SOCR ≫

DSPA ≫

DSPA2 Topics ≫

DSPA2: Data Science and Predictive Analytics (UMich HS650)

Appendix 11: Foundational Generative Artificial Intelligence Models (GAIMs)

Expanding on DSPA Chapter 14 (Deep Learning) and the GAIM GitHub website, this DSPA2 appendix demonstrates the utilization (application) of pre-trained large language models, such as foundational generative artificial intelligence models (GAIMs). Specifically, we show various SOCR GAIMs and use LM Studio, an integrated development environment for GAIM applications, that parallels what to RStudio does for computational data modeling and analytics but for generative AI.

This DSPA GAIM tutorial demonstrates AI-augmented human intelligence and provides specific examples of four complementary applications of AI decision-support including:

• Theoretical research and abstract scientific discovery,
• Applied and exploratory scientific inquiry,
• Interdisciplinary studies, and
• Health science scholarship and clinical decision-support.

1 SOCR GAIM Applications

The SOCR HTML5 Apps website includes a number of interactive data science, computational modeling, statistical analysis, and ML/AI applications. Several of these apps demonstrate GAIM implementations and foudational AI services, e.g.,

• The SOCR AI Bot provides access to SOCR data analytics, synthetic text generation, image and code generation, and R-training)
• The SOCR GAIM 2D/3D Synthetic Brain Image/Volume Generation App is used for simulating realistic normal and pathological brain scans/images (or hypervolumes) based on user specified participant profiles
• The SOCR Dimensionality Reduction app includes functionality fo r linear and nonlinear projections of high dimensional data into 3D and 3D spaces
• The Interactive Virtual Hospital/Simulated Patient Data (VH/SimPat) App is used for simulation of realistic patient profiles as digital twins of real electronic medical records
• The Interactive Pressure Injury Prediction Model (PIPM) App demonstrated the use of ML/AI for on-demand prediction and quantification of risks for developing a hospital-acquired condition - bed-ulcers (skin pressure injuries)
• The SOCR Analytical Toolkit (SOCRAT) is a scalable web platform for in-browser applications for interactive data analysis and visualization
• The SOCR 2D/3D Hierarchical Clustering and Gaussian Mixture app Modeling and the General Clustering App offer mechanisms for unsupervised segmentation, clustering, and tessellation.

2 LM Studio