such knowledge are studied using models consisting of networks of simple processing units with properties that are believed to be functionally essential in biological neurons. As the information processing capacity of these networks is due to the adaptive, modifiable connections between the units, the rules governing the activity-dependent modification of these connections are studied. One class of such ‘learning rules’, local learning rules, are particularly important for understanding the nervous system. Specific hypotheses about the form of these rules are studied in four ‘unsupervised’ learning tasks, in which the goal is not to implement a mapping between a predefined set of given input and output patterns, but to discover statistical structure in the input distribution without external guidance or supervision: 1 - An ‘anti-Hebbian’ synaptic modification rule is demonstrated to be able to adaptively form an uncorrelated representation of the correlated input signal. This mechanism can match the distribution of input patterns to the actual signalling space of the representation units, achieving information-theoretically optimal signal on noisy units. An uncorrelated, equal variance signal also makes optimally efficient least-mean- square error correcting learning possible. 2 - A combination of Hebbian and anti-Hebbian connections is demonstrated to implement a form of the statistical method of Principal Component Analysis, which reduces the dimensionality of a noisy Gaussian signal while maximising the information content of the representation, even when the units themselves are noisy. 3 - A similar arrangement of biologically more plausible, nonlinear units is shown to be able to adaptively code inputs into a sparse representation, substantially reducing the higher-order statistical redundancy of the representation without considerable loss of information. Such a representation is advantageous if it is to be used in further associative learning stages. 4 - A Hebbian rule modified by a trace mechanism is studied, that allows processing units to respond in a way which is invariant with respect to commonly occurring transformations of the input signal. University of Cambridge 1991-10-01 Thesis Doctoral Doctor of Philosophy (PhD) https://www.repository.cam.ac.uk/handle/1810/392204 https://www.repository.cam.ac.uk/bitstreams/63cadd90-0531-442a-9287-657e2affc0c5/download 089d2b1df0e2dd75e308924e51950889 https://www.repository.cam.ac.uk/bitstreams/edadc857-5d20-4cbc-84a6-dac88dd0ef1e/download eb9d2d75cbfff5683a827dd2229be5c0 http://purl.org/NET/rdflicense/allrightsreserved/