Contact
email: fpasha@fas.harvard.edu
About
I am a Research Associate at the Harvard University in Cambridge, MA, where I am also a fellow in CBS-NTT Program in Physics of Intelligence. I am interested in various problems at the intersection of Neuroscience and Artificial Intelligence, and specifically how these two fields can benefit each other.
I received my PhD from the Johns Hopkins University, was an Instructor in Medicine at the Harvard Medical School, and later a Swartz Fellow in Theoretical Neuroscience at the Cold Spring Harbor Laboratory in New York. For more information on some of my past and current projects please see below.
I received my PhD from the Johns Hopkins University, was an Instructor in Medicine at the Harvard Medical School, and later a Swartz Fellow in Theoretical Neuroscience at the Cold Spring Harbor Laboratory in New York. For more information on some of my past and current projects please see below.
Research Highlights
Representational drift in neural network models of the brain
In this project, I investigate the theoretical basis of representational drift in neural networks. Specifically, I am interested in how the dynamics of learning (e.g. via the Stochastic Gradient Descent) can lead to drift in the high-dimensional space of parameters, all while maintaining a stable task performance. A preprint of this work, which is accepted in ICML 2023, is available on arXiv.
Geometrical basis of reentry formation in the cardiac fibrotic tissue
In this work, I studied the structural basis of reentry (spiral) formation in cardiac fibrotic tissue, a problem with vast clinical implications, as it can lead to sudden cardiac death. I created a novel conceptual framework that combined functional aspect of wave propagation and the geometrical clusters of fibrosis, and showed that the probability of reentry could be accurately predicted across different conditions. This work was featured on the cover of Physical Review Letters. The paper can be accessed here.
High-resolution diffusion tensor imaging of human atria
The structure of the cardiac tissue is highly anisotropic, with this anisotropy modulating the electrical and mechanical activity of the heart. In this project, I designed a high-resolution Diffusion Tensor Imaging (DTI) methodology and used that to image the 3D myofiber architecture of intact human atria. Such reconstruction of fiber field was a first of its kind and revealed fascinating and intricate geometry of fiber bundles in thin atrial walls. See the main paper here, and more related to this work here.