Speakers
Overview
This session focused on how computational models are used to explain and predict decision-making, and why modelling choices matter for the interpretation of our findings. Speakers highlighted different modelling traditions, from decision neuroscience (neuroeconomics) and learning-based accounts to reinforcement learning frameworks and clinically informed approaches. The discussion emphasised what models add to the experimental approach, such as identifying latent processes, testing competing mechanisms, informing theories, and improving generalisation. A recurring theme was how to connect formal models to behaviour in real tasks and real populations.
Key Points
• Computational models are useful tools for generating plausible, mechanistic explanations.
• Different fields focus on different targets: learning and adaptation, control and effort, clinical relevance, and value-based choice.
• Model comparison and validation are central to deciding which explanation is credible.
• The session highlighted the tradeoff between model simplicity and interpretability versus complexity, fit, and parsimony.
• Open question: how can we reliably link model parameters to outcomes that matter outside the task, such as symptoms, functioning, or real-world behaviour?
Next Steps
Possible collaborations: linking learning and control models within shared task designs, connecting model parameters to individual differences or clinical markers, and sharing validation practices.
If you're interested in computational modelling of decision-making and would like to connect with the speakers or suggest a speaker for another session, contact Catalina.