Motivation#

Network AI Models/Algorithms Development Cycle#

../_images/motivation.png

Challenges Faced by Network AI Developers#

  1. real-world dataset controlled by network operator, difficult to acquire, not aligned with specific usage or requirement.

  2. dataset by itself not enough, also need environment to train/test AI models, e.g., Reinforcement Learning, etc.

NetworkGym’s Approach to Addressing this Challenge

✔️ Currently, NetworkGym environment enable 3 use cases: multi-access traffic splitting, QoS-aware traffic steering, and (cellular) RAN slicing.

  1. network simulation tools (e.g., ns-3, etc.) often very complex and difficult to use, especially for Network AI researcher and developer.

NetworkGym’s Approach to Addressing this Challenge

✔️ NetworkGym enables agent training without the requirement of network simulation expertise.

  1. lack of common simulation environment with simple APIs to develop, evaluate, and benchmark Network AI models and algorithms.

NetworkGym’s Approach to Addressing this Challenge

✔️ NetworkGym adheres to the standard gymnasium API for AI model training and additionally offers an API for network simulation configuration.