Evolutionary Game Theory Squared: Evolving Agents in Endogenously Evolving Zero-Sum Games
Authors: Stratis Skoulakis, Tanner Fiez, Ryann Sim, Georgios Piliouras, Lillian Ratliff11343-11351
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In artificial intelligence (Wang et al. 2019; Garciarena, Santana, and Mendiburu 2018; Costa et al. 2019; Miikkulainen et al. 2019; Wu et al. 2019; Stanley and Miikkulainen 2002) as well as biology, sociology, and economics (Stewart and Plotkin 2014; Tilman, Plotkin, and Akçay 2020; Tilman, Watson, and Levin 2017; Bowles, Choi, and Hopfensitz 2003; Weitz et al. 2016), the rules of interaction can themselves adapt to the collective history of the agent behavior. For example, in adversarial learning and curriculum learning (Huang et al. 2011; Bengio et al. 2009), the difficulty of the game can increase over time by exactly focusing on the settings where the agent has performed the weakest. Similarly, in biology or economics, if a particular advantageous strategy is used exhaustively by agents, then its relative advantages typically dissipate over time (negative frequency-dependent selection, see Heino, Metz, and Kaitala 1998), which once again drives the need for innovation and exploration. In all these cases, the game itself stops being a passive object that the agents act upon, but instead is best thought of as an algorithm itself. Similar to online learning algorithms employed by agents, the game itself may have a memory/state that encodes history. However, unlike online learning algorithms that receive a history or sequence of payoff vectors and output the current behavior (e.g., a probability distribution over actions), an algorithmic game receives as input a history or sequence of agents behavior and outputs a new payoff matrix. Hence, learning and games are dual" algorithmic objects which are coupled in their evolution (Figure 1). |
| Researcher Affiliation | Academia | 1 Singapore University of Technology and Design 2 University of Washington |
| Pseudocode | No | The paper describes algorithms conceptually but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at github.com/ryanndelion/egt-squared |
| Open Datasets | No | The paper focuses on theoretical models and simulations of game theory dynamics (e.g., generalized Rock-Paper-Scissors model, polymatrix games) rather than traditional datasets. Therefore, no information about publicly available or open datasets is provided. |
| Dataset Splits | No | The paper analyzes theoretical models and provides simulations of game dynamics, not experiments on traditional datasets with explicit training, validation, and test splits. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the simulations. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers. |
| Experiment Setup | No | The paper describes the mathematical models and the dynamics being simulated but does not provide specific experimental setup details such as hyperparameters or system-level training settings. |