Project description
How the brain learns to optimize human decision making
Training and experience can lead to long-lasting improvements in our ability to make decisions across different domains. For example, consider learning how to inspect a noisy X-ray image to issue an accurate diagnosis or how to choose between different stock options to maximize financial returns. These seemingly disparate learning scenarios have so far been studied in isolation. The aim of the EU-funded DyNeRfusion project is to develop a unified framework for integrating these different lines of research and characterize the neurobiological processes underlying learning and decision making in the human brain. To do so, it will develop state-of-the-art brain imaging (fusion of EEG and fMRI), by leveraging novel machine-learning techniques and mathematical modeling of human behavior. The project will offer a comprehensive account of how our brains learn to optimize decisions, extending beyond what could be inferred with more traditional brain imaging techniques.
Objective
Training and experience can lead to long-lasting improvements in our ability to make decisions based on either ambiguous sensory or probabilistic information (e.g. learning to diagnose a noisy x-ray image or betting on the stock market). These two processes are referred to as perceptual and probabilistic/reward learning, respectively. Despite considerable efforts to uncover the neural systems involved in these processes, perceptual and reward learning have largely been studied in separate lines of research using divergent learning mechanisms. The primary aim of this proposal is to develop a unified framework for integrating these lines of research and understand the extent to which they share a common computational and neurobiological basis. Specifically, we will test the proposition that both the perceptual and reward systems could be understood in a common framework of “reward maximization”, whereby a domain-general reinforcement-guided learning mechanism – based on separate prediction error representations – facilitates future actions and adaptive behavior. To offer a comprehensive spatiotemporal characterization of the relevant networks and their computational principles we will adopt a state-of-the-art multimodal neuroimaging approach to fuse simultaneously-acquired EEG and fMRI data, via machine-learning-inspired multivariate single-trial analysis techniques and computational modelling. The project’s ultimate goal is to empower a level of neuronal and mechanistic understanding that extends beyond what could be inferred with each of these modalities in isolation. We will achieve this goal by exploiting endogenous trial-by-trial electrophysiological variability to build parametric fMRI predictors that can offer additional explanatory power than what can already be achieved by stimulus- or behaviorally-derived predictors, allowing us to go over and beyond what has been reported previously in the literature.
Fields of science
Keywords
Programme(s)
Funding Scheme
ERC-COG - Consolidator GrantHost institution
G12 8QQ Glasgow
United Kingdom