Cluster-Trained Machine Learning for Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial neural networks used in image processing applications are computationally expensive, consuming large processing power and memory, and often require a trade-off between accuracy and efficiency.
Innovation Solution
The method involves training machine learning tools to associate images with clusters of labels or categories, using pseudolabels to reduce the number of computations needed for image classification, by grouping labels that are commonly confused with each other and assigning a pseudolabel to these clusters, allowing for subsequent training to identify images within these clusters efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional deep neural networks are used for image processing, then accuracy and confidence in image classification are improved, but computational cost and processing time increase significantly
Solution Approach 1:
The patent segments the image classification task into multiple stages: first using a computationally efficient machine learning tool to identify candidate labels, then using a deep neural network only to refine the classification among those candidates. This segmentation reduces the overall computational burden while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary filtering using a machine learning tool trained to identify candidate labels from training data before applying the computationally expensive deep neural network. This preliminary action reduces the search space and computational requirements for the subsequent classification step.
2Measurement precision
If larger and deeper neural networks are used, then classification accuracy is improved, but memory consumption and processing resources increase
Solution Approach 1:
The patent divides the classification system into two components: a simple machine learning tool for candidate identification and a deep neural network for final classification. This segmentation allows the use of complex network architecture only where necessary, reducing overall system complexity.
Solution Approach 2:
The patent introduces an intermediary machine learning tool that acts as a bridge between the input image and the deep neural network. This intermediary filters and prepares candidate labels, reducing the burden on the deep neural network and simplifying the overall system architecture.
3Measurement precision
If more computational resources are allocated to image processing, then processing accuracy is improved, but energy consumption and processing cost increase
Solution Approach 1:
The patent performs preliminary filtering using a computationally efficient machine learning tool before applying the energy-intensive deep neural network. This preliminary action reduces the amount of computational work required in subsequent steps, lowering overall energy consumption.
Solution Approach 2:
The patent applies the deep neural network only partially - specifically, only for refining classification among candidate labels identified by the machine learning tool, rather than applying it to all possible classification tasks. This partial application reduces energy consumption while maintaining necessary accuracy.
Data Source
AI summary
Image classification and related imaging tasks performed using machine learning tools may be accelerated by using one or more of such tools to associate an image with a cluster of such labels or categories, and then to select one of the labels or categories of the cluster as associated with the image. The clusters of labels or categories may comprise labels that are mutually confused for one another, e.g., two or more labels or categories that have been identified as associated with a single image. By defining clusters of labels or categories, and configuring a machine learning tool to associate an image with one of the clusters, processes for identifying labels or categories associated with images may be accelerated because computations associated with labels or categories not included in the cluster may be omitted.