Minimal MERINA+
A compact, readable reimplementation of meta-RL adaptive bitrate streaming.
A compact, self-contained reproduction of the core ideas in MERINA+ (IEEE TCSVT 2025), a meta-reinforcement-learning approach to adaptive bitrate (ABR) video streaming. The goal is to make the method easy to read, run, and reason about: a pure-Python simulator, three small networks, and a PPO + variational-information-bottleneck training loop that runs in a few minutes on a free Colab GPU or a laptop CPU.
This is a teaching-oriented reimplementation focused on the algorithmic core rather than the paper’s exact benchmark numbers. Original implementation: confiwent/merina-plus.
Stack: Python · PyTorch · PPO