direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Models of Neural Systems

PhD Theses

Neural Mechanisms and Computational Principles of Adaptive Sensory Processing
Citation key Wimmer20090
Author Klaus Wimmer
Year 2009
School Technische Universität Berlin
Abstract Neural systems compute dynamic representations of the environment: sensory neurons respond selectively to sensory stimuli, and adapt their ``code'' according to recently received sensory input. Despite the ubiquity of adaptation throughout neural processing across different species and sensory modalities, the functional benefits of adaptation often remain unclear. Ultimately, a complete understanding of the underlying computational principles requires insight into the neural circuitry and the physiological mechanisms that carry out these computations. Here, we study the generation and adaptation of sensory representations in low-level sensory processing in two model systems of different complexity: the auditory pathway of crickets and the primary visual cortex (V1) of cats. We first investigate the generation and dynamics of orientation selectivity in V1. Physiological and anatomical data indicate that the response properties of individual neurons can only be fully understood in the context of their local circuitry. Specifically, orientation tuned responses of a V1 neuron and its temporal response characteristics depend on the neuron's position in the orientation preference map and the recurrent inputs related to that position. We systematically explore a whole class of network models that incorporate the structure of the orientation preference map and conclude that only a network operating in a regime where excitatory and inhibitory recurrent inputs are balanced and dominate the feed-forward input is consistent with the experimental data. Furthermore, we argue that one functional benefit of this ``balanced recurrent'' operating regime is its enhanced sensitivity to modulations of the balance between excitation and inhibition. We then focus on intracortical synaptic depression as a potential mechanism through which adaptation alters recurrent processing and thus V1's representation of sensory stimuli. We simulate orientation adaptation in network models with different strengths of synaptic depression and select those models that predict adaptation-induced changes in orientation tuning and perceptual read-out consistent with experimental data. The relative strength of synaptic depression of the selected models is consistent with in vitro data, making intracortical synaptic depression a plausible mechanism of orientation adaptation. The best fit of the experimental data is found for the balanced recurrent network, which also correctly predicts an enhanced capacity for adaptive changes close to pinwheel centers. Finally, we study the optimality of adaptive systems in the simpler auditory pathway of the cricket. Unlike in primary visual cortex, where information is encoded by populations of neurons, just a single neuron conveys information about high frequency sounds to the cricket's brain. This enables us to accurately quantify the mutual information between a sensory stimulus and its neural representation. We design an experiment that allows us to distinguish whether this representation follows an infomax principle (a form of optimal coding) or a selective coding principle (a form of figure-background separation). We find that adaptation leads to a reduction rather than to an overall enhancement in information transmission, inconsistent with the infomax principle. However, adaptation also selectively decreases the amount of information that is transmitted about background signals, facilitating the detection of behaviorally relevant signals.
Link to publication Download Bibtex entry

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions