Yashvir Sabharwal

International Research Olympiad

Cambridge, Massachusetts, United States · Hybrid

Part of the Student Board at the International Research Olympiad. Reaching over 10,000 students from 89 countries.

Civil Air Patrol

Manassas, Virginia, United States · Hybrid

$142.5MM in budget appropriations approved in Congress.

• International Air Cadet Exchange - United Kingdom (2025)
• Specialized Undergraduate Pilot Training Familiarization Course – Texas (2022)
• Cadet Executive Officer (2023-till date)
• VAWG CAC Chair (2023-2024)

Alljoined

San Francisco, California, United States · Remote

Worked on the largest EEG reconstruction dataset & project. Backed by Khosla Ventures, Conviction, Embed, Jeff Dean, Guillermo Rauch, Compound, Artis Ventures, Oliver Cameron, Xtal, First Spark Ventures, and Topology.

MedARC

Worldwide · Remote

Neuroimaging & AI Lab @ MedARC. Working on MindEye, a fMRI reconstruction and retrieval project. Partnered with Stability AI, Stanford AIMI Center, and Fast.ai.

Samyak Science Society

Portland, OR · Remote

Part of the Samyak Science Society, a 501(c)(3) non-profit organization focused on spreading STEAM awareness to inspire & and ignite the spark of curiosity with youth worldwide in local communities.

AB, Neuroscience and Economics

Yashvir Sabharwal, Balaji Rama

arXiv preprint arXiv:2412.19999 · December 2024

This paper systematically reviews 1800 EEG-to-output studies using PRISMA guidelines, analyzing state-of-the-art generative methods (GANs, VAEs, Transformers), evaluation metrics, and data challenges, while proposing a roadmap for improving decoding accuracy and cross-subject generalization.

Jonathan Xu, Bruno Aristimunha, Max Emanuel Feucht, Emma Qian, Charles Liu, Tazik Shahjahan, Martyna Spyra, Steven Zifan Zhang, Nicholas Short, Jioh Kim, Paula Perdomo, Ricky Renfeng Mao, Yashvir Sabharwal, Michael Ahedor Moaz Shoura, Adrian Nestor

arXiv preprint arXiv:2404.05553 · April 2024

We present Alljoined1, a dataset of 46,080 epochs of 64-channel EEG responses from 8 participants viewing 10,000 natural images each, designed specifically for EEG-to-image decoding with response-based timing, repetition protocols, and data quality scores.

Yashvir Sabharwal

arXiv preprint arXiv:2302.05830 · February 2023

We develop NephroNet, a ResNet-18-based classifier for RCC subtypes and a diffusion model-based tool that generates realistic RCC histopathology images in under four seconds, provided as an open-source package for medical training and algorithm testing.

Connor Lane, Daniel Z Kaplan, Tanishq Mathew Abraham, Paul S Scotti (Yashvir Sabharwal, contributor)

arXiv preprint arXiv:2510.13768 · October 2025

We transform 4D volumetric fMRI data into 2D flat map videos and train Vision Transformers on 2.3K hours of Human Connectome Project data using masked autoencoding, demonstrating power-law scaling and rich representations for state and trait decoding across subjects.

Paul S. Scotti, Atmadeep Banerjee, Jimmie Goode, Stepan Shabalin, Alex Nguyen, Ethan Cohen, Aidan J. Dempster, Nathalie Verlinde, Elad Yundler, David Weisberg, Kenneth A. Norman, Tanishq Mathew Abraham (Yashvir Sabharwal, contributor)

NeurIPS 2023 · arXiv:2305.18274

We present MindEye, an fMRI-to-image model with parallel retrieval (contrastive learning) and reconstruction (diffusion prior) submodules that achieves state-of-the-art performance, enabling accurate image retrieval from large-scale databases like LAION-5B.