Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Working Memory Graphs
Authors: Ricky Loynd, Roland Fernandez, Asli Celikyilmaz, Adith Swaminathan, Matthew Hausknecht
ICML 2020 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate WMG in three environments featuring factored observation spaces: a Pathfinding environment that requires complex reasoning over past observations, Baby AI gridworld levels that involve variable goals, and Sokoban which emphasizes future planning. We find that the combination of WMG s Transformer-based architecture with factored observation spaces leads to significant gains in learning efficiency compared to baseline architectures across all tasks. |
| Researcher Affiliation | Industry | 1Microsoft Research AI, Redmond, Washington, USA. |
| Pseudocode | No | The paper describes the WMG model and training process using text and mathematical equations, but it does not include an explicitly labeled pseudocode block or algorithm. |
| Open Source Code | Yes | To encourage further work and comparative studies, we provide WMG s source code and pre-trained models at https://github.com/microsoft/wmg_agent. |
| Open Datasets | Yes | We evaluate WMG in three environments featuring factored observation spaces: a Pathfinding environment..., Baby AI gridworld levels..., and Sokoban... Baby AI domain (Chevalier-Boisvert et al., 2018)... Sokoban (Guez et al., 2019). |
| Dataset Splits | No | The paper mentions hyperparameter tuning and |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper describes the model and training process but does not specify any software dependencies with version numbers (e.g., Python version, PyTorch/TensorFlow version, specific library versions). |
| Experiment Setup | Yes | For this task, WMG is configured with Memos but no Factors. The number of Memos is a tuned hyperparameter, equal to 16 in this experiment. (See Appendix C for all settings.) |