ASCRIBE = AI + VR

ASCRIBE has developed innovative platforms for immersive Virtual Reality (VR) and Extended Reality (XR) that revolutionize the exploration of complex multi-scale scientific images by integrating advanced AI algorithms with intuitive 3D visualization. Designed to overcome the limitations of traditional 2D exploration tools, ASCRIBE tools support multimodal image analysis and human-in-the-loop discovery in domains like materials science and biomedical imaging. This research project explores advanced computational methods, ranging from Vision Transformers to Laplacian Mesh Smoothing to deliver a cohesive environment for interactive analysis, segmentation and digital twin construction. To improve algorithm performance, we are building hyperparameter optimization pipelines using Bayesian optimization and Gaussian Processes to fine-tune these models and quantify uncertainty in AI-driven workflows.

ASCRIBE-VR

An immersive VR platform designed in Unreal for fast exploration and measurements of data reduced meshes using the Meta Quest 3/3s [More]

ASCRIBE-XR

An immersive XR platform designed in Godot for advanced exploration of volumes and meshes allied to HPC data analysis using the Meta Quest 3/3s [More]

Recent press

Autonomous Research for Real-World Science

ARROWS Workshop - May 19-21, 2025

Colorado School of Mines

Golden, CO. More

Recent publications

Ushizima et al 2025, "Immersive Geometries and Structures in Motion for Scientific Exploration", SC25 Art of HPC. More

Sordo et al 2025, "Synthetic Scientific Image Generation with VAE, GAN, and Diffusion Model Architectures", Journal of Imaging. More

ASCRIBE VR & XR

Our virtual reality platforms, compatible with Meta Quest, offer mechanisms for scientific data analysis, immersive, interactive tools that enhance data comprehension. We are increasingly integrating results from AI-driven algorithms to deliver high-quality visualization of scientific imagery (e.g. X-ray CT, MRI, electron microscopy) for improving human intuition in materials research, connecting human-in-the-loop with digital twins.

Generative AI

GenAI can tackle data scarcity in research centers reliant on experimental data recorded as pictures. For that reason, ASCRIBE team has partnered with other centers to explore challenging scientific image sets to test our algorithms that combine both scientific text and imaging data for creation of new images with expected properties but that have not been acquired yet.

GP optimizes AI/ML

Bayesian optimization can improve image segmentation results by fusing GP with Bayesian Optimization to efficiently explore and identify the optimal training parameters for segmentation methods and apply the model to high-resolution imaging, e.g. micro-CT. Some AI methods have millions of parameters to be trained, therefore parameter optimization is key in order to maximize metrics such as IoU and other science-domain specific performance indexes.

Report abuse