Papers

2025

A Hybrid Random Forest and CNN Framework for Tile-Wise Oil-Water Classification in Hyperspectral Images

Authors: S.Mohammad S.Ahmadi, Mehdi Nickzamir
Abstract: This paper introduces a hybrid framework that improves oil-water classification in hyperspectral images by training a Convolutional Neural Network on Random Forest probability maps to integrate spatial context with pixel-wise spectral data.
PDF | arXiv

---

Extending Chronos: Fine-Tuning for Financial Forecasting and Consistency Calibration

Authors: S.Mohammad S.Ahmadi, Mehdi Nickzamir, Erfan Bayat, Andrea Vasco Grieco
Abstract: This work enhances the Chronos transformer model for financial forecasting by integrating domain-specific fine-tuning on stock and synthetic data with a perturbation-based consistency calibration (C3) method, resulting in a 15% increase in accuracy and more reliable probability estimates.
PDF | Code

---

2024

Natural Language Queries in Egocentric Videos

Authors: S.Mohammad S.Ahmadi, Yasaman Noshirvanbaboli, Amirhossein Lotf Ranaei
Abstract: This project enhances natural language query (NLQ) performance in egocentric vision by identifying the optimal combination of EgoVLP features and the VSLNet model, while further extending the framework with a fine-tuned Gemma 2B model for automated query generation.
PDF | Code