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..
Heterogeneous Adversarial Play in Interactive Environments
Authors: Manjie Xu, Xinyi Yang, Jiayu Zhan, Wei Liang, Chi Zhang, Yixin Zhu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental validation across multi-task learning domains demonstrates that our framework achieves performance parity with state-of-the-art (SOTA) baselines while generating curricula that enhance learning efficacy in both artificial agents and human subjects. We evaluate HAP s scalability across increasingly complex task distributions, encompassing open-world scenarios and environments featuring intricate task dependencies and interconnections. (From Abstract and Section 4 'Experiments') |
| Researcher Affiliation | Academia | 1 Institute for Artificial Intelligence, Peking University 2 School of Psychological and Cognitive Sciences, Peking University 3 School of Computer Science & Technology, Beijing Institute of Technology |
| Pseudocode | Yes | Algorithm 1: Training loop of the Heterogeneous Adversarial Play (HAP) Data: Initial θ, ϕ; learning rates α, β |
| Open Source Code | No | The code will be released in our project website. |
| Open Datasets | Yes | Minigrid (Chevalier-Boisvert et al., 2019) provides a highly configurable grid-world platform..., CRAFT (Andreas et al., 2017) represents a classic multi-task RL environment..., Crafter (Hafner, 2022) introduces open-world elements..., CIFAR-100 (Krizhevsky, 2009)) and natural language processing (Recognizing Textual Entailment (RTE) from GLUE (Wang et al., 2019a)). |
| Dataset Splits | No | The paper uses several datasets/environments (Minigrid, CRAFT, Crafter, CIFAR-100, RTE) but does not provide specific train/test/validation splits for these datasets. It mentions splitting *tasks* into difficulty levels, but not dataset samples into training, validation, or testing sets with percentages or counts. |
| Hardware Specification | Yes | conducting all experiments on a single NVIDIA A100 GPU. |
| Software Dependencies | No | The paper mentions using open-source implementations from Stable-Baselines3 and the official CRAFT and Crafter Repo, but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | We have included these details in the paper. (from NeurIPS Paper Checklist Section 6). More specifically, in Section C and tables A5, A6, A7, A8, which list numerous model parameters and training configurations. |