Format and topics

Probabilistic approaches to learning, perceiving, and acting are at the core of highly influential theoretical neuroscience proposals, including the “probabilistic brain”, the “predictive brain”, and the brain as an optimiser of “cost functions”.

Making connections in this rapidly developing field is challenging since researchers are using different methods, studying the brain at different levels of analysis, and do not attend the same meetings. The aim of this workshop is to facilitate dialogue among researchers at the frontier of this new field, to identify gaps in our current models and approaches, and to shape the future agenda for research.

We have organised talks along two broad themes: Learning and development, and Representing and computing with uncertainty. We have identified key open questions in these areas. We invite the speakers to address these in their talks, and the discussants to focus on them in interactive discussions after each session.

Mechanisms for probabilistic learning and development

• What limits the abilities of current models of statistical learning and reinforcement learning to explain human behaviour and development?
• Can computational approaches to these problems inform our understanding of the brain, and vice-versa?
• When can perceptual learning and development be understood as improvement of probabilistic inference (vs simply improved sensitivity), and why do different computational abilities (e.g. weighted averaging) develop when they do? Why and when do these develop atypically?
• How might humans or machines “learn to learn” – e.g. which knowledge is needed; which functions to optimise?
• How efficiently do people learn probabilities compared to ideal learners?
• How well can people report their state of knowledge about uncertainty (their confidence)?

Mechanisms for representing and computing with uncertainty

• When do populations of neurons represent likelihood functions, simpler summary statistics, or carry out sampling? What is the format of uncertainty?
• Can we relate these distinctions to different neuroanatomical pathways and the hierarchical nature of the brain?
• What other limitations are there to the brain’s precision at computing solutions to probabilistic perceptual and motor problems? Can we represent very small and very large probabilities?
• What role does metacognition (knowing own confidence / uncertainty) play in decisions – and how is this knowledge encoded?
• To what extent is probabilistic information used at early (perception) vs later (response / decision) stages?
• Are the mechanisms for probabilistic representation common or different across domains?
• At which level (e.g. based on Marr’s levels) do current models represent uncertainty?