Video Delivery Decision Trees With Decision-Aware Splitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content delivery systems face challenges in developing cost-effective and discriminating tree structures for predicting the performance of decisions, such as content delivery network selection and initial profile selection, due to the lack of discrimination for decision features and the linear cost increase with multiple decisions, leading to increased storage and computing resources.
Innovation Solution
A unified model using a decision-aware splitting process and gain-based splitting process to build a tree structure that discriminates between decisions, ensuring each decision feature has a separate subtree, thereby reducing the need for multiple models and optimizing storage and computing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate prediction models are built for each decision, then the discrimination of decisions is improved, but the device complexity and computational cost increase linearly with the number of decisions
Solution Approach 1:
The patent merges multiple separate prediction models into a single unified tree structure by including decision features as additional features in the model. This unified structure processes multiple decisions simultaneously, eliminating the need for separate models for each decision while maintaining the ability to discriminate between different decision outcomes.
Solution Approach 2:
The unified tree structure serves multiple functions by handling various decision types (content delivery network selection, initial profile selection, bitrate selection) within a single model. The decision features enable the same structure to make different types of predictions depending on the input features provided.
2Measurement precision
If the decision feature is included as an additional feature in the tree structure, then the discrimination of decisions is improved, but the training complexity and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-defining the decision feature structure and integrating it into the training process. The decision features are incorporated as additional features with known relationships to the target decisions, which guides the tree structure generation process and reduces the complexity of discovering these relationships during training.
3Adaptability or versatility
If multiple decisions are needed at the same time, then the comprehensiveness of the system is improved, but the decision space increases dramatically resulting in increased storage and computational cost
Solution Approach 1:
The patent combines multiple decision-making processes into a single unified tree structure that handles content delivery network selection, initial profile selection, and bitrate selection simultaneously. By sharing common features and decision logic across these decisions, the system avoids the exponential growth in decision space that would result from treating them as separate independent models.
Data Source
AI summary
In some embodiments, a method receives a data structure for an auto decision process for predicting a value for a decision feature of a plurality of features. The data structure is generated using a decision aware splitting process that splits a node in the data structure using the decision feature and a metric-based process that splits other nodes of the data structure by evaluating features in the plurality of features using a metric. Feature values are determined for a combination of features for a playback session. The method evaluates the data structure using the combination of features to determine a plurality of prediction values for a plurality of feature values of the decision feature. The prediction values are used to generate a decision for the playback session by selecting a feature value in the plurality of feature values based on the plurality of prediction values.