CogFormer: Learn All Your Models Once

2026-03-20

AI Summary

Traditional simulation-based inference methods in cognitive modeling struggle to efficiently adapt to changes in model structure, parameters, priors, or experimental design, often requiring retraining for each adjustment and thereby forfeiting the benefits of amortization. To address this limitation, this work proposes CogFormer, a meta-amortized framework built upon the Transformer architecture that, after a single training phase, enables general-purpose parameter inference across a vast family of structurally similar cognitive models. CogFormer achieves unprecedented scalability by supporting flexible inference across diverse data types, parameter configurations, design matrices, and sample sizes. It delivers highly accurate parameter estimates for binary, multi-alternative, and continuous-response decision models while incurring minimal amortization overhead, substantially accelerating the cognitive modeling workflow.

Technology Category

• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Probabilistic Circuits and Graphical Models
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• User Modeling, Personalization and Recommendation: Psychology-informed user models and recommender systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Simulation-based inference (SBI) with neural networks has accelerated and transformed cognitive modeling workflows. SBI enables modelers to fit complex models that were previously difficult or impossible to estimate, while also allowing rapid estimation across large numbers of datasets. However, the utility of SBI for iterating over varying modeling assumptions remains limited: changing parameterizations, generative functions, priors, and design variables all necessitate model retraining and hence diminish the benefits of amortization. To address these issues, we pilot a meta-amortized framework for cognitive modeling which we nickname the CogFormer. Our framework trains a transformer-based architecture that remains valid across a combinatorial number of structurally similar models, allowing for changing data types, parameters, design matrices, and sample sizes. We present promising quantitative results across families of decision-making models for binary, multi-alternative, and continuous responses. Our evaluation suggests that CogFormer can accurately estimate parameters across model families with a minimal amortization offset, making it a potentially powerful engine that catalyzes cognitive modeling workflows.

Problem

• simulation-based inference
• cognitive modeling
• amortization
• model retraining
• transformer

Innovation

• meta-amortization
• transformer-based architecture
• simulation-based inference
• cognitive modeling
• model generalization