Modeling Lateral Inhibition
Authors
Dhillon, Navinder Singh ; El-Omar, Barak
Term
4. term
Publication year
2018
Pages
81
Abstract
Lateral inhibition sharpens sensory perception and supports localization; prior studies report that localizing noxious stimuli is more difficult than innocuous ones. This thesis investigates these discriminatory differences by developing a simple artificial neural network model of lateral inhibition for laser and mechanical stimulation of the skin. A single-layer feed-forward network was implemented in MATLAB R2017b and trained with gradient descent on two-point discrimination data from Frahm et al. (2017). Validation indicated that the laser-stimulation model fitted the training data closely (prediction error 0.0102 mm), whereas the mechanical-stimulation model showed a higher training error (0.0618 mm). Across modalities, the models exhibited limited ability to generalize beyond the training data, with particularly poor generalization for mechanical stimulation. The work demonstrates the feasibility of ANN-based modeling of lateral inhibition and highlights modality-related differences that merit further study.
Sidehæmning skærper sanseindtryk og understøtter lokalisering; tidligere studier viser, at lokalisering af noxiske stimuli er vanskeligere end af uskadelige stimuli. Dette speciale undersøger disse forskelle ved at udvikle en simpel kunstig neuralt netværksmodel af sidehæmning for laser- og mekanisk stimulation af huden. Et enkeltlaget feed-forward netværk blev implementeret i MATLAB R2017b og trænet med gradientnedstigning på to-punkts diskriminationsdata fra Frahm et al. (2017). Validering viste, at modellen for laserstimulation passede træningsdata tæt (forudsigelsesfejl 0,0102 mm), mens modellen for mekanisk stimulation havde en højere træningsfejl (0,0618 mm). På tværs af modaliteterne var evnen til at generalisere ud over træningsdata begrænset, med særligt dårlig generalisering for mekanisk stimulation. Arbejdet demonstrerer anvendeligheden af ANN-baseret modellering af sidehæmning og peger på modalitetsrelaterede forskelle, der bør undersøges nærmere.
[This apstract has been generated with the help of AI directly from the project full text]