Dr. Brian Jalaian
- Position: Associate Professor; Director, Artificial Super Intelligence Lab
- Department: Computer Science; Intelligent Systems and Robotics
- Office Location: Campus: Building 4, Room 434;
- bjalaian@uwf.edu
- Campus: 850.474.2963
- Personal UWF Website
- Google Scholar
Biography
Dr. Brian Jalaian is the Director of the Artificial Super Intelligence (ASI) Lab at the University of West Florida, where he is an Associate Professor of Computer Science and Intelligent Systems & Robotics. The ASI Lab's Ph.D. students work on the reliability of agentic AI, model compression, and uncertainty quantification.
His research covers deep learning, large language and vision-language models, and uncertainty quantification. Much of it is on trustworthy AI: Bayesian deep learning, calibrated confidence, and robustness to adversarial and natural perturbations.
More recently, he studies how to evaluate AI agents honestly, and what post-training and compression change inside them. Many reported gains in AI are smaller than the run-to-run noise. Therefore, his group builds methods that tell real improvements from chance, and falsification methods that search simulated environments for the conditions where an autonomous vision system fails. His group also develops multi-agent systems on vision-language models. These systems reason over images and video, detect hallucination, estimate their own uncertainty, and abstain when unsure. He is building a research program on how post-training, distillation and compression change the safety and security of such agents, including whether safety monitors and interpretability probes keep working after a model changes.
Earlier, Dr. Jalaian was a senior AI research scientist at the DEVCOM Army Research Laboratory. There he led a research area on the uncertainty, robustness and assurance of AI in a multi-institution Army research alliance. He then served at the DoD Joint Artificial Intelligence Center, now part of the Chief Digital and Artificial Intelligence Office (CDAO). There he developed the recommended test and evaluation strategy for the program that became JATIC. His work has appeared at NeurIPS, UAI, ECML PKDD, the AAAI Symposium Series and IEEE venues. He holds a Ph.D. and M.S. in Electrical Engineering and an M.S. in Industrial and Systems Engineering from Virginia Tech, where he also served as adjunct faculty. He is a Senior Member of the IEEE.
Degrees & Institutions
Ph.D. Electrical Engineering, Virginia Tech
M.S. Industrial Systems Engineering, Virginia Tech
M.S. Electrical Engineering, Communication & Network Systems, Virginia Tech
Research
Broad interests: deep learning, large language and vision-language models, uncertainty quantification, and trustworthy AI. Current focus:
- Visual and Multi-Agent AI: Agents built on vision-language models that reason over images and video with tools. They estimate uncertainty across agents, detect hallucination and disagreement, and abstain or escalate when unsure.
- Evaluation and Measurement Science: Statistically valid evaluation of models and agents: error bars, noise floors, and decision reliability, so that real gains are separated from chance. Simulation-based falsification that searches for the conditions where autonomous perception fails.
- Post-Training, Compression and Safety: How fine-tuning, distillation and quantization change the safety and security behavior of agents: robustness to prompt injection and adversarial attacks on tool-using agents, and whether safety monitors and interpretability probes survive these changes. Small, efficient models with measured trade-offs between capability, cost and safety.
Learn more through the following platforms: