Best poster prize at ‘Subcortical sensory circuits’ – Course and Conference Office


Our mission is to train scientists. This blog is a platform for us to share updates on our annual programme, tips and tricks for scientists, new e-learning opportunities, and sometimes just something to make you smile.

Best poster prize at ‘Subcortical sensory circuits’

Last November we welcomed leading researchers in the field to discuss different subcortical circuits involving various sensory modalities and revisit their increasingly complex and interconnected contributions to perception, behavior, and cognition. The new EMBO Workshop ‘Subcortical sensory circuits’ featured two poster sessions from which two presenters received a poster prize for their outstanding work. We are pleased to introduce you to one of the winners today: Sacha Sokoloski!

A computational approach to visual ecology with deep reinforcement learning

Presenter: Sacha Sokoloski

Sacha Sokoloski, Hertie Institute for AI in Brain Health, University of Tübingen, Germany

Animal vision is thought to optimize various objectives from metabolic efficiency to discrimination performance, yet its ultimate objective is to facilitate the survival of the animal within its ecological niche. However, modeling animal behavior in a complex environment has been challenging. To study how environments shape and constrains visual processing, we have developed a deep reinforcement learning framework in which an agent moves through a 3-d environment that it perceives through a vision model, where its only goal is to survive. We consider a foraging task where the agent must gather food that sustains it, and avoid food that harms it. We first establish that the complexity of the vision model required for survival scales with the variety and visual complexity of the food in the environment. We then show how different model architectures learn distinct representations of the environment and task, and lead the agent to exhibit distinct behavioural strategies. Finally, we show that the vision model only encodes features that are relevant for agent survival and ignores task-irrelevant distractors, in conflict with existing efficient coding theories. In summary, we lay the foundation for a computational approach to visual ecology, provides extensive benchmarks for future work, and demonstrates how representations, behaviour, and relevance emerge from an agent’s drive for survival.

View poster

The EMBO Workshop ‘Subcortical sensory circuits: visual, auditory, somatosensory, and beyond‘ took place at EMBL Heidelberg and virtually from 28 November – 1 December 2023.