Hi, I am Aditya. I am a fourth-year B.Tech student in the Department of Computer Science and Engineering at the Indian Institute of Technology (IIT) Kharagpur.

My current research interests include Natural Language Processing and Machine Learning. In particular, I want to explore the ethical aspects of Artificial Intelligence models and ensure that they are unbiased, adversarially robust and trustworthy. Throughout my undergraduate years, I’ve worked on a lot of projects in the areas of Machine Learning, Natural Language Processing, Information Retrieval, Operating Systems, Databases and Web Development. Currently I am working on my Bachelor’s Thesis Project to deploy adversarial attacks on API-based NLP models using input-agnostic and task-agnostic input perturbations.

Experience

In addition to course projects, I have also been actively involved in various research projects at IIT Kharagpur. I have worked on a year-long research project under the guidance of Prof. Animesh Mukherjee. This project explored how the gender biases encoded in the Wikidata Knowledge Graph propagate to the downstream link prediction task. For my Bachelor’s Thesis, I am advised by Prof. Somak Aditya and Prof. Mainack Mondal. In this project, I am developing a framework to learn input-agnostic (universal) perturbations, called triggers, that can significantly degrade the performance of an instruction fine-tuned language model across diverse tasks and languages. Furthermore, an important constraint is that the attacker has only black-box API access to the model being attacked with zero knowledge about the model’s internals.

Contact

Please feel free to email me at: adityasoni9998@gmail.com.