**Pensieve: Neural Adaptive Video Streaming** **Author:** Hongzi Mao **Submitted to:** Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology **Date:** May 19, 2017 **Abstract:** Client-side video players use bitrate adaptation (ABR) algorithms to meet users' growing quality of experience (QoE) demands. These algorithms must balance multiple QoE factors, such as maximizing video bitrate and minimizing rebuffering times. Despite the abundance of recent ABR algorithms, state-of-the-art schemes suffer from two practical challenges: (1) inaccurate throughput predictions, which can lead to degraded performance, and (2) fixed heuristics that are tuned for specific deployment environments, limiting generalization across network conditions and QoE objectives. To address these challenges, Pensieve is developed to generate ABR algorithms using Reinforcement Learning (RL). Pensieve trains a neural network model that selects bitrates for future video chunks based on observations collected by client video players. Unlike existing approaches, Pensieve does not rely on pre-programmed models or assumptions about the environment. Instead, it learns to make ABR decisions through observations of the resulting performance of past decisions. As a result, Pensieve can automatically learn ABR algorithms that adapt to a wide range of environmental conditions and QoE metrics. **Evaluation:** Pensieve is compared to state-of-the-art ABR algorithms using trace-driven and real-world experiments spanning various network conditions, QoE metrics, and video properties. In all considered scenarios, Pensieve outperforms the best state-of-the-art scheme, with average QoE improvements ranging from 13.1% to 25.0%. Pensieve's policies generalize well, outperforming existing schemes even on networks not trained on. **Contributions:** - Pensieve uses RL to train a neural network model for bitrate adaptation. - Pensieve learns ABR decisions through observations of past performance, avoiding the need for pre-programmed models. - Pensieve generalizes well across different network conditions and video properties. - Pensieve achieves significant QoE improvements over state-of-the-art ABR algorithms. **Conclusion:** Pensieve is a robust and adaptive ABR system that leverages RL to learn from experience, outperforming existing schemes in various network conditions and QoE metrics.**Pensieve: Neural Adaptive Video Streaming** **Author:** Hongzi Mao **Submitted to:** Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology **Date:** May 19, 2017 **Abstract:** Client-side video players use bitrate adaptation (ABR) algorithms to meet users' growing quality of experience (QoE) demands. These algorithms must balance multiple QoE factors, such as maximizing video bitrate and minimizing rebuffering times. Despite the abundance of recent ABR algorithms, state-of-the-art schemes suffer from two practical challenges: (1) inaccurate throughput predictions, which can lead to degraded performance, and (2) fixed heuristics that are tuned for specific deployment environments, limiting generalization across network conditions and QoE objectives. To address these challenges, Pensieve is developed to generate ABR algorithms using Reinforcement Learning (RL). Pensieve trains a neural network model that selects bitrates for future video chunks based on observations collected by client video players. Unlike existing approaches, Pensieve does not rely on pre-programmed models or assumptions about the environment. Instead, it learns to make ABR decisions through observations of the resulting performance of past decisions. As a result, Pensieve can automatically learn ABR algorithms that adapt to a wide range of environmental conditions and QoE metrics. **Evaluation:** Pensieve is compared to state-of-the-art ABR algorithms using trace-driven and real-world experiments spanning various network conditions, QoE metrics, and video properties. In all considered scenarios, Pensieve outperforms the best state-of-the-art scheme, with average QoE improvements ranging from 13.1% to 25.0%. Pensieve's policies generalize well, outperforming existing schemes even on networks not trained on. **Contributions:** - Pensieve uses RL to train a neural network model for bitrate adaptation. - Pensieve learns ABR decisions through observations of past performance, avoiding the need for pre-programmed models. - Pensieve generalizes well across different network conditions and video properties. - Pensieve achieves significant QoE improvements over state-of-the-art ABR algorithms. **Conclusion:** Pensieve is a robust and adaptive ABR system that leverages RL to learn from experience, outperforming existing schemes in various network conditions and QoE metrics.