Algorithm Consolidation Reducing Computational Burden
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning algorithms become increasingly complex and computationally burdensome over time, especially when new versions require execution of all previous versions, leading to a significant increase in computational resources needed for execution.
Innovation Solution
A method is implemented to automatically consolidate machine learning algorithms by developing them from a complex state to a simpler state based on computational cost, using a machine learning process to determine if the algorithm can be reduced while maintaining performance, involving the execution of the algorithm on different input sets and determining matching values to adjust learning rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a machine learning process continuously develops the algorithm by adding new rules and versions, then the algorithm's performance and adaptability improve, but the computational burden and complexity increase significantly
Solution Approach 1:
The patent segments the algorithm development process into distinct phases: a first machine learning process adds new rules and versions to improve performance, while a second consolidation process periodically simplifies the algorithm by removing redundant elements. This segmentation allows the system to alternate between complexity-building and complexity-reducing operations, preventing uncontrolled growth while maintaining performance benefits.
Solution Approach 2:
The patent changes the parameter of algorithm complexity by introducing a consolidation operation that transforms the algorithm from a complex state (with many rules and versions) to a simplified state (with fewer, optimized rules). This parameter change is controlled by monitoring computational burden metrics and triggering consolidation when thresholds are exceeded, thus dynamically adjusting complexity levels.
2Reliability
If new versions of the algorithm require execution of all previous versions, then the algorithm incorporates cumulative learning, but the computational resources required increase with every new version
Solution Approach 1:
The patent changes the execution model parameter by transitioning from a cumulative execution model (where all previous versions must run) to a consolidated execution model (where a simplified version suffices). The consolidation process restructures the algorithm so that cumulative learning is preserved but executed more efficiently, reducing computational resources while maintaining reliability.
Solution Approach 2:
The patent extracts redundant and duplicate rules from the cumulative algorithm versions. The consolidation process identifies and removes unnecessary computational steps while retaining the essential learning outcomes, thus preserving cumulative learning benefits without the burden of executing all intermediate versions.
3Adaptability or versatility
If the algorithm encompasses a greater number of rules to handle variable environments, then its adaptability improves, but the computational burden increases
Solution Approach 1:
The patent segments rule management into rule creation (for adaptability) and rule consolidation (for efficiency). The first machine learning process creates detailed rules for handling variable environments, while the second consolidation process groups and optimizes these rules, maintaining environmental adaptability while improving execution efficiency by reducing redundant rule evaluations.
Data Source
AI summary
This disclosure relates to a computer-implemented method of controlling an algorithm, and a device for implementing said method, the method including developing the algorithm from a first state to a second state by a machine learning process; determining a second computational cost of executing the algorithm in its second state; determining if the second computational cost satisfies a trigger condition; and, if so, consolidating the algorithm from the second state to a third state, wherein a third computational cost of executing the algorithm in its third state is less than the second computational cost.