Ishaant Agarwal

Ongoing

Building a 3D network that can denoise complex 3D data distributions such as cryo-EM images or MRI scans using only noisy image pairs for training. Our goal is to be able to retain high frequency details after cleaning, an unadressed limitation of current 3D denoising schemes.

Developed a full package for processing and analyzing video data from a single-photon mini-microscope. Used an RNN along with traditional morphological processing to extract RoIs and calcium traces from these recordings and worked to register these cells to track them across sessions individually. In addition, we devised rules and methods (using a stochastic firing model) to quantify neuron behavior and conclusively identify and segregate HD cells. Assisted with behavioral tasks, including shock training and recall experiments as well.

Developed a highly simplistic (but effective!) and scalable probabilistic model of the inner ear, focussing on cochlear amplification and modelling auditory transduction. The model and accurately predictedd empirical responses to tones and even exhibited complex phenomena like two-tone suppression.

Employed times lapse confocal fluorescence imaging to study the transport of dye labeled dextran molecules of different sizes through the nuclear pore complexes. It includes analysis of single photon as well as time-averaged fluorescence data obtained through a confocal microscope examination of cells during the diffusion process.

Extensively studied and simulated the synchronization behaviour as seen in the Kuramoto Model for ’n’ weakly coupled oscillators. Each oscillator here represented a neuron and their synchronization depicted a seizure-like behavior in animals. We looked at and found topological events like fixed points and bifurcations and investigated their generation as a way of modulating seizure response.

We set up an experimental facility, including a waterbed to test the performance various modes of underwater acousticcommunication. Conducted a theoretical analysis using different encoding schemes and simulated results for our hardware andverified our findings using UnetStack3. This helped deploy the hydrophone setup at optimal configuration for testing.