AI
MoFlow: Multi-Objective Agentic Workflow Generation
Publication: cs.AI updates on arXiv.org
Author: Yining Lu, Aurelie Lozano, Xi Yang, Naoki Abe, Yu Deng, Meng Jiang
Source published: 01/10/2026, 04:00:00 UTC
Briefing created: 01/10/2026, 04:07:29 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 04:03:24 UTC.
Briefings
MoFlow is a new system for generating agentic workflows that optimize multiple objectives like accuracy, cost, and latency simultaneously. Unlike existing methods that optimize a single objective and require retraining, MoFlow uses a multi-objective Markov decision process with Convex-Hull Monte Carlo Tree Search to approximate the Pareto front. This allows it to return a workflow for any preference without retraining. The system was evaluated against six baselines on six benchmarks covering mathematics, code, and question answering, achieving the highest average hypervolume despite the difficulty of direct comparison due to the baselines' rerun setup.
Key points
- MoFlow formulates workflow generation as a multi-objective Markov decision process using Convex-Hull Monte Carlo Tree Search with optimistic set-valued backups.
- MoFlow achieves the highest average hypervolume against six baselines on six benchmarks spanning mathematics, code, and question answering.
Context and limitations
Evaluation setup favors baselines by having them rerun for each testing preference, which MoFlow never sees, making direct comparison difficult. Baselines are single-scalar optimizers by design, and the evaluation favors the baselines, which MoFlow never sees. Coverage is limited to six benchmarks spanning mathematics, code, and question answering.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 04:07:29 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeibkbogrsuwvspzzshnq22npgwihozcf7wwojoeqvnb3eszzsznhyi