Reinforcement learning of human vocal behavior
We study reinforcement learning of fundamental frequency (pitch) in songbirds and humans. When birds receive aversive reinforcement for low-pitch syllables they successfully learn to increase the syllables’ pitch.
We are working on theoretical mechanistic models of reinforcement learning to unravel the effect of reinforced vocalizations on future vocalizations via the correlation between exploration and reward. From behavioral data and by using the expectation maximization algorithm, we are able to estimate crucial learning-related parameters such as the fraction of behavioral variance accessible for learning.
We offer semester or MSc projects that aim to quantify reinforcement learning of vocal pitch in humans, in order to test for similarities of vocal learning between humans and songbirds. The goal is to design a mechanistic model on the human data and analyze the similarities and differences between the learning trajectories of birds and humans. Furthermore, we aim to estimate the influence of reinforced versus non-reinforced trials on future trials, to disentangle their possibly distinct roles in learning.
Anja Zai, zaia (at) ini.ethz.ch