MOCHA: Multi-Objective Chebyshev Annealing for Agent Skill Optimization
The paper introduces MOCHA, a new method for optimizing agent skills in artificial intelligence. MOCHA employs Multi-Objective Chebyshev Annealing to address the multi-faceted constraints of skill optimization. Experimental results demonstrate that MOCHA significantly outperforms existing optimizers in various tasks, achieving notable improvements in skill performance.
- ▪MOCHA stands for Multi-Objective Chebyshev Annealing and is designed for agent skill optimization.
- ▪The method addresses the challenges of multi-objective optimization by covering the full Pareto front, including non-convex regions.
- ▪In experiments, MOCHA achieved a 7.5% relative improvement in mean correctness over the strongest baseline.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.19330 (cs) [Submitted on 19 May 2026] Title:MOCHA: Multi-Objective Chebyshev Annealing for Agent Skill Optimization Authors:Md Mehrab Tanjim, Jayakumar Subramanian, Xiang Chen, Branislav Kveton, Subhojyoti Mukherjee, Anlan Zhang, Sungchul Kim, Somdeb Sarkhel, Sunav Choudhury View a PDF of the paper titled MOCHA: Multi-Objective Chebyshev Annealing for Agent Skill Optimization, by Md Mehrab Tanjim and 8 other authors View PDF HTML (experimental) Abstract:LLM agents organize behavior through skills - structured natural-language specifications governing how an agent reasons, retrieves, and responds.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
