Image Recognition Using Segmented Feature Maps for Ensemble Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image recognition, achieving high accuracy and independence in inference results with a small amount of training data is challenging due to biases in training data, and existing methods struggle to effectively combine feature amounts from multiple inference devices for group learning.
Innovation Solution
An image recognition method that generates a base feature map group from an input image, uses multiple machine-learned inference devices to derive inference results, and integrates these results using a specific manner, such as majority decision or weighted averaging, to produce a final inference result, where each inference input includes some or all base feature maps that differ from others, allowing for high independence and accuracy even with limited training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of training data is used for machine-learning inference devices, then inference accuracy is improved, but data preparation complexity and cost increase
Solution Approach 1:
The training data is segmented into multiple independent feature maps, each representing different characteristics of the input image. Multiple inference devices are trained on these segmented feature maps rather than the complete dataset, allowing parallel processing and reducing the complexity of preparing and managing large unified training datasets while maintaining high inference accuracy through diversified feature learning
Solution Approach 2:
The approach transitions from training a single inference device on a large 2D dataset to training multiple inference devices on segmented feature maps across an additional dimension (number of devices). This dimensional transformation allows the system to achieve high accuracy through ensemble learning while reducing the data preparation burden on each individual device
2Reliability
If multiple inference devices are used for group learning, then independence of inference results is improved, but system complexity increases
Solution Approach 1:
The system segments the training process by assigning different feature maps to different inference devices, ensuring each device learns independent features. This segmentation maintains result independence while managing system complexity through structured organization of devices and their corresponding feature maps
Solution Approach 2:
Multiple inference devices are trained using the same base feature maps but process them differently to achieve both independence and universality. Each device serves multiple purposes: learning specific features independently while collectively contributing to a comprehensive recognition system, thereby managing complexity through functional integration
3Reliability
If multiple inference devices are used for group learning, then reliability of inference results is improved, but computational resources required increase
Solution Approach 1:
The computational workload is segmented across multiple inference devices, each processing specific feature maps. This segmentation distributes computational resources efficiently, reducing the energy consumption burden on any single device while maintaining high reliability through ensemble decision-making
Solution Approach 2:
Each inference device processes only a subset of feature maps relevant to its specific function rather than all feature maps. This partial action approach reduces redundant computations and energy consumption while the collective output of multiple devices achieves the reliability equivalent to processing complete data comprehensively
Data Source
AI summary
An image recognition method includes a feature amount extracting step of generating, from an input image, a base feature map group including a plurality of base feature maps; an inferring step of deriving a plurality of inference results using each of a plurality of machine-learned inference devices for a plurality of inference inputs based on the base feature map group; and an integrating step of integrating the plurality of inference results by a specific manner to derive a final inference result, where each of the plurality of inference inputs has some or all base feature maps of the plurality of base feature maps, and each of the plurality of inference inputs has the some or all base feature maps that are different in part or whole from the some or all base feature maps of another inference input in the plurality of inference inputs.