Publications

2026

1. IEEE JBHI

   

   DiabLLM: An LLM-Based Framework for Blood Glucose Prediction in Type 1 Diabetes

   Amirhossein Mahmoudi, Ghazal Farahani, Peter Domanski, and 3 more authors

   IEEE Journal of Biomedical and Health Informatics, Aug 2026

   Abs DOI Bib Code

   Accurate blood glucose prediction is essential for enabling glycemic control in individuals with Type 1 Diabetes Mellitus, particularly within smart and connected health systems that integrate continuous glucose monitoring and automated insulin delivery. DiabLLM uses the Time-LLM and Chronos architectures for 30- and 45-minute forecasting. Evaluations on the OhioT1DM and D1NAMO datasets show improvements over established baselines, robustness to noisy and missing inputs through a denoising autoencoder, and substantial compression through knowledge distillation for deployment on resource-constrained edge devices.

   @article{mahmoudi2026diabllm,
     title = {{DiabLLM}: An {LLM}-Based Framework for Blood Glucose Prediction in Type 1 Diabetes},
     author = {Mahmoudi, Amirhossein and Farahani, Ghazal and Domanski, Peter and Farahani, Bahar J. and Firouzi, Farshad and Chakrabarty, Krishnendu},
     journal = {IEEE Journal of Biomedical and Health Informatics},
     year = {2026},
     month = aug,
     volume = {30},
     number = {8},
     pages = {6446--6459},
     doi = {10.1109/JBHI.2026.3658588},
   }

2. IEEE COINS

   

   Fair Teacher, Fair Student? Fairness Transfer in Clinical LLMs

   Amirhossein Mahmoudi and Bahar Farahani

   In 2026 IEEE International Conference on Omni-layer Intelligent Systems (COINS), Sep 2026

   Abs Bib Code Website

   Knowledge distillation is widely used to compress large models into compact students, but its effect on fairness during teacher–student transfer remains insufficiently characterized for large-language-model-based clinical time-series models. This study introduces Exposure-Balanced Teacher Distillation and Group-Conditional Affine Output Adaptation, evaluating their utility–fairness trade-offs for blood-glucose forecasting on OhioT1DM and ectopic-beat classification on the MIT-BIH Arrhythmia Database. The results show that fairness effects under knowledge distillation are task dependent and that a smaller fairness gap does not necessarily indicate improved group-specific performance.

   @inproceedings{mahmoudi2026fairteacher,
     title = {Fair Teacher, Fair Student? Fairness Transfer in Clinical {LLM}s},
     author = {Mahmoudi, Amirhossein and Farahani, Bahar},
     booktitle = {2026 IEEE International Conference on Omni-layer Intelligent Systems (COINS)},
     year = {2026},
     month = sep,
     address = {Bologna, Italy},
   }