BACKGROUND

My fascination with Artificial Intelligence (AI) began in my sophomore year, a time when AI was just starting to gain widespread attention. My curiosity about this emerging field was piqued, leading me to explore what AI really entailed.

I am deeply grateful to Mr. Vignesh Kothappalli, my basketball teammate and one of the pioneers of IITG.ai – a community of AI enthusiasts at IIT Guwahati. Through IITG.ai and Prof. Andrew Ng's courses on Coursera, I took my first steps in Machine Learning.

At IIT Hyderabad, under the mentorship of Prof. Sumohana Channappayya, I dived into Generative Models. This experience rapidly advanced my understanding of Generative Adversarial Networks (GANs). During my internship, I engaged in several projects utilizing GANs, culminating in a victory at the Engineering the Eye Hackathon.

Post-internship, I continued participating in various projects with the IITG.ai community and embarked on personal projects. My journey led me to an internship at Samsung Research Institute Noida, in the multimedia department of the smartphone division. Here, my team and I developed a facial attribute editing app using STGAN – a groundbreaking technology at the time. The project's success earned us recognition as one of the best projects and a coveted pre-Placement Offer.

Further expanding my skillset, I delved into computer vision and the mathematical modeling of AI models. Prof. M K Bhuyan's classes immensely helped me grasp the intricacies of computer vision. Additionally, Prof. Tony Jacob's course on Pattern Recognition and Machine Learning and Prof. Hanumant Singh Shekhawat's course on Applied Mathematics profoundly expanded my understanding of AI and ML, offering deep insights into their mathematical foundations.

Below, I detail a couple of my significant projects:

Project #1:

FACIAL ATTRIBUTE EDITING USING STGAN

Research Internship, Samsung Research Institute, Noida

Worked under Mr. Sangeet Asthana, Project Manager, Multimedia Division, SRI Noida.
Engaged in developing AI-based real-time filters for the Samsung camera application in Samsung devices.
Utilized the STGAN model with Selective Transfer Units, trained on the CelebA dataset, for facial attribute editing.
Developed a Python Tkinter Image Viewer GUI as a provisional application to showcase the results.

A representation of STGAN model being used

The proposed model achieved a 20% accuracy gain over all contemporary models, leading to being awarded a Pre-Placement Offer (PPO) for exemplary demonstration of technical skills during the internship.

(All images used are from the original paper on STGAN. Click here)

Project #2:

TARGET RECOGNITION USING EMBEDDINGS FROM RADAR CROSS-SECTION

Bachelor Thesis, Indian Institute of Technology, Guwahati

The Team: Narendra Pal, Kevin Jose

Objective and Approach: Focused on improving object identification accuracy using Radar Cross Section (RCS) data, regardless of object orientation or motion.
Initial Experiments: Began with experiments on metal plates to establish fundamental RCS principles and refine data modeling techniques.
Dataset Generation: Developed a synthetic dataset via simulations to simulate diverse target scenarios, overcoming real-world RCS data acquisition challenges.
Technological Implementation: Utilized advanced machine learning techniques and synthetic aperture radar (SAR) for high-resolution imaging, enhanced by multi-static radars for improved coverage.
Future Directions: Highlighted improvements in embedding generators and the potential use of Generative Adversarial Networks (GANs) for more accurate RCS-to-object mapping.

A representation of RCS model being used

FULL REPORT

EXPLORING AI: HIGHLIGHTS FROM MY DEEP LEARNING VENTURES