Skip to content

Machine Learning (ML) materials agent ​

Status: planned. Purpose: learn patterns for material property prediction and inverse design.

Develop and validate domain models for virtual screening, multi-objective optimization and experiment design. This is the intended workflow; future releases define actual pages and models.

Expected page content ​

AreaIntended content
DatasetScope, sources, features, labels, units and quality
ModelTask, training method, version and applicability
ValidationData splits, metrics, error and uncertainty
Design taskTarget properties, composition or process constraints and candidate scope
FeedbackCandidate plans, real test results and discrepancies

Target workflow ​

  1. Select authorized data; check missing values, units, duplicates and conditions.
  2. Define features and target properties; retain data versions and sources.
  3. Train and validate on independent data; record error and applicability.
  4. Predict properties, screen candidates or optimize multiple objectives.
  5. Generate candidate experiments under target properties and constraints for researcher selection.
  6. Save measured and predicted values separately, compare errors and return results to the database.

Prediction and inverse design ​

Prediction estimates properties from known composition or conditions. Inverse design searches compositions and conditions from target properties and constraints. Both require suitable models and validation; recommendations must be tested experimentally.

Platform vision ​

Formulation design → experiment execution → data accumulation → automated analysis → model learning → experiment recommendations → revalidation. New results enrich domain data for the next model and design cycle.

Matching learning resource ​

Built for experiment data workflows.