Are you sure where to go? Studying algorithm(s) for decision making and confidence in an ethological framework
We take decision with various degrees of confidence depending on context/sensory ambiguity, internal state (motivation, fear ...) and past experiences. In well-known/easy contexts, we quickly decide what the best action to execute is. In new or difficult situations, or following mistakes, we are cautious and take more time before acting. The understanding of the main constraints that influence the decision process and their neuronal underpinning is still fragmentary. The goal of this project is to combine behavioral science, engineering, and model-based data analysis to understand how rodents take decision in a whisker-guided choice task. The 1st objective will be to develop a decision making task in an ethologically valid environment allowing a 24x7 on-demand testing. In the 2nd objective, drift-diffusion modeling and reinforcement learning will be combined to characterize behavioral dynamics and generate prediction on decision making mechanisms.
Keywords: decision making, confidence, drift-diffusion model, whisker, cortex, basal ganglia, reinforcement learning theory
1) Design a modified Y maze based-task in which parameterized sensory cues (patterned texture) provide information about where animals should go when it reaches the maze’s intersection.
2) Include this task in an automatized 24h/7d on-demand environment to eliminate human intervention.
3) Analyze behavioral performance of animals using modeling approaches (drift-diffusion model and reinforcement learning)
Task: we will use a grating pattern that animals can detect with their whiskers when they run in the central stem of the Y-maze. The pattern direction (upward or downward) will provide an indication of which arm the animal must enter to obtain a reward. The ambiguity of this sensory signal will be controlled on a trial-by-trial basis by varying the angle of the pattern. We will use open-source software and hardware to control task events and collect data.
Data analysis: we will record animals’ positions through video tracking and photo-detectors to track decision and confidence (which correlates with reaction time and running speed). The data will be used to parameterize and optimize (using reinforcement learning) a drift-diffusion model of decision (Dunovan & Verstynen, bioRxiv, 2017) and provide mechanistic prediction and possible neural implementation in the cortical-striatal network.
PhD student’s expected profile
The 1st objective of the project is to develop an automatized task to study decision making in rodents. We are thus seeking for applicants with strong engineering and programming skills (required to design and build the behavioral apparatus, control of task events and data acquisition). The 2nd objective of the task is to perform model based-analysis (machine/reinforcement learning) of animal behavior. Applicants interested in animal behavior and neuroscience (decision making, neuroeconomics) with a taste for engineering and data analysis should also apply. Applicants with background in computer science are also encouraged to apply.