Fourth-year Ph.D. student at UC Riverside in the Dept. of Electrical Engineering with a focus on the application of AI towards the advancement of neuroscience and control theoryAdvisor: Prof. Erfan NozariProject: Modeling and Control of Responsive Neurostimulation (RNS) for the treatment of epilepsy>Developed low-compute (100 MB) models of the neural dynamics under electric stimulation using large-scale iEEG data>Current focus on enhancing the reliability of seizure forecasting systems using state-space models

Sensing Systems Lab, Intel Labs, IndiaProject: Multi-agent collaboration of industrial robot manipulators using Reinforcement Learning>Designed a library of robot action primitives using DDPG-HER and a customized Gym+MuJoCo simulator-based training environment>Reduced learning time by 50% by developing a custom human-to-robot deep-imitation toolkit based on ROS & OpenCV to learn robust policies>Generated IP by contributing to the invention of four novel concepts, resulting in one granted patent and three pending patent applications

Primary work involved brain tumor segmentation using Deep Learning Algorithms. Further, improving the segmentation model in terms of interpretability using Reinforcement learning so as to infer the decision process as understood by humans. Co-authored ‘Brain Tumor Segmentation using Dense Fully Convolutional Neural Network’ which was accepted at International Conference on Computer Assisted Intervention (MICCAI), Quebec, Canada, 20171) Improved DL interpretability using Recurrent Attention Modeling and gradCAM. 2) Achieved 87% accuracy in tumor segmentation of brain MRI using 103 layer Convolutional Neural Network3) Improved the prediction accuracy by ~1% using Conditional Random Field (CRF) post processing paradigm

Analyzed factors influencing the quality of image in regard to 3D-reconstruction through a correlation study. Used the features extracted to achieve an accuracy of 94% using Random Forest algorithm in Python to classify photos based on image quality. Reduced data required for outlier removal to ~1% by conceptualizing a smart, semi-supervised learning based algorithm which uses unsupervised techniques like DBSCAN.Pioneered the R&D for Automatic Visual Inspection in medium scale industries to achieve ~5% labor reduction

Project: Automated Guided Vehicles (AGV) system design & control >Implemented sensor fusion and optimal control, integrating IMU and motor-encoder data with an energy-usage model to improve battery life>Developed the AGV microcontroller codebase using C++ and decreased AGV path deviation by 20%