AI inversion#

The AI inversion guide documents the complete learned-inversion lifecycle in pyCSAMT: scientific assumptions, training-data generation, architecture selection, training, inference, validation, uncertainty, PINN workflows, agent-assisted execution, and reporting.

AI inversion is a peer of Forward Modelling and Classical model integrations. Forward modeling generates responses from known earth models; classical inversion estimates models through numerical optimization; AI inversion learns an inverse mapping from representative examples. These approaches can support one another, but their assumptions and validation requirements remain distinct.

Validation is part of the model

A network prediction is not trustworthy merely because inference is fast or a training loss is small. Field use requires representative training data, strict data separation, out-of-distribution checks, physical diagnostics, uncertainty assessment, and comparison with independent evidence.

Build the workflow#

Concepts

Understand surrogate inversion, supported dimensions, assumptions, limitations, and the relationship to forward and classical inversion.

AI inversion concepts
Data preparation

Build synthetic and field datasets, define features and targets, normalize consistently, split safely, and preserve provenance.

AI inversion data preparation
Model selection

Choose 1-D, profile-based 2-D, graph-based 3-D, or physics-informed approaches according to geometry, data volume, and scientific goals.

AI model selection
Training

Configure optimization, monitor learning, resume safely, diagnose overfitting, and retain reproducible checkpoints and histories.

Training AI inversion models
Inference

Apply a trained model to compatible observations, verify preprocessing, detect unsupported inputs, and export predictions with metadata.

AI inversion inference
Validation

Evaluate held-out synthetic tests, field responses, physical consistency, baselines, failure modes, and independent geological evidence.

AI inversion validation

Review and extend#

Uncertainty

Quantify predictive spread, calibration, ensemble behavior, domain shift, and uncertainty sources not captured by the network.

AI inversion uncertainty
PINN 2-D

Add physical residuals and constraints to learned 2-D inversion while keeping numerical assumptions and validation explicit.

Physics-informed 2-D inversion
AI inversion agents

Use pyCSAMT agents to configure and coordinate AI workflows without hiding the underlying science API or review requirements.

AI inversion agents
Reporting

Package datasets, configurations, checkpoints, metrics, predictions, uncertainty, limitations, and approvals into an auditable model record.

AI inversion reporting

Contents#