A research paper titled Modeling Recurrent Neural Networks in Serial Recall Paradigm with Dynamic Self-excitation, co-authored and presented by Purvikalyani Prasannah, UG Cohort of 2023–27, SIAS, with Dr Rakesh Sengupta, Assistant Professor, Psychology, SIAS, has been published by Springer, Cham, in Communications in Computer and Information Science (Vol 2921) as part of the proceedings of the 5th International Conference on Advanced Network Technologies and Intelligent Computing (ANTIC 2025). It was presented at ANTIC 2025, held at Atal Bihari Vajpayee Indian Institute of Information Technology and Management (ABV-IIITM), Gwalior, India. The paper emerged out of the final project completed as part of the undergraduate course Programming for Psychologists.
This project was conducted within the Computational Cognition Lab and represents a significant step in bridging artificial intelligence with cognitive neuroscience by modeling how the brain temporarily stores visual information. How does the brain remember a sequence of visual items? Past mathematical models assumed the neural mechanisms keeping these memories active were static. The team developed a Recurrent Neural Network (RNN) model where the neural “self-excitation” is dynamic—meaning it changes over time. The model successfully replicates human behavioral quirks, such as the “Primacy Effect” (remembering the first thing you saw) and the “Recency Effect” (remembering the last thing you saw), proving that a single decaying neural signal can explain these complex human memory traits.
In recognition of this work, Purvikalyani was also awarded the Young Scientist DST Travel Grant to present the paper at the conference.
