Neural Network Meta-Training for Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network models face challenges in accurately recognizing input data with diverse variations due to limited training data, leading to lower recognition accuracy, and traditional generative models often generate irrelevant images during training, rather than enhancing recognition performance.
Innovation Solution
The proposed method involves meta-training a neural network with a feature extractor, feature sample generator, and classifier using diverse training class data to extract and propagate feature vectors, and then performing class vector training to reduce the distance between class vectors and input data features, using a loss function that includes a softmax function and exponential distance ratios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional generative models are used for data augmentation, then training data volume increases, but recognition accuracy deteriorates due to generation of irrelevant images
Solution Approach 1:
The patent introduces an intermediary mechanism (feature space mapping and class vector guidance) between data generation and recognition training. Instead of directly generating images, the system generates feature vectors that are intermediaries between raw data and the recognition model, ensuring generated samples align with target class characteristics and improve rather than degrade recognition accuracy
Solution Approach 2:
The system changes the parameter space from image pixel space to feature vector space. By operating in the feature space defined by pre-trained extractors and guided by class vectors, the system transforms the data generation process to produce samples with controlled feature characteristics that directly enhance recognition performance
2Productivity
If limited training data is used, then training time and computational resources are reduced, but recognition accuracy deteriorates due to insufficient coverage of data variations
Solution Approach 1:
The system performs preliminary action by pre-training feature extractors and establishing class vectors before the main recognition training. This preliminary setup creates a structured feature space and class prototypes that guide subsequent training with limited data, enabling the model to achieve high accuracy even with small training sets by leveraging the pre-established feature representations
Solution Approach 2:
The system creates synthetic copies of training data through feature vector manipulation and generation. By copying and transforming feature vectors to generate additional training samples that maintain the statistical properties of the original data, the system effectively increases training data volume without requiring additional physical samples, thereby improving recognition accuracy with limited resources
Data Source
AI summary
A processor-implemented neural network method includes: extracting, by a feature extractor of a neural network, a plurality of training feature vectors corresponding to a plurality of training class data of each of a plurality of classes including a first class and a second class; determining, by a feature sample generator of the neural network, an additional feature vector of the second class based on a mean vector and a variation vector of the plurality of training feature vectors of each of the first class and the second class; and training a class vector of the second class included in a classifier of the neural network based on the additional feature vector and the plurality of training feature vectors of the second class.