Visual Intelligence Systems for Trustworthy AI
VISTA Lab develops AI that sees broadly, reasons deeply, and acts reliably
Visual Intelligence Systems for Trustworthy AI
The name VISTA evokes a broad and inspiring view of the world. It naturally connects to vision, perception, and deeper visual understanding. Inspired by the landscapes of Logan Canyon, VISTA Lab develops trustworthy visual and multimodal AI systems that can perceive, reason, and generalize across complex real-world environments.
Our research spans foundation models for perception and reasoning, biometrics, anomaly detection, digital forensics, healthcare AI, and autonomous driving. We are committed to advancing AI that is not only powerful, but also robust, interpretable, and aligned with human values.
VISTA Lab is part of the School of Computing at Utah State University.
VISTA Lab officially launches at Utah State University. We are actively recruiting motivated Ph.D. and Master's students!
Principal Investigator
Graduate Students
Ph.D. Student
Ph.D. Student
M.S. Student
Undergraduate Students
Undergraduate Researcher
VISTA Lab pursues research at the intersection of computer vision, multimodal AI, and trustworthy systems. Our goal is to build AI that is not only powerful, but also robust, interpretable, and safe for real-world deployment. Our work is organized around three complementary directions.
Developing foundation models that support robust perception, adaptive reasoning, memory, uncertainty, and continual learning.
This direction develops VLMs and MLLMs for robust perception, efficient adaptation, and trustworthy deployment, including multimodal in-context learning, prompt learning, test-time adaptation, and evidence-grounded and uncertainty-aware reasoning. Representative works include Hyper-ICL, DiSa, and Style-Pro.
Future directions include memory-augmented VLMs for long-horizon reasoning, physical-space reasoning, vision-language-action foundation models, closed-loop multimodal systems, test-time multimodal adaptation, and continual multimodal learning.
Building large-scale intelligent systems that are robust, efficient, and interpretable for real-world deployment.
This direction develops machine learning and multimodal AI techniques to safeguard identity and digital trust, advance long-range biometric recognition, and address multimodal safety, hallucination reduction, fairness-aware and privacy-preserving learning, and adversarial robustness.
Applying AI innovations to solve real-world challenges in healthcare, advanced manufacturing, autonomous driving, and beyond.
This direction leverages machine learning and multimodal AI to address critical challenges in healthcare, including MedViT for robust medical imaging; advanced manufacturing, including ROADS and PromptMAD for industrial anomaly detection and quality control; and autonomous driving, including robust perception, tracking, multi-teacher knowledge distillation, and risk-aware driving policies.
VISTA Lab is actively recruiting motivated Ph.D., Master's, and undergraduate students passionate about computer vision, multimodal AI, and trustworthy machine learning.
If you are interested in joining, please read Dr. Kashiani's homepage and send an email with your CV, transcripts, and a brief statement describing your research interests and background.
hossein.kashiani@usu.edu
