Reinforcement Learning for Streaming Channel Personalization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing streaming quality adjustment methods fail to learn user preferences and apply a one-size-fits-all approach, leading to suboptimal user experiences, especially in high-input scenarios like gaming.

Innovation Solution

A system that uses a reinforcement learning approach to dynamically adjust streaming channel parameters based on real-time user feedback and interaction history, personalizing the streaming experience for each user.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If heuristic solutions are used for streaming quality adjustment, then the system is simple to implement, but it cannot learn user preferences and applies a one-size-fits-all approach

Engineering Contradiction:
Improvesimplicity of implementationVSAvoidability to learn user preferences
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent implements a feedback mechanism where user responses (thumbs up/thumbs down) are collected during streaming and used to update the reinforcement learning model. This allows the system to continuously learn and adapt to individual user preferences for streaming quality, resolving the contradiction between simple implementation and adaptability.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system uses reinforcement learning to automatically adjust streaming quality parameters based on learned user preferences, eliminating the need for manual configuration or complex heuristic rules. The model self-optimizes by learning from user feedback, achieving both simplicity and adaptability.

Inventive Principle:
Principle #25Self-service

2Ease of operation

If reinforcement learning is used to personalize streaming quality, then user experience is improved, but system complexity increases

Engineering Contradiction:
Improveuser experience qualityVSAvoidsystem complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent changes the parameters being optimized from complex multi-dimensional quality metrics to simple binary user feedback (thumbs up/thumbs down). This simplifies the reinforcement learning problem while still achieving personalized quality optimization, resolving the contradiction between improved user experience and reduced system complexity.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If real-time user feedback is collected and processed, then streaming quality is optimized for individual users, but processing time and computational resources increase

Engineering Contradiction:
Improvestreaming quality optimizationVSAvoidprocessing time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by maintaining a pre-trained reinforcement learning model that can quickly infer optimal streaming parameters from user feedback. Rather than training from scratch with each user interaction, the pre-trained model enables rapid adaptation, reducing processing time while maintaining optimization quality.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4103297B1Streaming channel personalization
Publication Date: 2025.05.07 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4103297B1 patent drawingFigure 1
  • EP4103297B1 patent drawingFigure 2A
  • EP4103297B1 patent drawingFigure 2B

AI summary

The present disclosure relates to devices and methods for personalizing channel parameters for streaming content to a client device by dynamically adjusting channel parameters in response to learned user preferences. The devices and methods may receive context information from a client device and may send a rank and reward call to a reinforcement learning system for a recommendation for a value of the channel parameters. The rank and reward call may include the context information, a user vector, an item vector and a reward function error. The reinforcement learning system may use the information provided in the rank and reward call to the provide a recommendation for the value of the channel parameters. The devices and methods may use the recommendation to set the value of the channel parameters to stream the content to the client device.