(Funded by and in collaboration with “Reality Labs-Meta”)• Developed a novel AI algorithm (using PyTorch) to overcome the exploding/vanishing gradient problem in continuous-time natural signal inputs in a novel neural network. . Enabled the network to learn along a continuum of time-scales. Contrary to the currently available networks, this model does not need to adjust its parameters to learn the relevant time-scales. . The algorithm performs better compared to the other well-known solutions on several time-series prediction and decoding tasks.• Studied human behavior in memory retrieval tasks to show that history is encoded as logarithmically compressed timelines in the brain. . These findings from human memory enabled us to develop the above-mentioned novel AI model.

• As a consulting project, created an AWS-based Python pipeline to optimize the length of a survey on employee satisfaction in the workplace. . Performed coarse-grained sentiment analysis on free-text questions in the survey using pre-trained solutions such as Flair and BERT, to have the same score across all the questions. . Clustered the questions in the survey hierarchically based on their mutual information and Chi-square test to infer any complex relationship between the questions. . Created and implemented a malingering scorer to detect inconsistent respondents. . Optimized the length of the survey using ML techniques (e.g., linear regression, gradient boosting and random forest) to improve employee engagement. The number of questions decreased by a factor of 6 without any significant loss of performance. . Designed an online ML-based survey dashboard using Streamlit. • As a consultant for an industry-sponsored project in collaboration with academia, supervised a Ph.D. student to develop a risk score system for the early detection of COVID-19 cases. . Pre-processed and analyzed the data collected from employees of a company. . Created a set of operational constraints as input to our complex optimizer (e.g., IBM CPLEX) to increase the interpretability/explainability of the model. . The developed risk score is highly interpretable and easily scored to the extent that it does not require a computer or even a calculator.

• Utilized information theory (e.g., mutual information) to manipulate effective connectivity in neuronal networks. In this project, we improved our understanding of information flow in neuronal networks.• Applied Recurrent Neural Network and statistical analyses to clarify the confidence evolution in perceptual decision-making processes.• Supervised and led an M.Sc. student, as a thesis advisor, to assess information flow in neuronal networks to recognize the hubs of information flow.

In the NeuroTheory Center of Columbia University, • Studied supersaturation of neuronal response for high contrast stimuli in irregular/asynchronous patterns in Gamma rhythms in the visual cortex.

In the Buzsaki Lab at NYU,• Used time-series analytics and statistical methods to analyze causality in signals from different regions of the brain, using both simulated data and electrophysiological recordings from behaving animals.

In Theoretical Neuroscience Group at IASBS,• Investigated stable synchronous states in neuronal networks in presence of inhomogeneity.• Explained stimulus-dependent synchronization and dynamic channels of communication in neuronal networks, specifically temporal engagement and disengagement of neuronal activations.• Supervised and led an M.Sc. student, as a thesis advisor, to study the synchronized oscillations in presence of communication delay in balanced neuronal networks. • Served as a Teaching Assistant for “Introduction to Neuronal Networks”, “Synchronization in Neuronal Networks” & “Dynamical Systems and Chaos” courses.

• Developed a novel community detection algorithm based on noise injection to the graphs.• Derived an exact theoretical solution considering the stability of steady states in networks of phase oscillators.• Developed a theoretical framework to simply analyze Hamiltonian using Neother Theory and Symmetry.