Multi-agent Dynamic Algorithm Configuration
Authors: Ke Xue, Jiacheng Xu, Lei Yuan, Miqing Li, Chao Qian, Zongzhang Zhang, Yang Yu
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show the effectiveness of MA-DAC in not only achieving superior performance compared with other configuration tuning approaches based on heuristic rules, multi-armed bandits, and single-agent RL, but also being capable of generalizing to different problem classes. |
| Researcher Affiliation | Collaboration | Ke Xue1 , Jiacheng Xu1 , Lei Yuan1,2, Miqing Li3, Chao Qian1 , Zongzhang Zhang1, Yang Yu1,2 1 State Key Laboratory for Novel Software Technology, Nanjing University 2 Polixir Technologies 3 CERCIA, School of Computer Science, University of Birmingham |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https://github.com/lamda-bbo/madac. |
| Open Datasets | Yes | We use the MOPs with similar properties from the well-known MOP benchmarks DTLZ [9] and WFG [16] as the instance set of Ma Mo. In all of the following experiments, several arbitrary functions (here DTLZ2, WFG4 and WFG6) from the function set are used as the training set, and all the other functions are considered as the testing set. |
| Dataset Splits | No | The paper mentions training and testing sets based on problem instances but does not explicitly provide details about a separate validation set or its split percentages for model tuning within those instances. |
| Hardware Specification | Yes | All experiments are conducted on computers with Intel(R) Xeon(R) Gold 6230 CPU @ 2.10GHz processors and 256 GB of RAM. |
| Software Dependencies | No | The paper refers to various algorithms and frameworks (e.g., VDN, DQN, MA-UCB) and programming languages (e.g., Python implied by code link), but does not specify software library versions (e.g., PyTorch 1.9, numpy 1.20). |
| Experiment Setup | Yes | We use the Adam optimizer [26] with a learning rate of 1e-4. The batch size is set to 256, and the replay buffer size is 1e6. |