AlphaRTC - RL-based Network Bandwidth Prediction in Real-time Communication
A part of my work during the internship at Microsoft Research Asia
Challenges: Model's poor adaptability to different scenarios.
My Contributions:
1. Added supervised information into RL: Proposed unified reward function and classification loss to improve model performance in multiple scenarios.
2. Improved model's performance by 72% in ISP scenario, 55% in Burst Loss scenario.
Cold Page Reclaim & Oversubscription Algorithm on Cloud Container Platform
A part of my work during the internship at Alibaba Group
Challenges: Improve memory utilization in datacenter without undermining Latency-sensitive services’ quality of service.
To be updated……