The CDS Monthly Research Feature

The human brain processes information with remarkable efficiency, leaving most neural pathways inactive at any given moment to save energy while maintaining high cognitive performance. Researchers are now bringing this biological concept of “sparsity” to artificial intelligence, proving that computer vision models can ignore up to 90 percent of their own learned data without losing their ability to accurately classify images. This phenomenon is explored in a paper titled “Rectified LpJEPA: Joint-Embedding Predictive Architectures with Sparse and Maximum-Entropy Representations.” The work was co-authored by CDS PhD student Yilun Kuang, CILVR researcher Yash Dagade, former CDS Faculty Fellow Tim G. J. Rudner, researcher Randall Balestriero, and CDS founding director Yann LeCun.

Roughly 10 to 20 percent of patients referred with a multiple sclerosis diagnosis do not actually have the disease. CDS PhD student Jiajian Ma and his colleagues developed a new deep learning model that aims to lower this error rate by analyzing the parts of the brain that look healthy to the human eye. The paper, titled “Diagnosis of Multiple Sclerosis Using Multimodal Deep Learning Integrating Lesion and Normal-Appearing White Matter: A Retrospective Study With International Multicentre External Validation,” presents a method for distinguishing multiple sclerosis from its many mimics. An oral presentation at RSNA 2025 won the Kuo York Chynn Neuroradiology Research Award.

Large language models might pass medical licensing exams with flying colors, but they often struggle to answer the mundane yet critical questions that keep a hospital running: Will this patient be readmitted in thirty days? Will insurance deny this claim? CDS PhD student Lavender Jiang and her colleagues address this gap in their new paper, “Generalist Foundation Models Are Not Clinical Enough for Hospital Operations.” Working with CDS PhD alum Angelica Chen, CDS Professor of Computer Science and Data Science Kyunghyun Cho, and CDS-affiliated Associate Professor of Neurosurgery Eric Oermann, Jiang demonstrated that smaller, specialized models trained on internal hospital data significantly outperform massive generalist models on operational tasks.

When training AI models, the most difficult and specialized tasks are often precisely the ones with the least available data. To overcome this bottleneck, MIT PhD student Shobhita Sundaram and CDS Silver Professor of Computer Science, Mathematics, and Data Science Julia Kempe developed a method that allows AI models to generate their own intermediate practice problems to master concepts they previously could not solve.

Check out our Faculty Interview Series on YouTube. We sat down with several faculty members to discuss their advice for prospective students, what they're excited about in research, and much more! 

To learn more about our research, please visit the CDS Research Overview page.

Contact Email: cds-communications@nyu.edu

Copyright (C) 2026 Center for Data Science. All rights reserved.

60 5th Avenue, 7th Floor, New York, NY 10011

Unsubscribe  |  Manage your preferences