We are investigating three aspects of social cooperation within the framework of the Emmy-Noether research group. We are particularly interested in how these processes go awry in individuals with personality disorders. We aim at developing comprehensive neuro-computational models.
Before cooperating, humans usually learn about each others’ personality by generalizing across similar traits.
We have devised novel reinforcement learning (RL) models that capture how people generalize with different levels of granularity and references points (Frolichs, Rosenblau & Korn, 2023).
The basic problems of cooperation itself are encapsulated in the prisoner’s dilemma. Mutual cooperation yields the best joint payoff but unilateral defection is tempting.
Ongoing studies aim at modelling cooperative decisions in stylized as well as ecologically valid games (e.g., Kuper-Smith & Korn, preprint1, preprint2).
Once cooperation is established, partners have to decide how to share the benefits.
Initial studies model social value orientation in such allocation decisions and show the involvement of the amygdala (Doppelhofer et al., 2021).
These projects combine the topics listed below. All projects benefit from close interactions within the Department of General Adult Psychiatry and in particular with Sabine Herpertz and her lab. We closely collaborate with the lab of Jan Gläscher at the Institute of Systems Neuroscience (University Medical Center Hamburg-Eppendorf, UKE). Gabriela Rosenblau at George Washington University (Washington, DC) collaborates on projects involving social learning.
Heuristic versus optimal decision-making and learning
We are interested in understanding how humans combine myopic heuristic andrationally optimal policies. Such optimal policies necessitate integrating over multiple future time steps and across multiple social settings. We develop complex decision-making and learning tasks and models based on Reinforcement Learning (RL) and in particular based on Markov Decision Processes (MDPs).
We are interested in understanding how humans combine myopic heuristic andrationally optimal policies. Such optimal policies necessitate integrating over multiple future time steps and across multiple social settings. We develop complex decision-making and learning tasks and models based on Reinforcement Learning (RL) and in particular based on Markov Decision Processes (MDPs).
Results of studies from Christoph’s postdoc at the University of Zurich provide evidence for a trade-off between heuristic and optimal decision policies, which relate to activity in the medial prefrontal cortex (Korn & Bach, 2015; 2018, 2019). Such multi-step decisions play an integral role in many approach-avoidance conflicts, which are sensitive to anxiolytic medication and lesions of the amygdala and the hippocampus (Korn et al., 2017).
Apparent biases in (non-)social information processing
Studies from Christoph’s Master’s thesis in London and his PhD in Berlin demonstrate that humans process self-relevant information about their future and their character in an apparently positively biased way (e.g., Sharot*, Korn*, Dolan, 2011; Korn et al., 2012). These self-related positivity biases are reduced in individuals with psychiatric disorders, i.e., with depression (Korn*, Sharot* et al., 2014) and borderline personality disorder (Korn et al., 2016a, 2016b).
Studies from Christoph’s Master’s thesis in London and his PhD in Berlin demonstrate that humans process self-relevant information about their future and their character in an apparently positively biased way (e.g., Sharot*, Korn*, Dolan, 2011; Korn et al., 2012). These self-related positivity biases are reduced in individuals with psychiatric disorders, i.e., with depression (Korn*, Sharot* et al., 2014) and borderline personality disorder (Korn et al., 2016a, 2016b).
In his PhD thesis, Christoph developed a neurocognitive model that relates this type of biased information processing to the neural processing of social rewards and mentalizing (Korn et al., 2012; Korn et al., 2014).
Multimodal integration, RSA, and DNNs
In ongoing collaborative projects with colleagues from Hamburg, we harness Representational Similarity Analyses (RSA) of fMRI data and Deep Neural Networks (DNNs) to investigate how the human brain integrates multimodal stimuli (Korn et al., 2021).
In ongoing collaborative projects with colleagues from Hamburg, we harness Representational Similarity Analyses (RSA) of fMRI data and Deep Neural Networks (DNNs) to investigate how the human brain integrates multimodal stimuli (Korn et al., 2021).
Specifically, we obtained fMRI data from human participants presented with 72 audio-visual stimuli of actors/actresses depicting six different emotions. The same stimuli were classified with high accuracy in our pre-trained DNNs built according to a cross-channel convolutional architecture. Inspired by recent studies using RSA for uni-modal stimuli, we assessed the similarities between the layers of the DNN and the fMRI data. As hypothesized, we identified gradients in pattern similarities along the different layers of the auditory, visual, and cross-modal channels of the DNNs.
Model-based approaches to pupillometry
Together with Dominik Bach, Christoph has developed model-based procedures for analyzing pupil responses.
In a first step, we established and validated a parsimonious description of luminance-elicited pupil responses (on the basis of two response functions). This description furnishes a characterization of both prolonged and brief pupil responses (Korn & Bach, 2016). In a second step, we validated an analysis procedure for fear conditioning paradigms that provides better discrimination than previous methods (Korn, et al., 2016). These models have been summarized within a review paper (Bach, Castegnetti, Korn, et al., 2018).
Together with Dominik Bach, Christoph has developed model-based procedures for analyzing pupil responses.
In a first step, we established and validated a parsimonious description of luminance-elicited pupil responses (on the basis of two response functions). This description furnishes a characterization of both prolonged and brief pupil responses (Korn & Bach, 2016). In a second step, we validated an analysis procedure for fear conditioning paradigms that provides better discrimination than previous methods (Korn, et al., 2016). These models have been summarized within a review paper (Bach, Castegnetti, Korn, et al., 2018).
Commitment to open science
We believe that science should be open and reproducible.
Thanks to the efforts by Dominik Bach and many people in his lab, all developed models for analyzing psychophysiological data are incorporated in the MATLAB toolbox PsPM and most datasets are publicly available.
