Researchmodel distillationmobile deploymentmobapareto optimization
Distilled Game Agents Achieve Mobile Efficiency
7.1
Researchers Xionghui Yang et al. (arXiv v1, Feb 7, 2026) present a Pareto optimality–guided pipeline and mobile-focused student architecture search to compress Honor of Kings agents for practical on-device deployment. The distilled agent achieves 12.4× faster inference (under 0.5 ms/frame), 15.6× energy efficiency (under 0.5 mAh/game), and retains a 40.32% win rate versus the teacher, demonstrating feasible mobile MOBA deployment.
Key Points
- 1Distillates achieve 12.4× faster inference and 15.6× better energy efficiency on Honor of Kings.
- 2Introduces Pareto-guided pipeline and mobile-focused student architecture search to balance performance and efficiency.
- 3Enables deployment of complex multi-modal MOBA agents under 0.5ms latency and low energy consumption.
Scoring Rationale
Strong empirical efficiency gains and practical pipeline justify high impact, limited by single preprint source.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
