R.I.S.E. Optimizing Stimulus Parameters for Complex and Simple Behaviors in Drosophila melanogaster Using Model-Free Regret-Informed-Stimulus-Engineering Algorithm

Abstract

Choosing the optimal stimulus is an essential aspect of conducting behavioral research. In this paper, we discuss the advantages and drawbacks of different stimulus optimization methods. We propose the Regret-Informed-Stimulus-Engineering (RISE) algorithm as a potential alternative for stimulus optimization. RISE is a model-free algorithm that uses an estimated gradient to update stimuli in response to collected behavioral data of the past stimulus presented. RISE was tested on simple and complex behaviors in Drosophila melanogaster, starting with simple optomotor turning. RISE was successful in driving clockwise and counterclockwise behavior in Drosophila melanogaster. RISE was adapted with a Fourier series to optimize more complex stimulus sequences and tested with the goal of inducing variation in Drosophila melanogaster walking. Experiments conducted showed no significant difference of variation in walking compared to a control group. Lastly, optogenetic experiments were conducted to test RISE’s ability of stimulus optimization on internal states. Courtship tracking behavior was successfully induced, but it was inconclusive if RISE successfully optimized a visual stimulus to produce a more robust behavior. Despite time limitations in testing, the RISE algorithm shows promise in enhancing how researchers choose stimuli. RISE has the potential of automating the stimulus optimization process without relying on assumptions or prior knowledge in a time efficient manner.