NORA-Tech Student Poushali Sengupta Wins Best Oral Presentation at IEEE AAIML 2026
Poushali Sengupta, a PhD student and member of the NORA-Tech Research School Student Council, has been awarded the Best Oral Presentation Award at the IEEE International Conference on Applied Artificial Intelligence and Machine Learning (AAIML 2026), held in Tokyo, Japan. Her paper, "Correlation-Aware Feature Attribution-Based Explainable AI," was recognized for its significant contribution to the field of trustworthy and interpretable artificial intelligence.
Poushali Sengupta presenting at the conference
The recognition underscores the growing importance of trustworthy, human-understandable AI - a theme at the heart of Sengupta’s research:
"This recognition motivates me to continue working towards scalable, reliable, and interpretable AI systems that can be trusted in high-stakes applications."
-Poushali Sengupta
About the Research
As AI systems grow increasingly complex, the ability to explain why a model makes a given decision has become as critical as the decision itself. Existing explainability methods, however, have faced persistent challenges including high computational cost, instability when features are correlated and limited scalability in real-world settings.
Sengupta and her collaborators addressed these shortcomings by proposing ExCIR(Correlation Impact Ratio) - a lightweight, correlation-aware feature attribution framework. Key capabilities of ExCIR include:
Capturing sign-aligned co-movement between features and model outputs
Handling feature correlation effectively via BLOCKCIR to avoid double-counting
Operating in a single pass, making it highly scalable
Reproducing full-model explanations using only 20–40% of data, running 3 to 9 times faster while maintaining strong agreement with complete-dataset results
The method was validated across text, tabular, image, and time-series datasets, demonstrating consistent, stable, and efficient explanations well-suited for real-world deployment.
Further Reading
The full paper is available as an arXiv preprint:
Sengupta, Poushali, et al. "Correlation-Aware Feature Attribution Based Explainable AI." arXiv preprint arXiv:2511.16482 (2025).
Everyone at NORA and the NORA-Tech Research School joins in congratulating Poushali Sengupta on this well-deserved award and the strong contribution her work makes to trustworthy AI.
