Veeraraju ElluruI am a Computer Science Senior at the Indian Institute of Technology, Jodhpur. I love AI research and Math. My main areas of interest are Generative Modeling for Computer Vision, Scene Understanding, and Multimodal Mechanistic Interpretability. When it comes to research, I try to do my part to ensure multimodal systems respect and preserve privacy. I am extremely grateful to be jointly mentored by Dr. Shivang Agarwal and Dr. Mayank Vatsa at the Image Analytics and Biometrics (IAB) Lab, IIT Jodhpur. Throughout my undergraduate studies, I actively work on side-quests where AI research can help siginificantly contribute. This has given me the privilege to work at Thoughtworks AI Labs (TAILS), under the mentorshop of Shayan Mohanty and team, on Fine-grained Incompleteness Evaluation of SLM-generated Summaries; under Dr. Yinglun Zhu, UC Riverside, on mutlimodal mechansitic interpretability; with Dr. Tiago Bresolin, at the Center for Digital Agriculture (UIUC), on Foundation Models for Livestock Image Segmentation; and under Dr. Deepak Mishra at the MAISys research group, IIT Jodhpur, on Deep Learning-based simulation of High Energy Physics with CERN. I'm a strong believer of Bruce Lee's 10000-hour to mastery rule; and am forever grateful to all my mentors who have shaped my research. |
|
News
|
Publications |
|
Bias-Aware Machine Unlearning: Towards Fairer Vision Models via Controllable ForgettingSai Siddhartha Chary Aylapuram, Veeraraju Elluru, Shivang Agarwal paper / code / program page Deep neural networks often rely on spurious correlations in training data, resulting in biased or unfair predictions, particularly in safety-critical applications. While conventional bias mitigation methods typically require retraining from scratch or redesigning the data pipeline, recent advances in machine unlearning offer a promising alternative for post-hoc model correction. In this work, we explore the trifecta of efficiency, fairness, and model utility post unlearning via Bias-Aware Machine Unlearning, a paradigm that selectively forgets biased samples or feature representations to address various forms of bias in vision models. |
|
|