Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Offline Meta-learning for Real-time Bandwidth Estimation

Aashish Gottipati, Sami Khairy, Yasaman Hosseinkashi, Gabriel Mittag, Vishak Gopal, Francis Y. Yan, Ross Cutler

TL;DR

Ivy addresses the challenge of variable network conditions and data drift in real-time video bandwidth estimation by introducing an offline meta-learning metapolicy that selects among multiple BWE algorithms to maximize MOS-based QoE. It leverages Implicit Q-Learning on offline telemetry to learn a policy that makes 6-second at-a-time decisions, avoiding live exploration and reducing training data needs. In Microsoft Teams, Ivy improves QoE by up to 11.4% over online QoS meta-heuristics and over 5.9%–11.2% over individual estimators, with up to 28% better data efficiency than online methods. The work demonstrates the practicality of offline meta-learning for production systems facing nonstationarity and data drift and points to future hybrids that combine offline robustness with online adaptation.

Abstract

Real-time video applications require dynamic bitrate adjustments based on network capacity, necessitating accurate bandwidth estimation (BWE). We introduce Ivy, a novel BWE method that leverages offline meta-learning to combat data drift and maximize user Quality of Experience (QoE). Our approach dynamically selects the most suitable BWE algorithm for current network conditions, enabling effective adaptation to changing environments without requiring live network interactions. We implemented our method in Microsoft Teams and demonstrated that Ivy can enhance QoE by 5.9% to 11.2% over individual BWE algorithms and by 6.3% to 11.4% compared to existing online meta heuristics. Additionally, we show that our method is more data efficient compared to online meta-learning methods, achieving up to 21% improvement in QoE while requiring significantly less training data.

Offline Meta-learning for Real-time Bandwidth Estimation

TL;DR

Ivy addresses the challenge of variable network conditions and data drift in real-time video bandwidth estimation by introducing an offline meta-learning metapolicy that selects among multiple BWE algorithms to maximize MOS-based QoE. It leverages Implicit Q-Learning on offline telemetry to learn a policy that makes 6-second at-a-time decisions, avoiding live exploration and reducing training data needs. In Microsoft Teams, Ivy improves QoE by up to 11.4% over online QoS meta-heuristics and over 5.9%–11.2% over individual estimators, with up to 28% better data efficiency than online methods. The work demonstrates the practicality of offline meta-learning for production systems facing nonstationarity and data drift and points to future hybrids that combine offline robustness with online adaptation.

Abstract

Real-time video applications require dynamic bitrate adjustments based on network capacity, necessitating accurate bandwidth estimation (BWE). We introduce Ivy, a novel BWE method that leverages offline meta-learning to combat data drift and maximize user Quality of Experience (QoE). Our approach dynamically selects the most suitable BWE algorithm for current network conditions, enabling effective adaptation to changing environments without requiring live network interactions. We implemented our method in Microsoft Teams and demonstrated that Ivy can enhance QoE by 5.9% to 11.2% over individual BWE algorithms and by 6.3% to 11.4% compared to existing online meta heuristics. Additionally, we show that our method is more data efficient compared to online meta-learning methods, achieving up to 21% improvement in QoE while requiring significantly less training data.
Paper Structure (14 sections, 2 equations, 5 figures, 2 tables)

This paper contains 14 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Design Decisions
  5. Problem Formulation
  6. Data Collection
  7. Training Details and Implementation
  8. Evaluation
  9. Setup
  10. Comparison with Baseline Estimators
  11. Offline Learning vs. Online Meta Heuristics
  12. Strength in Multiple Estimators
  13. Offline vs. Online Learning
  14. Conclusion

Figures (5)

  • Figure 1: Ivy Network Architecture.
  • Figure 2: Comparison between Ivy and Individual Estimators.
  • Figure 3: Ivy vs. Online QoS-based Meta Heurstics.
  • Figure 4: Comparison between Ivy and Individual Estimators in Nonstationary Environments.
  • Figure 5: Offline vs Online Meta-learning.