MrsElla Zarandi

AI in Healthcare and Biomedical Applications: Application of large language models (LLMs) and machine learning techniques to healthcare and biomedical domains, with emphasis on clinical text mining, diagnostic support, and biomedical knowledge discovery, while addressing challenges such as hallucination detection and reliability.

Large Language Model (LLM): Development, benchmarking, and evaluation of diverse LLM families, including:

OpenAI (GPT-3.5-Turbo, GPT-4o), Google DeepMind (Gemini Pro 1.0, Gemini-1.5-Flash-8B, Gemini-2.0-Flash-Exp), Meta (LLaMA-2-7B/ 13B), Mistral AI (Mistral 7B, Mistral 8x7B), TII ( Falcon-7B), Google/FLAN (Flan-T5 small/Large/ XL), Transformers (BERT family) BERT, BioBERT, SciBERT, 

Conduct comparative evaluation, domain adaptation, and performance benchmarking across semantic, lexical, statistical, and information coverage metrics.

Fine-Tuning and Model Optimisation: Investigation of advanced parameter-efficient fine-tuning methods, including LoRA variants (LoRA-Small, LoRA-Alpha, LoRA-Beta) and related optimisation techniques for efficient deployment of LLMs across specialised domains.

Hallucination Detection and Mitigation: Research into frameworks for identifying and reducing hallucinations in LLM outputs, using methods such as semantic alignment, lexical and statistical analysis, and information coverage metrics, with a particular focus on high-stakes domains like healthcare.

Synthetic Data Generation: Development and benchmarking of synthetic data generation methods for tabular and biomedical applications using GAN-based approaches (CTAB-GAN, CTAB-GAN+, TableGAN, CTGAN, CT-GAN, MEDGAN, GANBLR++, GNBLAR), transformer-based approaches (TabPFN, TabPFNGen), and diffusion-based approaches (TABDDM). Research interests include assessing synthetic data quality with respect to statistical fidelity, privacy preservation, fairness, and downstream machine learning performance.