Marina Abd El Shahid

One of eight team members, exploring avenues of creating an accurate and debiased classifier model, able to differentiate between healthy and unhealthy communication within Reddit forums. The initial deliverable will be debiasing an existing model. Time permitting, this could involve creating a classifier model that takes BERT model embeddings to recognize and moderate harmful messages within Reddit forums with high accuracy. Currently, I've researched and created documentation on adversarial learning, helping my team and I gain an understanding of the different measures of fairness and how that dictates the information given to the adversary.