Interested in reusing summarization resources in other languages, we investigated whether summary evaluations remain valid after applying machine translation to a dataset. Our findings were published in the Proceedings of the NAACL 2022 conference.

As language models and multimodal models become more powerful, domain-specific finetuning may no longer be the dominant approach to powering AI applications. I looked into the growing literature on AI ensembles and built a Streamlit demo of these approaches to understand their abilities and failure modes.

Predicting blood platelet demand using online and GP regression

Experiments in editing large, pretrained language models using meta-learning techniques

Working through a common NLP benchmark using classic ML and transformer-based methods

Exploration of deep learning architectures used for generative text style transfer