Analysis of Machine Reading Comprehension Models
Conducted an extensive study of the state of the art Machine Reading Comprehension neural models(MRC) and benchmarked the MRC datasets to understand the peculiarities of neural models by performing qualitative and quantitative analysis. This study was done under the guidance of Dr. Mathew Lease in the Crowdsourcing and Information Retrieval lab at the University of Texas at Austin
​
Database: SQL
Algorithms: Recurrent Neural Network, Long-Short-Term-Memory Network
​
Fake Visuals Recognition: Human vs Computer
This goal of this project is to evaluate the ability of humans and computers in recognizing fake visuals. The study was threefold,
1. Developed a synthetic image generator using Generative Adversarial Network that was trained on thousands of actual human portraits.
2. Trained a Convolutional Neural Network(CNN) classifier that distinguishes real and fake visuals
3. Conducted a survey from participants to identify real and fake visuals.
The results concluded that the CNN classifier was better at classifying fake visuals with an accuracy of 91% over humans at 41%. The project won the third place for "Popular Vote for Best Project Award" in the Crowdsourcing in Computer Vision class.
​
Database: SQL
Algorithms: Nvidea Style GAN Architecture, Inception V3 (for CNN).
Dell Smart Policy Advisor
This tool captures insights from various sources on Dell laptops and PCs from sources such as 'Dell Support Assist' and 'PC Doctor' to rank alerts leading to a hardware dispatch. The tool is currently predicting failures of PC commodities such as Hardware and Battery. An iterative machine learning pipeline using PySpark and Hadoop was developed to periodically assess alert trends.
​
Database: Hive(Hadoop)
Algorithms: Random Forest, SMOTE, Convolutional Neural Networks