Dynamic Early Classification for Precision-Time Trade-off
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing identification apparatuses face a trade-off between improving the precision of identifying input data classes and reducing the time required for identification, as current objective functions do not effectively balance these two factors, often resulting in insufficient reductions in time or precision.
Innovation Solution
An identification apparatus and method that utilize a learnable learning model updated by an objective function based on the relevance between an accuracy index and a time index, optimizing both precision and time through a balanced evaluation curve.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the learning model is optimized to improve identification precision, then the accuracy of class identification is improved, but the time required for identification increases
Solution Approach 1:
The patent introduces a dynamic early classification mechanism that adjusts the classification process based on the input data's characteristics. The system dynamically determines whether to perform full classification or early classification by evaluating the input data against predetermined criteria, making the identification time flexible rather than fixed. This resolves the contradiction by allowing the system to adapt its processing depth to each input, achieving high precision when needed while reducing time when possible.
Solution Approach 2:
The patent changes the parameter of classification depth from a fixed value to a variable that can take different states (full classification or early classification). By introducing a classification determination unit that evaluates input data characteristics and selects appropriate classification depth, the system can optimize the balance between precision and time based on data-specific parameters rather than using a one-size-fits-all approach.
2Loss of time
If the identification process is simplified to reduce time, then the identification time is reduced, but the precision of identification deteriorates
Solution Approach 1:
The system dynamically selects between early classification and full classification based on input data evaluation. When the determination unit assesses that early classification is sufficient for the given input, the system performs only early classification, achieving fast results. When full precision is needed, the system proceeds to full classification. This dynamic selection resolves the contradiction by matching the identification depth to the actual requirements of each input case.
Solution Approach 2:
The patent implements partial classification (early classification) as a sufficient action for certain input types, rather than always performing complete classification. The determination unit identifies cases where partial classification is adequate, allowing the system to perform less processing than the maximum possible, thereby reducing time without sacrificing the precision needed for that specific input.
3Device complexity
If separate loss functions are used for precision and time, then the optimization is simpler, but the balance between precision and time is insufficient
Solution Approach 1:
The patent segments the classification process into distinct stages: early classification, determination of whether early classification is sufficient, and full classification if needed. This segmentation allows independent optimization of each stage while maintaining overall balance. The determination unit acts as a gatekeeper that separates cases suitable for fast processing from those requiring thorough processing, achieving reliable balance without complex unified optimization.
Data Source
AI summary
A learning apparatus includes: an identification unit that identifies a class of input data by using a learnable learning model; and an update unit that updates the learning model, by using an objective function based on relevance between a first index value for evaluating accuracy of a result of identification of the class of the input data and a second index value for evaluating time required to identify the class of the input data.