Navigation auf uzh.ch

Suche

Institute of Neuroinformatics

Statistical inference of social learning strategies

The acquisition of a song template - A special case of learning and memory

template
How do juvenile birds memorize a template (TEM=T1,…,T5) of adult song that they later use as learning target of their immature song (S1,…,S5)? We are working towards mathematically describing this template formation process.

Juvenile songbirds are capable of learning a courtship song by imitation of an adult tutor. In 1965, Marc Konishi suggested that young songbirds imitate an adult tutor by comparing the auditory feedback of self-produced sounds to a memorized auditory template (1).  This hypothesized template is some kind of memory trace of the tutor song’s auditory features. Despite decades of research, the exact feature extraction mechanism involved in template formation remains elusive. While there is evidence that reinforcement learning (RL) governs error-driven vocal adaptation during song learning, little is known about the feature extraction mechanism involved in template formation.

In collaboration with Dina Lipkind at Cuny, we perform controlled experiments in which we provide diverse song targets to young birds. From their behavioral learning trajectory, we try to infer the template formation mechanism. Our assumption is that the template is defined as the song the bird ends up singing as an adult.

Concretely, we want to find out whether the template TEM is an instance memory or a composite memory. In the context of Figure 1, where the tutor sings a song syllable i with diverse variants Ti* and Ti#, an instance memory means that the song target TEMi of that syllables satisfies either TEMi=Ti* or TEMi=Ti#; a composite memory means that the syllable target is a more complex function F of the syllable variants:  TEMi=F(Ti*,Ti#). A characterization of the unknown function F will advance our understanding of how the developing brain extracts motor targets from sensory inputs. Such insights will likely benefit the treatment of developmental speech disorders.           

Student Project

If you are interested in this project for an MSc Thesis or Semester Project, please get in touch with Anja Zai

Prerequisites: Knowledge in programming (Matlab, Python), data-science methods (many TB of data), neural networks, reinforcement learning, probability and estimation theory, and in particular Bayesian inference. An interest in Biology, Neuroscience, developmental psychology, language science, or ethology is a big plus.

Reference

  1. Konishi M. The role of auditory feedback in the control of vocalization in the white-crowned sparrow. Z Tierpsychol. 1965 Dec;22(7):770–783.