GPT as a Therapists
Developed a distilled generative neural network using transformer architecture (distil-gpt2) to generate empathetic reflections in response to patient cases, addressing challenges in data acquisition and psychology domain expertise. Implemented training, evaluation, and sequence generation processes with insights into attention mechanisms for model behavior analysis and future optimization considerations, aiming for multilingual understanding and enhanced generative capabilities.
The Disambiguator
The goal of this project is to develop a bi-LSTM model trained to distinguish between the proper use of the and a in a given text. The disambiguation model is tested on Charles Dickens’ “A Tale of Two Cities” and resides in the project’s repository. Our training dataset is created from other Dickens’ works, by masking every instance of the and a in the texts and training the model to predict the correct instance label. For further details on this project have a look at the Report.