R2-B2: Recursive Reasoning-Based Bayesian Optimization for No-Regret Learning in Games
Authors: Zhongxiang Dai, Yizhou Chen, Bryan Kian Hsiang Low, Patrick Jaillet, Teck-Hua Ho
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The performance and generality of our R2-B2 algorithm are empirically demonstrated using synthetic games, adversarial machine learning, and multi-agent reinforcement learning. This section empirically evaluates the performance of our R2-B2 algorithm and demonstrates its generality using synthetic games, adversarial ML, and MARL. |
| Researcher Affiliation | Academia | Zhongxiang Dai 1 Yizhou Chen 1 Bryan Kian Hsiang Low 1 Patrick Jaillet 2 Teck-Hua Ho 3 1Department of Computer Science, National University of Singapore, Republic of Singapore 2Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, USA 3NUS Business School, National University of Singapore, Republic of Singapore. |
| Pseudocode | Yes | Algorithm 1 R2-B2 for attacker A s level-k reasoning 1: for t = 1, 2, . . . , T do 2: Select input action x1,t using its level-k strategy (while defender D selects input action x2,t) 3: Observe noisy payoff y1,t = f1(x1,t, x2,t) + ϵ1 4: Update GP posterior belief using (x1,t, x2,t), y1,t |
| Open Source Code | No | No explicit statement about providing access to source code for the methodology, nor a link to a code repository, was found. |
| Open Datasets | Yes | We apply our R2-B2 algorithm to black-box adversarial ML for image classification problems with deep neural networks (DNNs) using the MNIST and CIFAR-10 image datasets. |
| Dataset Splits | No | The paper mentions using MNIST and CIFAR-10 datasets but does not explicitly provide details about training, validation, and test splits (e.g., percentages or specific sample counts) or references to predefined splits for reproducibility beyond simply naming the datasets. |
| Hardware Specification | No | No specific hardware details (e.g., CPU/GPU models, memory, or cluster specifications) used for running the experiments were provided in the paper. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies or libraries used (e.g., Python, PyTorch, TensorFlow, scikit-learn), which would be necessary for reproducibility. |
| Experiment Setup | Yes | In our experiments, variational autoencoders (VAE) (Kingma & Welling, 2014) are used by both A and D to project the images to a lowerdimensional space (i.e., 2D for MNIST and 8D for CIFAR-10). The maximum allowed perturbation to each pixel added by either A or D is no more than a pre-defined value ϵ where ϵ = 0.2 for MNIST and ϵ = 0.05 for CIFAR10. the results are averaged over 10 initializations of 5 randomly selected actions with observed payoffs. |