Cognition and Computational Psychiatry
Our vision is to understand the neural codes and computational principles enabling us to build, represent, and update a mental model of the world.
This work is informed by theoretical treatments, particularly ideas derived from reinforcement learning. Understanding how such models are constructed and represented in the brain is a fundamental question for neuroscience, and has particular importance for psychiatric research. Our research programme has two major strands:
Decision making and its impairments in psychopathology
Our previous work on this topic has concerned model-based reasoning, in particular its integration with model-free reasoning. This has motivated us to address the question of intermediate points on a spectrum between the two. Here we have shown a greater degree of complexity than implied in previous accounts, particularly in relation to credit assignment (Moran et al., 2019) with striking explanatory implications for psychopathology (Shahar et al., 2019). In social decision making we have described a novel effect on preferences engendered by having to make inter-temporal choices for a partner. This observation inspired us to provide a theoretical treatment of this effect, one that focused on uncertainty regarding one’s own values (Moutoussis et al., 2016). In work under review we have shown in our NSPN cohort that this has important implications for the trajectory of social development during adolescence.
An important methodological innovation, opening avenues for us to address more sophisticated questions in relation to decision making, has been our ability to capture the course of model-based planning through decoding of MEG signals. Our initial work here includes revealing the temporal structure of associative retrieval as well as identifying fast, non-spatial, sequence replay using MEG (Kurth-Nelson et al. 2016). Building on this, using a decoding strategy, we have shown forward and reverse replay in an MEG signal (Liu et al., 2019). In related research we examined the neural representation of serial and parallel computation, both of which are involved in model-based planning (Eldar et. al, 2016). Most recently, we have characterised the relative contribution of on-task and off-task replay to model based and model free decision making.
Neuromodulation and psychopathology
Previously in theoretical work we have proposed a mechanistic framework for how affect-learning interactions contribute to mood dynamics (Eldar et al., 2018). Empirically, we showed how unexpected outcomes alter affective state, including providing evidence that a varying reward sensitivity is predictive of subsequent fluctuations in mood (Eldar et al., 2016). Theoretically, this two-way relationship can set in train an escalating positive feedback loop, one wherein good outcomes improve mood which, in turn, improves perception of subsequent outcomes leading to further mood elevation. Building on this idea we have revealed that a positive impact on mood is best accounted for by a boost in subjective reward perception during learning, and this results in a delayed mood response. Importantly, this effect is amplified by SSRI’s, in a manner that can explain a delayed impact of these treatments. Thus, instead of influencing affect or reward sensitivity directly, SSRIs amplify a bilateral interaction between mood and reward perception.
Max Planck UCL Centre for Computational Psychiatry and Ageing Research
G. Elliott Wimmer
- Differential replay of reward and punishment paths predicts approach and avoidance Nature Neuroscience, 26 (4), 627-637 DOI: 10.1038/s41593-023-01287-7
- Risk taking for potential losses but not gains increases with time of day Scientific Reports, 13 DOI: 10.1038/s41598-023-31738-x
- View all publications by the Cognition and Computational Psychiatry team