初心
I am primarily interested in Natural Language Processing with a focus on Reasoning, Alignment and AI-Safety. I obtained my Masters in Computer Science from the Georgia Institute of Technology and have industry experience as an Applied Scientist at Amazon. I have previously worked on topics such as Semantic Parsing, Cross-Lingual Transfer Learning, addressing bias and fairness in Large Language Models (LLMs) and more recently, Red-Teaming LLMs. I was a co-organizer for the TrustNLP workshop at NAACL 2022 and have served as a committer to the Apache Software Foundation in the past.
During undergrad, I participated in the Google Summer of Code program, where I worked on PhyloGeoRef, an open-source project for visualizing Phylogenetic trees in Google Earth. After the program ended, I started contributing to the Apache Hama project which was an open source implementation of the Google Pregel paper. This experience helped me build a solid foundation in concurrency and synchronization primitives in distributed systems.
As I entered engineering, I was fortunate to work with amazing peers at Bloomreach who introduced me to technologies such as Staged Event-Driven Architecture (SEDA), Map-Reduce, NoSQL Databases, Data Warehousing, Search and Information Retrieval. This foundational software-engineering knowledge has been invaluable in transforming my research into practical production systems.
In 2013, I learned about the ImageNet Challenge and the advancements made by Geoffrey Hinton's group. To get a better understanding of deep learning and machine learning theory, I decided to apply for graduate school. I subsequently enrolled at Georgia Tech in 2015 after taking some time to save up funds, explore my interests, research the graduate application process and improve my english for GRE and TOEFL.
During my time at Georgia Tech, I had the opportunity to take the first deep-learning course offered on campus, where I built a system to detect skin cancer from skin lesions. Back in 2016, as deep learning was just starting out, this system was already delivering an impressive accuracy of 89%, which was comparable to that of a human expert. The models were trained using the Torch Lua framework. I also collaborated on two research papers with Professor Le Song's group. Our research focused on detecting changes in dynamic events over networks (Li et al., 2017) and guiding information diffusion in online networks (Wang et al., 2018). Other projects I worked on during this time included Multi-Hop Question Answering using Key-Value Memory Networks, and Epileptic Seizure Detection through LSTMs.
After graduating, I started working at Amazon Boston as an Applied Scientist on the Alexa team. During 2017-2018, I focused on developing core pipelines to facilitate scaling out Alexa NLU model training. From 2018 to 2020, I worked on creating a successful feature for hierarchical parsing of long-form Alexa utterances. However, due to visa issues, I had to move back to India in 2020. I was able to continue working for Amazon Alexa and worked on building Multilingual models for Hindi utterances. 2021 was a tough year with the pandemic in India. I returned to Amazon Boston in 2022 to join the Alexa Trustworthy AI team where I contributed to multiple initiatives focused on evaluating and mitigating bias and fairness issues in Large Language Models (LLMs) (Soltan et al., 2022), (Gupta et al., 2022), (Krishna et al., 2022).
In 2023, I joined a promising startup after the large-scale layoffs at Alexa. As a technology worker, I was lucky to retain my position throughout the pandemic despite facing visa challenges. I now empathize with those who suffered job losses due to the pandemic. Even though I was saddened by the news, I had a strong belief in the Alexa vision. I am viewing this as a chance to focus on my health and delve into new fields within machine learning.
As of 2024, I joined Bloomberg Law as a Senior Machine Learning Engineer and have also enrolled in a doctoral program at NJIT. I am quite happy with the work-life balance at Bloomberg and their focus on philanthropy. One of my recent achievements is writing a comprehensive paper on Red-Teaming Attacks against LLMs. This paper provides an extensive review of various attack methods and defense mechanisms, as well as strategies for successful red-teaming (Verma et al., 2024).
My present research focuses on, (1) Model Alignment, (2) AI Content Detection, (3) Reasoning and Hallucination, and (5) AI Safety. If you are interested in working with me on a research problem, please send me an email including a brief biography and the topic and question you would like to investigate together.
- [1]AlexaTM 20B: Few-Shot Learning Using a Large-Scale Multilingual Seq2Seq Model.In ArXiv, vol. abs/2208.01448, 2022.
- [2]Mitigating Gender Bias in Distilled Language Models via Counterfactual Role Reversal.In Findings of the Association for Computational Linguistics: ACL 2022, Dublin, Ireland, pp. 658–678, 2022.
- [3]Measuring Fairness of Text Classifiers via Prediction Sensitivity.In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Dublin, Ireland, pp. 5830–5842, 2022.
- [4]Detecting Changes in Dynamic Events Over Networks.In IEEE Transactions on Signal and Information Processing over Networks2017.
- [5]A Stochastic Differential Equation Framework for Guiding Online User Activities in Closed Loop.In Proceedings of the Twenty-First International Conference on Artificial Intelligence and Statisticsvol. 84, , pp. 1077–1086, 2018.
- [6]Operationalizing a Threat Model for Red-Teaming Large Language Models (LLMs).In arXiv preprint arXiv:2407.14937, 2024.