Get in touch

CogFormer All you need is one meta-model.

Explore a large variety of cognitive models with CogFormer, which handles model changes across any experimental designs.

Try the demoSource code

With one trained network, CogFormer can perform inference on

Simulate all models.

Experimental designs change parameters. Parameters define a model family. Related families form a model class. CogFormer brings these levels into one inference workflow, across all models and designs in a domain.

Model hierarchy: experimental covariates define a model species, parameters define a model family such as the DDM, and related families such as DDM, RDM, and CDM form a sequential-sampling model class.Model hierarchy: model species, model families, and the sequential-sampling model class.

DDM · 1 / illustrative recovery

Make your designs. Infer with one model.

CogFormer adapts to your design so one meta-model can estimate across the configurations it was trained to support.

Four DDM designs. Four RDM designs. Four CDM designs. Each varies the effects of two regressors and their interaction, sharing one four-by-eight recovery grid. Each design appears on its own, followed by the full ensemble. Select a design to pause and inspect it. This animation uses mock data.

Now make it yours.

Choose a pretrained estimator, build your design, and simulate data. Run inference and explore the posterior, recovery, and calibration for yourself.

Open in new tab

Documentation & evidence

  1. CogFormer paper

    Method, model-family construction, training setup, and empirical evaluation.

  2. CogFormer repository

    Source code and installation.

  3. CogFormer interactive demo

    Pretrained estimators, design configuration, simulated data, and inference diagnostics.

All products