AI Demos by industry

Research & scientific AI AI Demos

Develop and test reproducible methods for model diagnostics, information theory, and scientific learning. Explore matrix and tensor methods and quantum-assurance evidence across disciplines. Browse interactive demonstrations relevant to this area, with methods, screenshots, and resource links.

Browse by industry

Choose an industry to explore relevant demonstrations. Research & scientific AI brings together methods used across industries.

9 of 9 entries

Browse entries by name

EnviTrace’s U.S. Geothermal Temperature & Stress Model

EnviTrace has recently developed a comprehensive U.S. Geothermal Temperature & Stress Model. This demo presents the results obtained. It supports regional geothermal screening by bringing continental temperature and stress model information into a common map. Users can compare available model layers, inspect geographic variation, and relate predicted conditions to the underlying inputs. The shared view helps researchers identify areas and model differences that merit closer examination during geothermal resource studies.

Maps geothermal inputs and model predictions across the continental United States.
Compares EnviTrace thermal and stress models with the published Stanford Thermal Model.
Lets users switch input and output layers while inspecting the same geographic area.
Provides regional screening and comparison workflow using existing model layers, rather than a new site-specific simulation.
Research papers summarizing the methods and results are in preparation.

Geysers Seismicity

The Geysers workflow presents seismic activity within a major operating geothermal field, connecting earthquake locations with their spatial organization over a decade of observations. We have performed various clustering and spatiotemporal ML analyses. We present 3D views of our closest server to make it easier to examine where events concentrate and how different groups occupy the field. The demo provides an interactive starting point for discussing induced seismicity and selecting patterns for further scientific investigation.

Explores earthquake observations from the Geysers geothermal field over 2010–2019.
Groups events into prepared spatial clusters to reveal differences in event density and location.
Supports pan, zoom, hover, and comparison of two clustering views.
Uses prepared observational visualizations; browsing does not fit a new seismic model or establish a causal explanation.
A recent research paper titled “Deep-Embedded-Clustering of Microseismicity Identifies Multiple Failure Mechanisms at The Geysers Geothermal Field” summarizes preliminary ML results.

Model Diagnostics

A close match to observed data leaves important questions about a model's reliability and the information contained in its parameters. This demo combines calibration with held-back checking, sensitivity analysis, and an examination of whether individual inputs can be constrained. The resulting views help users distinguish influential parameters from weakly supported ones and identify where additional evidence could improve an analysis.

Combines model calibration with a held-back data check.
Examines which inputs influence predictions and which parameters are poorly constrained.
Includes an optional unused input to illustrate an identifiability warning.
Encourages joint review of fit, validation, sensitivity, and identifiability rather than judging a model by fit alone.
Applies our open-source code Mads.jl

Data Imputation

Incomplete geologic (geospatial) measurements can make regional patterns difficult to interpret and limit the comparisons available to a modeling team. This demo organizes the reconstruction of missing Great Basin data and the evaluation of spatial predictions within a common analysis workflow. Its comparison tools help users examine how selected methods perform when observations are deliberately withheld for checking.

Examines incomplete Great Basin geothermal measurements and their spatial reconstruction.
Organizes dataset preprocessing, held-back splits, comparison methods, and evaluation metrics in one workflow.
Supports regional prediction grids and comparison of reconstructed values with observations withheld from fitting.

Information Theory Analyses

Scientific datasets often contain more detail than a particular analysis needs, yet excessive simplification can remove useful structure. The Information Theory Explorer lets users examine this balance through synthetic measurements arranged along a line or across a surface. Visual comparisons connect the amount of retained information with recognizable patterns, making resolution choices easier to understand and discuss.

Explores how scientific patterns appear in data along a line and across a surface.
Varies the character of synthetic measurements and the goal of the information summary.
Uses TensIT comparisons to balance useful retained detail against grid size.
Makes resolution choices inspectable through visual comparisons of the original evidence and its simplified representation.

Tensor Information (TensIT)

TensIT addresses the practical question of how much spatial and temporal detail a scientific dataset should retain. TensIT tool provides capabilities to compare candidate tensor grids against an information target and a computational budget while accounting for structure in the data. Its recommendations and supporting comparisons help users select a manageable representation and review the evidence behind that choice.

Compares candidate tensor resolutions for a reproducible scientific-data scenario.
Balances an information-retention target and cell budget with spatial and temporal structure.
Reports a recommended grid together with retained information and computational-cost comparisons.
Keeps the resolution-selection evidence and saved run history available for review.

Feature Extraction (temporal)

A measured time series may combine several underlying processes whose individual contributions are difficult to distinguish. NMFk is applied here to show how unknown signatures can be extracted from mixed synthetic observations. Comparing the recovered signals with the original mixtures provides a practical introduction to interpretable feature extraction and the choice of decomposition size.

Separates mixed synthetic time-series observations into interpretable latent temporal signals.
Uses NMFk to compare candidate signal counts rather than assuming a single decomposition size.
Allows a computed run or an explicitly precomputed example, with progress and cancellation for live execution.
Supports comparison of recovered components with the original observations for a chosen value of k.

Feature Extraction (spatial)

Large geoscience datasets can contain recurring combinations of attributes that vary across a region. This demo uses NMFk to identify latent spatial factors and groups from selected measurements, with configurable preprocessing and factorization settings. Maps, biplots, and reconstruction views help users connect the extracted mathematical structure with the locations and attributes being studied.

Extracts latent spatial factors and clusters from selected dataset attributes with NMFk.
Configures the factor-count range, restart count, and normalization before factorization.
Provides matrices, latent maps, biplots, and reconstruction evidence to interpret recovered structure.
Supports optional regional interpolation after results exist; a factorization does not by itself establish a geological cause.

Q-Sure Quantum ML Assurance

Q-Sure explores how evidence can guide the use of quantum machine-learning methods when uncertainty and resource limits matter. The demo replays seeded offline-emulator records and presents assurance, prediction, and audit comparisons within a shared evidence workflow. Users can examine how available information supports accepting a result, allocating additional effort, or abstaining when the evidence is insufficient.

Replays seeded offline-emulator evidence for quantum machine-learning assurance.
Compares assurance savings, quantum-data prediction evidence, and an NMF audit.
Makes uncertainty, evidence budgets, and abstention decisions visible.
Uses frozen offline records, not live quantum hardware; the NMF audit reports no quantum advantage.