We explored data manipulation techniques, explainability, prompt engineering, and unsupervised approaches (using attention matrices) to detect and mitigate hallucinations in Neural Machine Translation. (Repo Link)
Cognitive biases impact the decision-making of patients with a chronic illness. We analyzed the presence of language patterns and psycholinguistic markers characteristic of cognitive biases within Reddit posts by individuals who have self-disclosed a chronic illness using visual, interrupted time series and n-gram analyses of associated LIWC dimensions. (Report Link)
A Visual Question Answering System to help the visually impaired with navigation. We explored large pre-trained models (VGGNet and BERT) for images and text to answer queries by the visually impaired. (Link to project site)
We analyzed different distance metrics including static and learned distance metrics in Prototypical Networks. (Link to paper)
We analyzed patterns in mental discourse of the student community in Reddit. We performed topic modeling and time series modeling and analysis that quantifies relationships between the qualitative topics. (Paper Link)
We used social media data from Reddit to understand
audience’s perception about biases in movies.
(Presentation Link)