CogPonder is a flexible, differentiable model of cognitive control that is inspired by the Test-Operate-Test-Exit (TOTE) architecture in psychology and the PonderNet framework in deep learning. CogPonder functionally decouples the act of control from the controlled processes by introducing a controller that wraps around any end-to-end deep learning model and decides when to terminate processing and output a response, thus producing both a response and response time.
CCN2023 Paper · CCN2023 Poster
Warning
This is a work-in-progress. Model architecture and results are subject to change as we continue to develop and refine the model.
To install the dependencies, you need Conda (or even better, Mamba). Then, create a new environment with the dependencies as follows:
mamba env create -f environment.yml
mamba activate cogponder
dvc update --rev master -R data # import the SRO data using DVCTo install additional GPU dependencies run the following:
mamba env update -f environment_gpu.yml --pruneThe notebooks are organized as follows:
notebooks/N-Back.ipynb: training a single-task 2-back agent.notebooks/Stroop.ipynb: training a single-task Stroop agent.
see data/Self_Regulation_Ontology/README.md
@Conference{ansarinia2023cogponder,
title={CogPonder: Towards a Computational Framework of General Cognitive Control},
author={Ansarinia, Morteza and Cardoso-Leite, Pedro},
year={2023},
month={August},
doi={10.32470/CCN.2023.1697-0},
booktitle={2023 Conference on Cognitive Computational Neuroscience, Oxford, UK}
}