Some of the research projects I undertook:
Contributed to the “Scene Summarization” project, mentored by Prof. Chen Feng. This project, SceneSum, focuses on condensing extensive image collections (scenes) into a concise set of key images that highlight spatial diversity. The work is currently in the process of review for the Conference on Computer Vision and Pattern Recognition (CVPR), and its details can be accessed via the arXiv link: Scene Summarization on arXiv .
Successfully completed my Master’s thesis, supervised by Prof. Yao Wang, on the topic “Segmenting Metastatic Brain Tumors Using Deep Learning”. This research involved an in-depth analysis of the NYUMets dataset, employing various temporal segmentation techniques in neural networks. The thesis is publicly available and can be accessed here: Master’s Thesis by Ankush Pratap.
Some of the course projects I undertook:
MuGAN – Adversarial Music Generation and Genre Transformation: Developed an innovative GAN-based neural network, MuGAN, which specializes in creating unique music pieces in specified genres or transforming existing music into different genres. This project showcases the fusion of artificial intelligence and creative arts in music generation. Explore the project on GitHub.
Backdoor Attack Detection for BadNets: Designed a specialized detector for identifying backdoor attacks in BadNets, using a dataset sourced from YouTube faces. This detector enhances network security by employing a pruning defense strategy, effectively removing select network parameters to bolster resistance against sophisticated adversarial attacks. Learn more about this project.
Advanced Recommender System Development: Engineered a cutting-edge Recommender System, initially using Popularity-based methods and subsequently advancing to Collaborative Filtering approaches. This development significantly enhanced key performance metrics such as precision, NDCG, and Catalog coverage. The project also entailed benchmarking analyses contrasting single-machine and cluster-based approaches, leading to substantial reductions in model fitting time. Check out the code here.
Foundational Work in CNN, NLP, and Image Segmentation: Undertook the challenging task of developing some of the most fundamental architectures in Convolutional Neural Networks (CNN), Natural Language Processing (NLP), and Image Segmentation from the ground up. This work demonstrates a deep understanding of and practical skills in these critical areas of machine learning and artificial intelligence. Check github for codes.