Learning to Hand Off: Provably Convergent Workflow Learning under Interface Constraints
The paper discusses a new approach to workflow learning in multi-agent systems where agents hand off control through a shared artifact. It introduces an asynchronous decentralized Q-learning algorithm called IC-$Q$, which operates under interface constraints. The authors provide a finite-sample bound for this algorithm, demonstrating its effectiveness through various experiments.
- ▪The study focuses on workflow learning in multi-agent systems with interface constraints.
- ▪IC-$Q$ is a decentralized Q-learning algorithm designed for agents that do not observe joint trajectories.
- ▪The authors establish a finite-sample bound for neural IC-$Q$ and validate it through experiments.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.19140 (cs) [Submitted on 18 May 2026] Title:Learning to Hand Off: Provably Convergent Workflow Learning under Interface Constraints Authors:Jiayu Li, Enpei Zhang, Dawei Zhou, Elynn Chen, Yujun Yan View a PDF of the paper titled Learning to Hand Off: Provably Convergent Workflow Learning under Interface Constraints, by Jiayu Li and 4 other authors View PDF HTML (experimental) Abstract:We study workflow learning in a setting where specialized agents hand off control through a shared artifact, each agent observes only a local function of that artifact and its own private state, and no centralized learner accesses joint trajectories -- the operating regime of multi-agent LLM pipelines that span organizational, vendor, or trust boundaries.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
