Syntra: Synthesizing Cross-Layer Controllers for Low-Latency Video Streaming
summary
A tool called Syntra synthesizes a cross-layer controller from a symbolic worst-case model and a declarative, lexicographic objective, using bounded-horizon minimax search over a partially observable game and distilling the search policy via imitation learning into a decision-tree policy. Models capture encoder overshoot and jitter as non‑deterministic bounds; synthesis runs in 3–4 hours. The produced controller jointly picks bitrate, send rate, FEC grouping, and frame skips, and on six real cellular traces it cuts tail latency significantly versus three baselines while keeping quality and frame rate comparable.
AI-written summary · about
chapters (8)
- 0:00Syntra intro and streaming challenges
- 3:53Classical controllers fail
- 5:58Non-deterministic worst-case modeling
- 7:23Encoder and network modeling
- 9:30Performance and safety goals
- 10:31Two-player game synthesis
- 12:36Evaluation on fixed and complex links
- 14:45Baselines and takeaways
related talks
16mSwift: Adaptive Video Streaming with Layered Neural Codecs
22mA Hybrid Meta-Learning Framework for Adaptive Safe Controller Synthesis of Dynamical(…)
14mARTEMIS: Adaptive Bitrate Ladder Optimization for Live Video Streaming
31mVehicle with Time: Signal First-Order Logic for Closed-Loop Controller Synthesis
21mLearning in situ: a randomized experiment in video streaming
17mCrystalBox: Efficient Model-Agnostic Explanations for Deep RL Controllers
23mYou Deserve Low-Latency Sports and We Serve It via Content-Aware Playback
26mMagic Mirror: Understanding the Diversified Network Conditions of Twitch Community