Metaheuristic Optimisation Algorithms for Tuning a Bioinspired Retinal Model.

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Engineering

Background:

• Understanding visual information encoding in the retina is crucial for developing advanced bioinspired sensors and artificial retinal systems.
• Existing computational models require refinement to accurately mimic vertebrate retinal behavior.

Purpose of the Study:

• To tune a computational bioinspired retinal model for improved mimicry of retinal ganglion cell responses.
• To identify the optimal metaheuristic algorithm for fine-tuning the retinal model.

Main Methods:

• An automatic multi-objective optimization strategy using four biological-based metrics was developed.
• Population-based search heuristics including genetic algorithms (SPEA2, NSGA-II, NSGA-III), Particle Swarm Optimization (PSO), and Differential Evolution (DE) were explored.
• Performance comparison was based on the hypervolume metric, with nonparametric statistical tests for rigorous evaluation.

Main Results:

• The Particle Swarm Optimization (PSO) algorithm achieved the best performance.
• PSO demonstrated superior results based on the largest hypervolume, well-distributed Pareto front elements, and high numbers on the Pareto front.
• Rigorous statistical tests confirmed PSO's effectiveness over other tested metaheuristics.

Conclusions:

• The proposed multi-objective optimization strategy effectively tunes bioinspired retinal models.
• Particle Swarm Optimization (PSO) is the most effective algorithm for enhancing the accuracy of artificial retinal systems.
• This research advances the development of sophisticated bioinspired sensors and artificial retinas.