Project Scientist — IIT Delhi
Hi — I’m Jayesh.
Distribution Shifts and AI/ML Model Robustness.
I build methods to detect, diagnose, and adapt models when data changes.
About
I am a Project Scientist at IIT Delhi working on distribution-shift detection, robust machine learning for remote sensing, and reproducible evaluation pipelines in applied fields.
Selected Projects
-
Cybersecurity — Explainable Drift Detection & Early Retrain Paper (DIMVA 2025)
Methodologies to detect concept drift in malware/security pipelines, provide human-interpretable explanations for drift sources, and trigger targeted retraining to restore performance. Designed experiments on Android malware corpora (DREBIN, AndroZoo) with reproducible evaluation.
Key outputs: DIMVA 2025 paper describing drift-detection + explanation pipeline and experiments showing improved sustained model performance.
-
Disaster Robotics — Damage Assessment Paper (FAccT 2025)
End-to-end pipelines for post-disaster imagery: assembled and annotated a NADIR satellite building-damage dataset (~10k buildings), ran community annotation campaigns, and published the dataset to accelerate research on building damage assessment.
Key outputs: ACM FAccT 2025 dataset publication, Hugging Face dataset release, Labelbox annotation workflow and community workshops.
-
Climate Tech — Forest Cover Modelling
Temporal forest-cover and biomass estimation pipelines for carbon-credit estimation. Built satellite-based temporal models and validated them with field surveys across multiple village sites (Dhenkanal, Odisha). Working on integration into operational monitoring stacks.
Publications
- Jayesh Tripathi, Heitor Gomes, and Marcus Botacin, “Towards Explainable Drift Detection and Early Retrain in ML-Based Malware Detection Pipelines”, DIMVA 2025.
- Thomas Manzini, Priyankari Perali, Jayesh Tripathi, and Robin R. Murphy, “Now you see it, Now you don’t: Damage Label Agreement in Drone & Satellite Post-Disaster Imagery”, FAccT 2025.
Contact
Email: jt (dot) jayesh98 (at) gmail (dot) com