ALSO: Adversarial Online Strategy Optimization for Social Agents
The paper presents a new framework called ALSO for optimizing strategies in multi-agent social simulations. This framework addresses the challenges posed by non-stationary environments, where agents must adapt their strategies dynamically. Experiments show that ALSO outperforms existing methods, demonstrating its effectiveness in enhancing social intelligence among agents.
- ▪ALSO stands for Adversarial Online Strategy Optimization.
- ▪The framework treats multi-turn interactions as an adversarial bandit problem.
- ▪ALSO introduces a lightweight neural surrogate to predict rewards from interaction histories.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.15768 (cs) [Submitted on 15 May 2026] Title:ALSO: Adversarial Online Strategy Optimization for Social Agents Authors:Xiang Li, Liping Yi, Mingze Kong, Min Zhang, Zhongxiang Dai, QingHua Hu View a PDF of the paper titled ALSO: Adversarial Online Strategy Optimization for Social Agents, by Xiang Li and 5 other authors View PDF HTML (experimental) Abstract:Social simulation provides a compelling testbed for studying social intelligence, where agents interact through multi-turn dialogues under evolving contexts and strategically adapting opponents. Such environments are inherently non-stationary, requiring agents to dynamically adjust their strategies over time.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
