Although individuals with schizophrenia show impaired feedback-driven learning on probabilistic r... more Although individuals with schizophrenia show impaired feedback-driven learning on probabilistic reversal learning (PRL) tasks, the specific factors that contribute to these deficits remain unknown. Recent work has suggested several potential causes including neurocognitive impairments, clinical symptoms, and specific types of feedback-related errors. To examine this issue, we administered a PRL task to 126 stable schizophrenia outpatients and 72 matched controls, and patients were retested 4 weeks later. The task involved an initial probabilistic discrimination learning phase and subsequent reversal phases in which subjects had to adjust their responses to sudden shifts in the reinforcement contingencies. Patients showed poorer performance than controls for both the initial discrimination and reversal learning phases of the task, and performance overall showed good test-retest reliability among patients. A subgroup analysis of patients (n = 64) and controls (n = 49) with good initia...
Background: Anxiety disorders often entail exaggerated threat prediction. Here, we address comput... more Background: Anxiety disorders often entail exaggerated threat prediction. Here, we address computational algorithms for learning threat prediction, using discriminative delay fear conditioning in healthy humans. Fear conditioning is suggested to implement a prediction error-based reinforcement learning (PERL) algorithm, but more general theories of brain function imply probabilistic computations. Methods: In three experiments (overall N ¼ 68), we recorded skin conductance (SCR) and pupil size responses (PSR). We used random-effects Bayesian model selection of maximumlikelihood fits. In a fMRI study (N ¼ 22) with 4 CS, we examined neural representations of prediction error signals in an axiomatic approach. Results: Trial-by-trial SCR and PSR trajectories over 160 trials were best described by a probabilistic learning model, while extant PERL models provided a quantitatively and qualitatively worse fit. The best-fitting probabilistic learning algorithm maps a linear combination of threat prediction and its uncertainty onto SCR, and threat prediction onto PSR (protected exceedance probabilities 0.99, 0.85, 0,98, respectively for SCR, and 0.68 for PSR). Using fMRI, we found no neural signals fulfilling necessary conditions for representation of prediction errors. Quantities from the probabilistic learning model were represented in several brain regions (p < .05 FWE). Conclusions: Overall, a probabilistic learning model provided a parsimonious description of the data, while we found no evidence that threat prediction learning relies on a PERL algorithm. These findings extend Bayesian brain theories to subcortical threat learning systems.
for helping to administer the assessments and run the SCIT groups. Enormous thanks also to Lesa H... more for helping to administer the assessments and run the SCIT groups. Enormous thanks also to Lesa Hoffman for statistical guidance and assistance. I would like to thank Mandi Daws, Marylyde Kornfeld, and Joe Swoboda for their invaluable insight and support. I would like to thank J. Rock and Mary Sullivan for reminding me of the grace that is nestled in hard work and perseverance. I have had many mentors and supervisors in the last several years, all of whom I owe thanks. The members of my doctoral committee, David DiLillo, Cynthia Willis-Esqueda, and Helen Moore deserve a special recognition for helping me to create and finish this project. Finally, I would like to acknowledge with the deepest gratitude my mentor, Will Spaulding, for his guidance and brilliance as a teacher, a doctor, and a scientist. This dissertation is dedicated to my family, and to Charles.
Although individuals with schizophrenia show impaired feedback-driven learning on probabilistic r... more Although individuals with schizophrenia show impaired feedback-driven learning on probabilistic reversal learning (PRL) tasks, the specific factors that contribute to these deficits remain unknown. Recent work has suggested several potential causes including neurocognitive impairments, clinical symptoms, and specific types of feedback-related errors. To examine this issue, we administered a PRL task to 126 stable schizophrenia outpatients and 72 matched controls, and patients were retested 4 weeks later. The task involved an initial probabilistic discrimination learning phase and subsequent reversal phases in which subjects had to adjust their responses to sudden shifts in the reinforcement contingencies. Patients showed poorer performance than controls for both the initial discrimination and reversal learning phases of the task, and performance overall showed good test-retest reliability among patients. A subgroup analysis of patients (n = 64) and controls (n = 49) with good initia...
Background: Anxiety disorders often entail exaggerated threat prediction. Here, we address comput... more Background: Anxiety disorders often entail exaggerated threat prediction. Here, we address computational algorithms for learning threat prediction, using discriminative delay fear conditioning in healthy humans. Fear conditioning is suggested to implement a prediction error-based reinforcement learning (PERL) algorithm, but more general theories of brain function imply probabilistic computations. Methods: In three experiments (overall N ¼ 68), we recorded skin conductance (SCR) and pupil size responses (PSR). We used random-effects Bayesian model selection of maximumlikelihood fits. In a fMRI study (N ¼ 22) with 4 CS, we examined neural representations of prediction error signals in an axiomatic approach. Results: Trial-by-trial SCR and PSR trajectories over 160 trials were best described by a probabilistic learning model, while extant PERL models provided a quantitatively and qualitatively worse fit. The best-fitting probabilistic learning algorithm maps a linear combination of threat prediction and its uncertainty onto SCR, and threat prediction onto PSR (protected exceedance probabilities 0.99, 0.85, 0,98, respectively for SCR, and 0.68 for PSR). Using fMRI, we found no neural signals fulfilling necessary conditions for representation of prediction errors. Quantities from the probabilistic learning model were represented in several brain regions (p < .05 FWE). Conclusions: Overall, a probabilistic learning model provided a parsimonious description of the data, while we found no evidence that threat prediction learning relies on a PERL algorithm. These findings extend Bayesian brain theories to subcortical threat learning systems.
for helping to administer the assessments and run the SCIT groups. Enormous thanks also to Lesa H... more for helping to administer the assessments and run the SCIT groups. Enormous thanks also to Lesa Hoffman for statistical guidance and assistance. I would like to thank Mandi Daws, Marylyde Kornfeld, and Joe Swoboda for their invaluable insight and support. I would like to thank J. Rock and Mary Sullivan for reminding me of the grace that is nestled in hard work and perseverance. I have had many mentors and supervisors in the last several years, all of whom I owe thanks. The members of my doctoral committee, David DiLillo, Cynthia Willis-Esqueda, and Helen Moore deserve a special recognition for helping me to create and finish this project. Finally, I would like to acknowledge with the deepest gratitude my mentor, Will Spaulding, for his guidance and brilliance as a teacher, a doctor, and a scientist. This dissertation is dedicated to my family, and to Charles.
Uploads
Papers by LENA REDDY