Relevant History Length Selection Using Conditional Entropy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive technologies face challenges in determining the optimal length of historical data required for accurate predictions, often resulting in excessive data storage or inaccurate predictions due to either storing too much data or discarding too much information.
Innovation Solution
A system and method that determine the length of a relevant history of instances of data by calculating the conditional entropy of historical data sequences and adjusting the sequence length to maintain reliable predictions while minimizing data storage, using a device with a second data set unit, a first data set unit, a first conditional entropy unit, a threshold unit, and a length determining unit to find the smallest sequence length within a defined entropy increase threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a long data history is used to train the machine, then prediction accuracy is improved, but data storage requirements and processing power increase significantly
Solution Approach 1:
The patent extracts and removes redundant historical data points from the dataset, keeping only the essential information needed for accurate predictions. This is achieved through automated algorithms that identify and eliminate unnecessary data, thereby reducing storage requirements while preserving prediction accuracy.
Solution Approach 2:
The patent changes the parameter of history length from a fixed long duration to a dynamically optimized length. By adjusting the history length parameter based on actual predictive needs rather than using maximum available history, the system achieves accurate predictions with reduced data storage requirements.
2Reliability
If a long data history is processed when new data arrives, then training accuracy is improved, but processing time and computational power increase
Solution Approach 1:
The patent extracts only the relevant portion of historical data needed for training, rather than processing the entire history. This selective extraction of necessary data points significantly reduces processing time while maintaining training accuracy.
Solution Approach 2:
The patent applies partial action by processing only the sufficient amount of historical data required for accurate predictions, rather than processing all available historical data. This partial processing approach reduces computational overhead and processing time.
3Reliability
If the whole stored data history is used for predictions, then prediction reliability is improved, but processing power requirements increase
Solution Approach 1:
The patent extracts and retains only the critical historical data points necessary for reliable predictions, removing redundant information. This extraction process reduces the volume of data that needs to be processed during predictions, thereby reducing processing power requirements while maintaining reliability.
4Device complexity
If a single history length is determined for all sensors, then system complexity is reduced, but prediction accuracy decreases because sensor-specific optimal lengths are not achieved
Solution Approach 1:
The patent applies local quality by determining optimal history lengths specifically for each sensor based on its individual characteristics and contribution to predictions. This sensor-specific optimization improves prediction accuracy by tailoring the history length to each sensor's needs rather than applying a uniform approach.
Data Source
AI summary
A device (100) for and a method of determining a length of a history of instances of a first type of data are provided. The device comprises a second data set unit (104), a first data set unit (102), a first conditional entropy unit (106), a threshold unit (108), and a length determining unit (114). The first data set unit, the second data set unit and the threshold unit obtain input data. The first conditional entropy unit determines a first conditional entropy for a second data set given a first data set. The determining unit searches for a length of the relevant history by determining a smallest length for sequences of a reduced first set of data for which a second conditional entropy for the second set of data, given the reduced first set of data, is within a maximum entropy increase threshold from the first conditional entropy.