Embedded Classification Processing With Inverse-Transformed Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The computational strain caused by exponential computations in Softmax functions when implementing classification models on embedded devices with limited resources, leading to increased computational load and reduced recognition accuracy.
Innovation Solution
Implementing an inverse-transformable function to convert classification model outputs into non-normalized classification thresholds, reducing the need for exponential computations by using a second transform to derive a first classification threshold for classifying data into classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Softmax function is used to transform classification model output into classification probability values, then classification accuracy is improved, but computational load increases due to exponential computations
Solution Approach 1:
The patent extracts and removes the Softmax function from the classification pipeline, replacing it with a simple threshold comparison operation. This extraction eliminates the exponential computations while preserving the essential classification functionality through direct comparison of model outputs with pre-determined thresholds.
Solution Approach 2:
Instead of transforming model outputs into probability values using Softmax and then comparing with thresholds, the patent inverts the approach by directly comparing raw model outputs with transformed thresholds. This reversal eliminates the need for exponential computation while maintaining classification accuracy.
2Loss of information
If Softmax function is applied to normalize classification outputs, then probability interpretation is improved, but device complexity increases due to additional computation steps
Solution Approach 1:
The Softmax normalization step is extracted and removed from the system. The patent demonstrates that probability interpretation can be achieved through alternative means that do not require the Softmax function, thereby reducing computational steps and device complexity.
Solution Approach 2:
The patent performs preliminary transformation of classification thresholds during the training phase, storing these transformed thresholds for direct comparison during inference. This preliminary action eliminates the need for runtime probability normalization, reducing device complexity while preserving interpretability.
3Measurement precision
If classification probability values are computed for each unit class, then classification precision is improved, but processing time increases due to multiple exponential computations
Solution Approach 1:
The patent removes the time-consuming Softmax computation step that calculates probability values for each unit class. By extracting this step and replacing it with direct threshold comparison operations, the system maintains classification precision while dramatically reducing processing time.
Solution Approach 2:
Classification thresholds are pre-computed and stored during the training phase, eliminating the need for runtime probability calculations. This preliminary action allows for rapid inference by directly comparing model outputs with pre-determined thresholds, reducing processing time while maintaining precision.
Data Source
AI summary
An information processing device includes a processor. The processor obtains a first classification threshold for classifying data into at least one of a plurality of classes, and outputs a classification result of classifying the data into at least one of the plurality of classes based on an output of a trained classification model and the first classification threshold. The first classification threshold is obtained by a second transform performed on a second classification threshold, the second transform being an inverse transform of a first transform. The first transform corresponds to transforming the output of the trained classification model into a classification probability value of each of a plurality of unit classes constituting the plurality of classes. The second classification threshold is set based on the classification probability values of the plurality of unit classes.