Tranformers as predictive models of the retina

Design a transformer network for the retina


We train neural networks to predict responses of retinal ganglion cells and then further use them to gain insights into the retinal circuitry. So far, we mostly build convolutional neural networks that do not reflect the transformer revolution that happened in the field of deep learning.


In this project you could design and build a transformer that is can predict retinal ganglion cell responses and benchmark it against our state-of-the-art models.


Within the context of this project, several research questions could be explored: - Is a transformer architecture able to outperform CNNs in predicting neuronal activity of the retina? - How does the attention mechanism change the predictions compared to convolutions?


To apply please email Michaela Vystrčilová stating your interest in this research and detailing your relevant skills.

Neural Data Science Group
Institute of Computer Science
University of Goettingen