Building the data science and machine learning applications on telemetry, events, and logs big data from ground-up

May 2020: I was selected for the Rock Star Spot Award for stellar contribution and performance in View Inc. COVID-SENSE AI-VISION product to fight the Corona-virusCOVID-SENSE:Developing AI-based solutions for re-entry to the workplace after COVID pandemic including Face Mask Detection, Sneeze & Cough Detection based on audio sensors, Sanitization Reminders based on Human Activity Detection, and Social Distancing Monitoring based on Human Detection and Pose EstimationOptimizing the AI models and CV algorithms to run efficiently on edge devices such as NVIDIA Jetson Nano, Google Coral, and Raspberry PiData Quality Module:Anomaly detection & missing value imputation in time-series data of temperature, humidity, CO2, light, and sound sensors via GAN \& LSTM models to increase accuracy by 15-30%Adjust for Drift and Obstruction of Air Quality, Light, Acoustics, and Thermal SensorsThermal Wellness:Fever Detection based on the combination of thermal IR and visual cameras via facial landmarks detection modelsExtract patterns & insights from office buildings thermal and humidity sensors time-series data to enhance occupants' wellness

Research: Distributed Adversarially Robust Machine Learning via Saddle DynamicsWe propose a novel discrete-time dynamical system-based framework for achieving adversarial robustness in machine learning models. Our algorithm is originated from robust optimization, which aims to find the saddle point of a min-max optimization problem in the presence of uncertainties. The robust learning problem is formulated as a robust optimization problem, and we introduce a discrete-time algorithm based on a saddle-point dynamical system (SDS) to solve this problem. Under the assumptions that the cost function is convex and uncertainties enter concavely in the robust learning problem, we analytically show that using a diminishing step-size, the stochastic version of our algorithm, SSDS converges asymptotically to the robust optimal solution. The algorithm is deployed for the training of adversarially robust deep neural networks. Although such training involves highly non-convex non-concave robust optimization problems, empirical results show that the algorithm can achieve significant robustness for deep learning. We compare the performance of our SSDS model to other state-of-the-art robust models, e.g., trained using the projected gradient descent (PGD)-training approach. From the empirical results, we find that SSDS training is computationally inexpensive (compared to PGD-training) while achieving comparable performances. SSDS training also helps robust models to maintain a relatively high level of performance for clean data as well as under black-box attacks.

Software Development: Setting up a pipeline for data collection, data extraction, noise reduction, feature extraction, and faulty wafer detection by tree-based and support vector machines classification methodsEfficiency Achievement: Boosted the run-time speed of data analysis software for silicon wafers classification (3x improvement from 6.5 to 2 hours)Theoretical Contribution: Extracting ‘twice accurate’ characteristics from the magnetic field data, thanks to exploring the intrinsic properties of datasetData Visualization: Coming up with new ways of interactive and declarative data visualization as per team's desire with Bokeh and Altair Python librariesBig Data Processing: Analyzing the big data of manufacturing and development processes of silicon wafers in a parallelized manner for more efficiency (the largest data-set worked with: TB order)User-Friendly Software: Setting up a GUI to make data analysis and visualization easily accessible by the team with limited knowledge of data science