Harshvardhan Gazula

This study shows that speakers' brains generate information-rich words through extra neural computation, evidenced by stronger neural activity before articulation, and that LLMs similarly require deeper computation for improbable words.

The paper shows that larger language models better capture linguistic structure and predict neural activity, with performance peaking in earlier layers as model size increases.

This paper introduces a decentralized linear mixed-effects model for neuroimaging analysis, enabling large-scale collaboration without data pooling, and shows it achieves similar performance to centralized models.

The paper introduces Path2MR, a pipeline for 3D histology reconstruction without MRI, revealing distinct tau and amyloid-beta pathology gradients along the hippocampal axis in Alzheimer's Disease.

The paper develops open-source tools for 3D analysis of brain dissection photographs and surface scans, achieving accurate reconstructions and segmentations comparable to MRI.

The paper uses a deep speech-to-text model to predict neural responses during spontaneous conversations, revealing a distributed cortical hierarchy for speech and language processing.

This paper introduces COINSTAC Vaults to enhance collaborative neuroimaging research through federated analysis and reproducibility by hosting standardized datasets.

The paper uses a novel explainability framework to analyze sex differences in brain network structure using fMRI data, identifying modular and localized differences in temporal brain activity.

This study uses federated analysis in COINSTAC to examine functional network connectivity in adolescent brains linked to smoking and alcohol consumption, finding hypoconnectivity and reduced synchronization.

This study finds a correspondence between the layered structure of deep language models and the temporal structure of neural activity during language processing in the human brain.

This work proposes a decentralized method for brain age estimation using MRI data and shows it achieves similar performance to centralized models.

The paper shows that brain embeddings from the IFG share geometric structure with DLM contextual embeddings, enabling zero-shot encoding and decoding of unseen words.

This paper uses COINSTAC to perform decentralized VBM analysis on multi-site neuroimaging data to identify age, BMI, and smoking-related structural brain changes.

In nine epilepsy patients, cortical activity carries next-word information up to 800 ms before onset, tracks GPT-2 surprise after onset, and fits GPT-2 contextual embeddings better than static ones.

The paper analyzes resting-state fMRI data across different frequency bins to assess functional network connectivity and finds distinct patterns at higher frequencies.

The paper introduces a decentralized classification method that reduces network load while maintaining accuracy, applied to brain imaging data.