y0news
← Feed
Back to feed
🧠 AI🟢 Bullish

CoME: Empowering Channel-of-Mobile-Experts with Informative Hybrid-Capabilities Reasoning

arXiv – CS AI|Yuxuan Liu, Weikai Xu, Kun Huang, Changyu Chen, Jiankun Zhao, Pengzhi Gao, Wei Liu, Jian Luan, Shuo Shang, Bo Du, Ji-Rong Wen, Rui Yan||1 views
🤖AI Summary

Researchers introduce Channel-of-Mobile-Experts (CoME), a new AI agent architecture that uses four specialized experts to handle different reasoning stages for mobile device automation. The system employs progressive training strategies and information gain-driven optimization to improve mobile agent performance on complex tasks.

Key Takeaways
  • CoME introduces a novel mobile agent architecture with four distinct experts handling screen summary, subtask planning, action decision, and action function.
  • The system uses output-oriented activation to selectively engage appropriate experts for different reasoning stages.
  • Progressive training strategy includes Expert-FT, Router-FT, and CoT-FT phases to enhance capabilities and coordination.
  • Info-DPO method uses information gain to evaluate intermediate steps and reduce error propagation in reasoning.
  • Experimental results show CoME outperforms existing dense mobile agents and MoE methods on AITZ and AMEX datasets.
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Connect Wallet to AI →How it works
Related Articles