RecNet
Modeling Strong and Human-Like Gameplay with KL-Regularized SearchAthul Paul Jacob, David J. Wu, Gabriele Farina, Adam Lerer, Hengyuan Hu, Anton Bakhtin, Jacob Andreas, Noam Brown
2022
Read
Robert Hawkins recommended on on 1/23/2024
introduces a policy-regularized hedge algorithm that searches for equilibria relatively close to a specified 'anchor' policy (via a KL term). I mostly liked the theoretical contribution, but the empirical results are also nice.
Exploring the hierarchical structure of human plans via program generationCarlos G. Correa, Sophia Sanborn, Mark K. Ho, Frederick Callaway, Nathaniel D. Daw, Thomas L. Griffiths
2023
Read
Robert Hawkins recommended on on 12/12/2023
Studies a game called LightBot where human participants plan action sequences in a domain-specific language with explicit hierarchical structure. They propose a model of hierarchy choice as the induction of a grammar over actions.
© 2024 RecNet. All rights reserved.