My research lies at the intersection of human cognition and artificial intelligence. I use computational and experimental methods to investigate how people perceive, learn, represent, and remember information, and what these processes can teach us about intelligence more broadly.
My work so far has examined human memory and attention across visual and linguistic domains, combining behavioral experiments, eye-tracking, neural network models, and semantic network analyses. These investigations have also shaped my interest in broader questions: what constitutes understanding or intentionality in artificial systems, how human and machine reasoning differ, and whether principles of human cognition can inform the development of AI systems that learn and generalize more effectively.
Hover or select a word. Then switch languages and watch the same words regroup.
People differ in how good their memory is, yet they tend to agree on which items are memorable. I study which properties of words and videos, and of their representations, predict that agreement.
A model can predict a behavioral measure without processing information the way people do. I compare where models and people attend, and how model-derived structure relates to human memory.
If memorability reflects something general about items, it should survive translation. If it depends on how a language organizes meaning, it may not. I test this with English and Hindi.
Peer-reviewed, WACV 2025
When a model is trained only to predict how memorable a video is, does it end up attending to the same things people look at? We built a CNN–Transformer memorability model whose attention can be read out over space and time, collected eye-tracking data from people doing a video memory task, and compared the two using saliency metrics, temporal analyses, and panoptic segmentation.
Manuscript in preparation
Some words are remembered better than others, consistently across people. I ran a paired-associate memory experiment with 763 eligible native Hindi speakers on 221 Hindi words derived from an English memorability study, and compared item-level memorability across the two languages. In this dataset the English and Hindi scores were not significantly correlated. I am now examining how memorability relates to the structure of language-specific semantic networks, and how much of that relationship is carried by word frequency.
Research notes, reading trails, and ideas worth examining more closely. The notes below are in draft.