The Value Equivalence Principle for Model-Based Reinforcement Learning

Authors: Christopher Grimm, Andre Barreto, Satinder Singh, David Silver

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate the benefits of value-equivalent model learning with experiments comparing it against more traditional counterparts like maximum likelihood estimation. (Abstract) We now present experiments illustrating the usefulness of the value equivalence principle in practice.
Researcher Affiliation Collaboration Christopher Grimm Computer Science & Engineering University of Michigan crgrimm@umich.edu André Barreto, Satinder Singh, David Silver Deep Mind {andrebarreto,baveja,davidsilver}@google.com
Pseudocode No The paper includes mathematical formulations, such as equation (6) for the value-equivalence loss, but it does not present any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statement about releasing the source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets No The paper mentions using well-known domains like "four rooms [37], catch [25], and cart-pole [4]" for experiments. However, it does not provide specific access information (e.g., URLs, DOIs, or dataset names with formal citations including author and year for direct access) for publicly available datasets used.
Dataset Splits No The paper states that for some experiments, "we collected 1000 sample transitions" or "10000 sample transitions" (Appendix A.2). However, it does not specify how these samples were split into training, validation, and test sets with percentages, sample counts for validation, or reference to predefined splits.
Hardware Specification No We did not use any hardware specific setup or software dependencies other than standard libraries for Python (e.g., NumPy, TensorFlow/PyTorch).
Software Dependencies No We did not use any hardware specific setup or software dependencies other than standard libraries for Python (e.g., NumPy, TensorFlow/PyTorch). (Appendix A.2)
Experiment Setup Yes All neural networks were trained using Adam optimizer with a learning rate of 0.001.