Split Neural Network Training for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training neural networks, such as deep face recognition networks, require large amounts of data, computational resources, and time, making them inefficient and impractical for use on devices with limited resources.
Innovation Solution
The training process is split into component neural networks, where each network is trained separately to determine specific transformations (e.g., alignment and representation models) before being combined, reducing the need for extensive resources and data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is trained for object recognition, then recognition accuracy can be achieved, but training requires large amounts of data, computational resources, and time
Solution Approach 1:
The patent divides the single neural network training task into two separate neural networks: a first neural network for alignment determination and a second neural network for object recognition. This segmentation allows each network to be trained independently with specialized focus, reducing the overall training complexity and resource requirements while maintaining or improving recognition accuracy through the combined system
2Productivity
If training data and computational resources are reduced, then training becomes practical for devices with limited resources, but training efficiency and effectiveness improve
Solution Approach 1:
By segmenting the training task into two separate neural networks with distinct functions (alignment and recognition), the patent enables more efficient training processes. Each network can be trained with optimized data subsets and computational configurations, improving overall training efficiency while requiring less total training data compared to training a single comprehensive network
Solution Approach 2:
The first neural network performs preliminary alignment determination on input images before the second neural network performs object recognition. This preliminary action of aligning images beforehand improves the effectiveness of the subsequent recognition process, allowing the system to achieve better training efficiency and accuracy with reduced computational resources during the main recognition task
Data Source
AI summary
Technical solutions are described for training an object-recognition neural network that identifies an object in a computer-readable image. An example method includes assigning a first neural network for determining a visual alignment model of the images for determining a normalized alignment of the object. The method further includes assigning a second neural network for determining a visual representation model of the images for recognizing the object. The method further includes determining the visual alignment model by training the first neural network and determining the visual representation model by training the second neural network independent of the first. The method further includes determining a combined object recognition model by training a combination of the first neural network and the second neural network. The method further includes recognizing the object in the image based on the combined object recognition model by passing the image through each of the neural networks.