Shashi Bhushan Jha

• Conduct research on machine learning algorithms and techniques to improve accuracy and efficiency of models. • Acquire novel aerospace composite components image dataset from aerospace research lab. • Develop and implement new deep learning models and algorithms for various applications such as quality control for aerospace manufacturing and flight delay prediction for US Airlines. • Write research papers, publish findings in reputed journals. • Collaborate with other researchers to advance the field. • Analyze and interpret large datasets using machine learning models and statistical techniques. • Stay up to date with the latest research and technology trends in machine learning and related fields.

• Collaborated within a workstream alongside a Machine Learning Engineer/Modeler and cross-functional partners. • Conducted analysis on large datasets using Python/MySQL and scripting languages to uncover significant insights. • Provided quantitative assessment of the advantages and disadvantages of different model classes of Graph ML • Managed a massive dataset (billions of records) for GML using machine type of 4 TB RAM and 160 CPUs. • Worked on PyTorch BigGraph (PBG) using a distributed system. • Methodically tested and validated advanced ML models. • Utilized statistical and mathematical modeling techniques to study and comprehend customer behavior. • Developed code for efficient data processing, cleansing, and integration across multiple sources, and deployed in a production environment.

DISSERTATION WORK • Conducted a literature review of machine vision techniques for automated optical inspection systems. • Acquired novel aerospace composite components image dataset from the aerospace research lab, and labeled the image dataset. • Designed and implemented a new hybrid deep CNN-based models, utilizing MobileNetV2 and Random Forest, for an automated optical inspection system to identify defects in Boeing aircraft components. • Trained and evaluated pre-trained, fine-tuned, and end-to-end deep CNN supervised models using an image dataset of aerospace composite materials. • Currently working on unsupervised pixel-level segmentation techniques to defect defects in images.

Projects Involved in Ph.D.: Improved CNN Model Performance by Incorporating Image Gradients as Secondary Input: • Proposed a new CNN architecture that utilizes multiple forms of input, including original images and gradients, by sharing layers across all input forms. • Incorporated image gradient as a secondary input associated with the original input image. • Trained and tested the proposed model using diverse datasets (e.g., MNIST, CIFAR10, CIFAR100) and achieved better performance than benchmark models. Flight Delay Prediction Using Hybrid Machine Learning Approach: A Case Study of US Airlines: • Collected and preprocessed 27 months of recent flight data of US Airlines. • Conducted data analysis and visualization on the dataset. • Proposed a novel hybrid machine learning approach that combines deep learning and random forest/XGBoost to predict flight delay in terms of departure, arrival, and total delay. Housing Market Prediction • Collected and preprocessed housing data from the Florida Volusia County Property Appraiser database, including socio-economic factors such as GDP, CPI, PPI, HPI, EFFR. • Developed a housing price prediction model using XGBoost, Lasso, Decision Tree, and Random Forest. • Proposed a prediction model that classifies whether a closing price is greater or less than the listing price using XGBoost, Random Forest, Voting Classifier, SVM, and Logistic Regression. Stock Market Price Prediction • Used Multiple Linear Regression Model (in a Statistics Course Project). Formal Modeling of Cyber-Physical Resource Scheduling in IIoT Cloud Environments: • Used discrete state-machine diagrams to model resource reliability and availability status, and logistics timing purposes. Published/Preprint articles can be found here: • https://scholar.google.com/citations?user=A1NjNCEAAAAJ&hl=en

Thesis: A Multi-objective Meta-heuristic Approach for TNDFS Problem • Studied a multi-objective transit network design and frequency setting (TNDFS) problem with the aim of determining a set of routes and frequency on each route for public buses. • Generated a set of routes using a novel initial route set generation (IRSG) procedure combined with genetic algorithm. • Formulated a multi-objective model to assign the frequency on each route with the objectives to minimize the passenger time and the operating cost simultaneously. • Developed several multi-objective algorithms, such as NSGA-II and MOPSO, to solve the problem. Outstanding Contribution Award in reviewing by CAIE Journal, Dec. 2016 Published articles can be found here: • https://scholar.google.com/citations?user=A1NjNCEAAAAJ&hl=en

Dissertation Title: Deep CNN-Based Automated Optical Inspection for Aerospace Components

Thesis Title: A Multi-Objective Metaheuristic Approach for Transit Network Design and Frequency Setting Problem in a Bus Transit System.

Oustanding Bachelor Student Award, Department of Computer Science, Nagpur University, Apr. 2010