Publications

Mahan Veisi, Kian Anvari, Shahabedin Nabavi, Mohsen Ebrahimi Moghaddam

Developed a cascaded MRI reconstruction model utilizing Frequency Mining with state‑of‑the‑art transformer‑based U‑Net models for k‑space data. This approach efficiently handles multi‑contrast MRI and is targeted for the MICCAI 2025 reconstruction challenge.

Mahan Veisi, Sadra Berangi, Mahdi Shahbazi Khojasteh , Armin Salimi‑Badr

Integrated FinBERT within a PPO‑LSTM model, achieving 134.39% cumulative return and a 1.46 Sharpe ratio, outperforming standard benchmarks.

Armin Salimi-Badr, Mahan Veisi, Sadra Berangi

Applied SincNet layers to classify Parkinson’s Disease from gait data, achieving 98.7% accuracy. After training, filters were pruned using K‑means clustering and silhouette scores to highlight key frequency bands significant in disease detection. This study enhances explainable AI by identifying critical signal components in Parkinson’s diagnosis.

Mohammad Hesam Nasiri, Seyed Hassan Majid Zonouzi, Arya Parvizi, Seyed Mostafa Atyabi, Seyedeh Rana Rokni, Sanaz Moosapour, Mahan Veisi, Kiarah Kowsari, Farbod Saghfi

Developed intelligent soccer‑playing agents optimized for roles like goalkeeping and defense in the RoboCup 2D Soccer Simulation League. The team secured second place at the 2023 RoboCup IranOpen.