Ipsita Ghosh

Rutgers University (DIMACS)

Investigated the application of low-rank optimization techniques to reinforcement learning from human feedback (RLHF). This work explores how low-rank structures can reduce computational costs in training large language models while maintaining alignment quality. Worked with Dr. Arunesh Sinha.

UNC Charlotte

Developed novel approaches to video summarization using state space models. Explored efficient temporal modeling techniques for understanding and summarizing long video sequences. Course project for Advanced Computer Vision with Dr. Srijan Das.

UNC Charlotte

Investigated various low-rank training techniques for deep neural networks, analyzing their impact on model performance, training speed, and memory requirements. Explored different low-rank parameterizations and their effectiveness across various architectures. Course project for Advanced Machine Learning with Dr. Li Yang.

UNC Charlotte

Analyzed the convergence properties of natural gradient descent methods in neural network optimization. Studied the geometric properties that enable faster convergence compared to standard gradient descent. Course project for Optimization for Machine Learning with Dr. Christian Kuemmerle.

Personal Project (Remote)

Developed a 2-stream adaptive graph convolution network for skeleton-based action recognition on the MSR Daily Activity 3D Dataset. Worked with Dr. Srijan Das from INRIA, France on this computer vision project.

Chennai Mathematical Institute

Implemented facial recognition using the EigenFaces method, exploring principal component analysis (PCA) for dimensionality reduction in face images. Worked with Prof. Kavita Sutar.

Chennai Mathematical Institute

Developed deep learning models for semantic segmentation to enable scene understanding for autonomous driving on Indian roads. Worked with Prof. Kavita Sutar.

Chennai Mathematical Institute

Built models to find words similar to a given word and constructed a knowledge graph to check connected components of keywords in COVID-19 research articles, enabling better understanding of research trends.

Indian Institute of Science Education and Research, Pune

Developed semi-Markov models to analyze market microstructure and understand the dynamics of financial markets. Worked with Prof. Anindya Goswami.