Incremental Real-Time Learning for Neural Network Data Tagging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional Deep Neural Networks require thousands or millions of iteration cycles for training, which is slow and prone to human error due to the tedious process of manual image tagging, making data preparation time-consuming and affecting the quality of subsequent learning.
Innovation Solution
A method and system that utilize a user interface and processors to learn and tag representations of objects in images, with a fast learning classifier that assists users by suggesting tags and positions, allowing for incremental and real-time updates, reducing manual labor and improving tagging speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual tagging is performed by human users, then tagging accuracy can be maintained through human judgment, but tagging speed becomes unacceptably slow and prone to human errors
Solution Approach 1:
The patent introduces an automated tagging system that acts as an intermediary between the image data and the final tagged output. This system uses machine learning models to perform the tagging function, eliminating the need for direct human manual tagging while maintaining acceptable accuracy through automated algorithms and iterative learning from user corrections.
Solution Approach 2:
The patent replaces the mechanical human manual tagging process with an automated computational system. Instead of humans manually examining and tagging images, the system uses algorithms and processing power to automatically generate tags, thereby dramatically increasing tagging speed while reducing human error.
2Reliability
If traditional deep neural networks are used for training, then comprehensive learning can be achieved, but the training process requires thousands or millions of iteration cycles making it time-consuming
Solution Approach 1:
The patent applies preliminary action by pre-tagging images using an automated system before formal training begins. This pre-tagging step creates an initial dataset that can be quickly processed, allowing the training process to start with already-labeled data rather than requiring extensive manual tagging and lengthy training iterations from scratch.
Solution Approach 2:
The patent implements continuous useful action through iterative learning where the system continuously refines its tagging capabilities. User corrections are immediately incorporated to improve the automated tagging system, creating a continuous feedback loop that progressively enhances learning quality without requiring repeated full training cycles.
Data Source
AI summary
Today, artificial neural networks are trained on large sets of manually tagged images. Generally, for better training, the training data should be as large as possible. Unfortunately, manually tagging images is time consuming and susceptible to error, making it difficult to produce the large sets of tagged data used to train artificial neural networks. To address this problem, the inventors have developed a smart tagging utility that uses a feature extraction unit and a fast-learning classifier to learn tags and tag images automatically, reducing the time to tag large sets of data. The feature extraction unit and fast-learning classifiers can be implemented as artificial neural networks that associate a label with features extracted from an image and tag similar features from the image or other images with the same label. Moreover, the smart tagging system can learn from user adjustment to its proposed tagging. This reduces tagging time and errors.