Neural Network Data Synthesis for Lower Edge Storage and Transmission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing environments face heavy burdens on network bandwidth and storage due to the transmission, storage, and analysis of large or complex data generated at edges, which is inefficient and resource-intensive.
Innovation Solution
Implementing a neural network, such as a generative adversarial network (GAN), at a central hub to generate synthetic data that is statistically similar to edge data, using a detector to suppress actual data transmission and storage by validating the generator's ability to replicate edge data accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If actual edge data is transmitted to the hub for storage and analysis, then query accuracy is maintained, but network bandwidth and storage resources are heavily consumed
Solution Approach 1:
The patent applies the copying principle by training a generator neural network to create synthetic data that replicates the statistical properties and patterns of actual edge data. The generator produces copies that are statistically indistinguishable from real data, allowing queries to be answered using synthetic copies rather than transmitting and storing voluminous actual data at the hub.
Solution Approach 2:
The patent introduces a detector neural network as an intermediary component that validates whether synthetic data accurately represents actual data characteristics. The detector acts as a mediator between the generator and the query system, ensuring that synthetic data maintains the necessary statistical properties to answer queries accurately without requiring actual data transmission.
2Productivity
If actual edge data is stored at the hub, then comprehensive analysis is enabled, but storage requirements and processing time increase significantly
Solution Approach 1:
The generator neural network creates synthetic data copies that preserve the statistical properties and patterns of actual edge data. These synthetic copies enable comprehensive analysis at the hub without requiring storage of voluminous actual data, thus maintaining productivity while reducing storage requirements.
Solution Approach 2:
The patent transforms actual data into synthetic data by changing its representation parameters through neural network generation. The synthetic data maintains essential statistical parameters and patterns needed for analysis while occupying minimal storage space, enabling comprehensive analysis capability with reduced data volume.
3Quantity of substance
If synthetic data is used to respond to queries, then data transmission and storage are reduced, but data accuracy and reliability may be compromised
Solution Approach 1:
The detector neural network provides feedback mechanisms to validate synthetic data quality. The detector continuously monitors whether synthetic data maintains statistical fidelity to actual data, and this feedback loop ensures that only accurate synthetic data representations are used to respond to queries, maintaining reliability while reducing data transmission.
4Productivity
If a neural network is trained at the hub using edge data, then synthetic data generation capability is achieved, but initial data transmission and computational resources are consumed
Solution Approach 1:
The patent applies preliminary action by training the generator neural network at the hub using initial edge data transmission before operational use. This preliminary training phase establishes the synthetic data generation capability, after which the hub can generate synthetic data locally without requiring continuous data transmission from edges, achieving long-term productivity with minimal initial resource consumption.
Data Source
AI summary
A hub of a computing environment obtains a training set from an edge of the computing environment. The training set that is obtained includes data from the edge and is used to train a neural network at the hub. The training of the neural network provides a detector and a generator at the hub. A determination is made as to whether the training of the neural network is complete. Based on determining that the training of the neural network is complete, the detector is sent to the edge. The detector at the edge is to facilitate suppression of additional edge data to the hub based on the detector at the edge determining that the additional edge data is statistically similar, based on one or more selected criteria, to data used to train the neural network.