SES & Machine Learning

During my summer internship at the Princeton Neuroscience Institute, I delved into the relationship between early life stress and socioeconomic status with the use of machine learning. Early life stress (ELS) has been highlighted as a risk factor for psychopathology throughout the lifespan, namely through alteration of the brain reward circuitry. One such source of ELS is low socioeconomic status (SES). For example, limited financial resources may cause significant stress due to feelings of uncertainty; relatively marginal losses may lead to insecurity in maintaining one’s survival resources. Thus, we hypothesized that individuals from a lower SES background will have greater sensitivity to rewards and losses, particularly in terms of promoting survival and risk-aversion. Despite there being empirical support for this notion among samples from low-SES groups specifically, work on general ELS and adversity falls short of addressing uncertainty and insecurity in reward processes. To assess this gap, this study proposed a task to model reinforcement-learning of rewards and losses in contexts of uncertainty. In a simulated environment, computational agents underwent a critical phase, wherein they were placed in one of two environments which differed in how rewarding they were. Each of these critical periods was designed to reflect the reward statistics of real-life aversive and positive experiences. Following the critical phase, all agents underwent the same post-critical phase, wherein they responded to stimuli to receive rewards and losses. Based on the agent’s learning in the critical phase, we expected differing behavior in the post-critical phase on the basis of the agent’s critical environment—with greater uncertainty leading to greater sensitivity to rewards, and thus stronger belief updating following received feedback. This study demonstrates how simulation can be used to assess hypotheses about early life stressors and later life reward and loss sensitivity.

I completed this project under the guidance of Dr. Rachel Bedder in Dr. Yael Niv’s “Niv” Lab at Princeton.

The code for this project can be accessed in my github repository, titled “SES-Machine-Learning”

This was the final poster: