
Tasfia Nuzhat
Research
I am a Ph.D. student in Industrial Engineering & Management Systems at the University of Central Florida, where I work in the Complex Adaptive Systems Lab under the supervision of Dr. Niloofar Yousefi. My research develops multi-objective optimization methods for generative diffusion models, with a broader interest in trustworthy and controllable generative AI.
I am also expanding my work into AI safety, reasoning, and alignment for tool-using LLM agents. As part of this effort, I collaborate on NEXUS, a runtime safety framework that monitors agent behavior and intervenes before potentially unsafe actions are executed. Before beginning my Ph.D., I spent two years as an AI engineer developing LLM-based retrieval, information-extraction, and scientific knowledge systems for pharmaceutical research.
- Multi-Objective Optimization & Generative Diffusion
Constrained molecule generation with motif-aware tokenization and role-aware corruption (MoCDiff, MotifRole-Diff). - AI Safety, Reasoning & Alignment
Collaborative work on runtime monitoring and intervention for tool-using LLM agents; safety under multi-agent composition (NEXUS, ChannelGuard, CAREBENCH). - Trustworthy & Biomedical AI
Temporal-conflict resolution in language models and biomedical document classification under limited data (Parametric Temporal Conflict, R-GAT).
Experience
Education
Publications
Full list on Google Scholar.
-
MoCDiff: Efficient Motif-Constrained Discrete Diffusion for Molecule Generation
-
MotifRole-Diff: Risk-Optimal Role-Aware Corruption for Masked Molecular Graph Diffusion
-
A Review on Machine Learning in Photocatalytic Degradation of Organic Dyes from Wastewater: Current Trends and Future Directions
-
A Multi-Scale Hyper Representation Network for Semantic Segmentation of Pulmonary Circulation Vessels in Pathology Assessment
-
Artificial Magnetic Conductor Unit Cell Design Using Machine Learning Algorithms
-
NEXUS: Structured Runtime Safety for Tool-Using LLM Agents
-
R-GAT: Cancer Document Classification Leveraging a Graph-Based Residual Network for Scenarios with Limited Data
-
Microwave Imaging: CPW-Fed Microstrip Patch Antenna with Defected Ground Plane Operating in ISM Band for Breast Tumor Detection
-
Right Knowledge, Wrong Answer: Characterizing Parametric Temporal Conflict in Open-Weight Language Models
-
ChannelGuard: Safe Models Do Not Compose into Safe Multi-Agent Systems
-
CAREBENCH: When Correct Diagnoses Come from Unsafe Reasoning
-
PEFT A2Z: Parameter-Efficient Fine-Tuning Survey for Large Language and Vision Models
Skills
Recent updates
Peer Review
- Reviewer, ICML 2026 Workshop on Generative & Agentic AI for Biology (GenBio)
- Reviewer, NeurIPS 2026 Women in Machine Learning Workshop