An explosion in the popularity of transformerbased language models (such as GPT-3, BERT, RoBERTa, and ALBERT) has opened the doors to new machine learning applications involving language modeling, text generation, and more. However, recent scrutiny reveals that these language models contain inherent biases towards certain demographics reflected in their training data. While research has tried mitigating this problem, existing approaches either fail to remove the bias completely, degrade performance ("catastrophic forgetting"), or are costly to execute. This work examines how to reduce gender bias in a GPT-2 language model by fine-tuning less than 1% of its parameters. Through quantitative benchmarks, we show that this is a viable way to reduce prejudice in pre-trained language models while remaining cost-effective at scale.

No abstract

Objective Driven exercise (DEx) is a serious and common feature of eating disorders (EDs), but current understanding of factors that give rise to and maintain DEx is limited. DEx may be reinforced through its effects on the threat reduction and reward systems. The current protocol is designed to evaluate acute psychobiological response to exercise among female participants (age 16‐22) with and without EDs. Method Twenty medically‐stable participants with restrictive‐spectrum EDs and 20 healthy control (HC) participants will complete study screening and three task visits which will include two 30‐minute bouts of aerobic exercise. Results We aim to validate and demonstrate feasibility of two tasks capturing exercise response in this sample. Further, we will estimate the degree to which a bout of exercise impacts state body image, affect, and circulating concentrations of biological markers among participants, and we will examine whether the impact of exercise on psychological outcomes may differ across ED and HC groups. Discussion Completion of this project will contribute to the conceptualization of DEx and how individuals' acute biological and affective responses to exercise contribute to risk for and maintenance of DEx.