We are looking for a PhD candidate to join our research project entitled “BrainShorts: Identifying neural mechanisms of dynamic video perception with deep learning”, funded by a NWO Vidi grant awarded to Iris Groen.

The public summary of the project is as follows: Videos are everywhere these days. The popularity of platforms such as TikTok shows how much these ‘stimuli’ appeal to our brain. Physically, video shorts consist of a few seconds of pixels changing luminosity over time. Our visual brain effortlessly distils a meaningful message from this, while even the most advanced AI techniques still struggle. Moreover, modern AI requires a lot of training data and energy to analyse a short video, while the brain continuously operates in a dynamic world on a few watts only. What clever computations does the brain use to achieve efficient processing of videos?

This PhD project aims to address this question by developing stimulus-computable models of human electroencephalogram (EEG) responses to video shorts. Together with a research assistant and post-doctoral researcher, you will collect an unprecedented, large-scale EEG dataset of dynamic brain measurements of the video-watching brain, and use these data to:

a) curate an open-science, neural prediction benchmark for deep learning video-AI models;

b) assess representational alignment of bio-inspired deep learning models to the human brain.

The bio-inspired models will be enriched with different temporal integration mechanisms hypothesized to underlie efficient neural processing of dynamic inputs. By directly comparing models with and without brain-inspired computations, you will identify the contribution of each type of computation to effective human video perception.