I am currently learning how to make computers "learn", so that they can leverage data to solve real-world problems that positively impact society.
Developed a Convolutional Neural Network (CNN) based model for Facial Emotion Recognition (FER) and a real-time FER model integrated with a webcam using Local Binary Patterns (LBP). Used Viola Jones Algorithm (Haar Filter / Cascades) for face detection and CNNs, Local Binary Patterns (LBP), & Fisher Face Algorithm for emotion detection in developing these models.
Created a Tableau based data visualisation system using OpenStreetMap (OSM) and GPS pings to analyse anomalous actions of food delivery executives. Used clustering of GPS pings and neighbour-based anomaly detection for route outliers to identify safety risks arising due to malpractices by food delivery executives.
Delivered a Web Application for the software development team using Flask along with RESTful APIs and integrated it with security features (JWTokens) using Flask extensions.
Applied decision tree and Naïve Bayes algorithms to solve classification problem as a use-case of condition-based maintenance of industrial products. Gained knowledge of text mining techniques and applied them to virtual assistant/chatbots to diagnose fault detection using service record text data. Used python to develop a simple chatbot that could answer the time zone of any city. Learned about AI applications in Healthcare used for detection of lung ailments, and tracking of mother and baby health during pregnancy.
Creating a Natural Language Processing (NLP) model to perform Named Entity Recognition on a corpus of unstructured COVID-19 scholarly literature. Utilising pre-trained word embeddings (like word2vec, Glove or BERT) to learn features using a recurrent network, such as a bidirectional LSTM, for prediction of 64 fine-grained entity types such as general types (person, organization), common biomedical entity types (genes, chemicals, diseases), and COVID-19 types (coronavirus, viral protein, substrates, immune responses). Deriving insights with effective data visualisation for downstream actions.
Developed an interactive data visualization system for analyzing individual player performance and team strategy for a soccer game. Used TSNE and JavaScript to leverage Social Network Theories such as betweenness and clustering coefficients to quantify the importance of a player. Applied advanced filtering, kernel density estimation, and spatiotemporal event plotting using IPython Kernel to understand team tactics. Demonstrated tactics used by a successful English Premier League soccer team and analyzed a loss to show how this knowledge can be exploited to defeat the team.
Developed a forgery detection model for security against facial image identity theft using UNet based deep learning approach and attained over 99% pixel-wise accuracy for forgery detection. The forgery transformation model used a Latent Vector approach and generated semantically consistent facial images when tested against both cropped out real images and forged images. The transformed images were qualitatively better than generative models from literature using cropped real images.
Created a model to predict a restaurant’s survivability using customer satisfaction metrics (obtained from Yelp and Zillow dataset) and used model agnostic LIME approach to demonstrate interpretability. Deployed Machine Learning and Deep learning approaches such as Support Vector Machine model (SVM), Long Short Term Memory (LSTM), and Convolutional Neural Networks (CNNs) for model creation to predict restaurant survival. Achieved directional insights that highlighted the need for inclusion of a more robust dataset and additional attributes to enhance the efficacy of the model.
Apart from my studies, I spend most of my playing and watching sports (ee sala cup namde!!!). Since COVID, I have also become obsessed with watching videos of mouth-watering food.
Spending quality time with family and friends is very important to me.