Be part of an interdisciplinary team and contribute to the development of neural networks mapping premotor neural activity to vocalizations for interpretability studies. The goal is a publication about the method and its application to our own neural recordings.
neural recording, vocal learning, bioacoustics, songbird, neural network, interpretability, signal processing
In our group we aim to understand vocal development in songbirds, with interest in shared mechanisms between songbirds and humans. We have extensive large-scale recordings of neural activity in song-relevant brain areas of songbirds during vocal production. These recordings promise insight into the neural mechanisms of vocal learning, but analyzing and drawing conclusions is a lengthy process and requires new tools that can handle the high-dimensional nature of these recordings.
The goal of this master thesis is to explore the feasibility of mapping neural activity to vocalizations, and interpret the intermediate representations of the pre-trained network activations, comparing them with the current understanding in the field, and assessing the alignment between the two.
Your tasks will be literature review, implement machine learning algorithms (Liquid neural networks, S4, Liquid S4, Hyena or similar), assess the performance and visualize the contribution of the recorded brain structures and neurons to the prediction. We expect proper documentation of your code and data, and a useful report that could lead to a publication. We have weekly meetings to discuss outcomes, ideas, and next steps. The thesis workload is designed for 6-month full-time work.
You will learn about neuroanatomy, biophysics, neural activity recordings, machine learning, and neural network interpretability methods. You will gain expertise in state-of-the-art deep learning and data analysis methods and work in an interdisciplinary group in fields spanning neuroscience, machine learning, bioacoustics, and signal processing.
We are looking for a student interested in machine learning, neuroscience, signal processing and programming. To apply, please send a CV and transcript of records to one of the contacts below.
Maris Basha: maris (at) ini.ethz.ch
Corinna Lorenz: corinna (at) ini.ethz.ch
Remo Nitschke: remo.nitschke (at) uzh.ch
Yigit Demirag: yigit (at) ini.ethz.ch
Prof. Dr. Richard Hahnloser: rich (at) ini.ethz.ch
Schneider, S., Lee, J. H., & Mathis, M. W. (2023). Learnable latent embeddings for joint behavioural and neural analysis. Nature, 1-9. https://doi.org/10.1038/s41586-023-06031-6
Hasani, R.M., Lechner, M., Amini, A., Rus, D., & Grosu, R. (2020). Liquid Time-constant Networks. AAAI Conference on Artificial Intelligence. https://doi.org/10.1609/aaai.v35i9.16936