From the India Conference at Harvard Hackathon to PReMI 2025
Highlighting research that originated at the India Conference at Harvard Policy Hackathon and was later published and presented at the International Conference on Pattern Recognition and Machine Intelligence (PReMI).
Aryan Nath, a final-year Computer Science major and Mathematics minor at Ashoka University, recently attended the International Conference on Pattern Recognition and Machine Intelligence (PReMI 2025) to present his research on climate vulnerability assessment using explainable deep learning. Held at IIT Delhi, the conference served as a professional forum for researchers and practitioners working in pattern recognition, machine learning, and artificial intelligence. His participation provided an opportunity to engage with the international research community, present the team’s findings, and discuss the policy relevance of interpretable machine learning approaches for climate risk analysis.
From the India Conference at Harvard Hackathon
A student team comprising Soumyajit Basu, Fiona Arora, Aryan Nath, and Karan Kumar, mentored by Prof. Lipika Dey, was shortlisted as the the only student group selected to present in person at Harvard University in the data science track of the Policy Hackathon, part of the India Conference at Harvard (ICH).
Among more than a thousand participating teams - including policy professionals and domain experts - the project stood out for its ability to translate complex climate and socio-economic datasets into meaningful analytical insights.
The research focused on identifying vulnerable communities across India and proposing targeted, data-driven adaptation strategies. By combining representation learning with explainability methods, the project demonstrated how machine learning can support climate-resilient policy development.
With support from the Mphasis AI & Applied Tech Lab, Fiona Arora represented the team at Harvard and presented the work before a distinguished panel of judges.
The Research: Assessing Climate Vulnerability Using Explainable Deep Clustering
The work was later expanded into a research paper titled:
“An Explainable Deep-Clustering Approach to Assess Climate Vulnerability in India.”
The approach combined:
- Latent space representation using a β variational auto-encoder
- Block-level clustering for fine-grained vulnerability assessment
- SHAP-based interpretability to identify key contributing factors
- Policy-oriented visualization and analysis
The framework addressed limitations of traditional vulnerability indices by providing localized insights and interpretable results for policymakers.
Presentation at PReMI 2025
The refined research was accepted at the International Conference on Pattern Recognition and Machine Intelligence (PReMI 2025), where Aryan Nath presented the work.
The conference brought together researchers working in applied machine learning, pattern recognition, health analytics, and artificial intelligence.
Presenting the work enabled discussions with faculty and researchers on emerging topics such as quantum artificial intelligence, soft computing, and swarm intelligence.
Poster Presentation and Technical Discussions
In addition to the oral session, the research was also presented in poster format. This created opportunities for detailed technical discussions on:
- Model interpretability
- Evaluation of vulnerability indices
- Policy applications of explainable AI
Acknowledgments
Aryan thanks his collaborators - Soumyajit Basu, Fiona Arora, and Karan Kumar - and advisor Prof. Lipika Dey for their collaboration and support.
He also acknowledges:
- India Conference at Harvard (ICH)
- International Conference on Pattern Recognition and Machine Intelligence (PReMI 2025)
- Mphasis AI & Applied Tech Lab