Home
FED3Bandit is an implementation of a 2-armed bandit task that uses a FED3 device. FED3Bandit uses FED3, which means it is customizable, can be used in teh home-cage, and requires minimal supervision (food and battery level checks every few days). In this page we first provide a brief introduction of a 2-armed bandit task and we then describe the features that FED3Bandit has and the ways in which it can be customized.
A two-armed bandit task is a special case of the k-armed bandit problem. It consists of an environment where there are two choices that are each associated to a probability of receiving a reward. Let these two choices be “Left” and “Right”. The reward probabilities of Left and Right may be fixed or come from a probability distribution, remain constant or change upon a condition, and may be dependent or independent from each other. When the reward probabilities of Left and Right change upon meeting a condition, the task is called nonstationary. A special case of nonstationary bandit task is The restless two-armed bandit task, where reward probabilities change slightly after each choice.
Examples of conditions that may be used to trigger the change in reward probabilities:
- After n rewards deliverd
- After m consecutive choices in the higher probability port
- After every choice (restless bandit task)
The two-armed bandit task has become increasingly popular in neuroscience as it can probe learning behavior, particularly learning by trial-and-error. The task is also suitable for the the use of reinforcement learning techniques and models.
FED3Bandit is an implementation of the two-armed bandit task that that uses the FED3 device (more information about FED3 can be found here: About FED3). In this section we will describe the FED3 features as they pertain to FED3Bandit. FED3 has two nose pokes, which are used in FED3Bandit as the two “arms”. Additionally, FED3 has a well through which food pellets are dispensed. FED3Bandit, thus, uses nose pokes as choices, and food pellets as rewards.
