Mohit Bansal is a computer science professor at the University of North Carolina at Chapel Hill (UNC Chapel Hill). His research focuses on Artificial intelligence, particularly Deep learning and Neural network approaches to natural language processing (NLP). He is known for contributions to multimodal learning, which combines text with images or video, and for improving the robustness of NLP models against adversarial inputs and distribution shifts.
Bansal leads the UNC NLP group, which has produced influential work on vision-language models, commonsense reasoning, and model interpretability. His research often bridges Machine learning theory and practical applications, with publications in top conferences such as ACL, EMNLP, NeurIPS, and ICML.
Academic Career and Education
Bansal completed his Ph.D. in computer science at the University of Illinois at Urbana-Champaign, where he worked under the supervision of Dan Roth. His doctoral research focused on structured prediction and semantic parsing, laying the groundwork for his later work in multimodal and robust NLP. After his Ph.D., he spent time as a postdoctoral researcher at the University of Washington, collaborating with researchers on grounded language understanding.
He joined UNC Chapel Hill as an assistant professor in 2016 and was promoted to associate professor, later becoming a full professor. His lab has received funding from national agencies and industry partners, supporting research on Large language model alignment and evaluation.
Multimodal and Vision-Language Research
A significant portion of Bansal's work addresses the integration of visual and textual information. His team developed models that learn joint representations from images and text, enabling tasks such as visual question answering and image captioning. They introduced methods for aligning visual features with linguistic structures, improving performance on benchmarks like VQA and COCO Captions.
His research also explores how Transformer (architecture) architectures can be adapted for multimodal inputs. By leveraging Multi-Head Attention and Cross-Attention mechanisms, his models effectively reason about relationships between objects in images and words in sentences. This work has implications for applications in autonomous systems and assistive technologies.
Robustness and Adversarial NLP
Bansal has made notable contributions to making NLP models more reliable. His group studies adversarial attacks - small, often imperceptible changes to input text that cause models to make errors. They have developed defense strategies, including Data Augmentation techniques and training procedures that improve Gradient Clipping and regularization.
His work on robust training has shown that models fine-tuned with adversarial examples generalize better to out-of-distribution data. This is critical for deploying Generative AI systems in real-world settings where inputs may be noisy or malicious. His findings have influenced best practices in model evaluation and safety.
Commonsense Reasoning and Interpretability
Another research thread involves teaching models to perform commonsense reasoning. Bansal's team has created datasets and models that require understanding implicit knowledge, such as physical or social conventions. They have used Curriculum Learning to gradually increase task complexity, improving learning efficiency.
He also investigates model interpretability, aiming to explain why neural networks make certain predictions. By analyzing attention patterns and learned representations, his work provides insights into the internal workings of Large language models. This research helps identify biases and errors, contributing to more transparent AI systems.
Impact and Recognition
Bansal's work has been widely cited and recognized in the NLP community. He has served as an area chair and senior area chair for major conferences, and his papers have received best paper awards and nominations. His research has been featured in industry collaborations, and he has given invited talks at academic and industrial venues.
His contributions have helped shape the direction of multimodal and robust NLP, influencing both academic research and practical deployments. As of 2025, he continues to lead an active research group at UNC Chapel Hill, mentoring students who have gone on to positions in academia and industry.
Selected Publications
Bansal has authored over 100 peer-reviewed papers. Notable works include studies on adversarial training for text classification, vision-language pretraining, and robust question answering. His collaborative papers with students and colleagues have advanced the understanding of how to build AI systems that are both capable and trustworthy.
His research has been supported by grants from the National Science Foundation and the Defense Advanced Research Projects Agency, among others. He is a member of the Association for Computational Linguistics and the Institute of Electrical and Electronics Engineers.
External Links and Further Reading
For more information, readers can consult his university faculty page and his Google Scholar profile, which list publications and ongoing projects. His lab's website provides resources and open-source code for reproducing results.
References
This article is based on publicly available information about Mohit Bansal's career and research. Specific citation details are omitted to maintain brevity, but interested readers can find primary sources through academic databases and his official university profile.