Neural Network Training Across Local and Cloud Computing Nodes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing artificial intelligence model training methods face challenges in balancing security and performance, with cloud-based training risking data leakage and on-premise training using low-performance GPUs, while lacking integrated user interfaces for efficient management across environments.

Innovation Solution

A method and user interface for dynamically allocating tasks between local and cloud computing resources, allowing flexible training based on idle resources and user access rights, using a continuously updated UI/UX for efficient task management.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If training is performed in cloud environment, then computing performance is improved, but data security deteriorates

Engineering Contradiction:
Improvecomputing performanceVSAvoiddata security
Core Design Contradiction:
PowerVSReliability

Solution Approach 1:

The patent segments the training task into multiple sub-tasks and distributes them across multiple computing nodes (both cloud and on-premises). This segmentation allows the system to leverage cloud computing power for performance while keeping sensitive data processing distributed and isolated across multiple nodes, reducing the security risk concentration.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local quality by allowing different computing nodes to have different roles and characteristics. On-premises nodes handle data-sensitive operations with high security requirements, while cloud nodes handle computation-intensive tasks. This differentiation enables the system to optimize for both performance and security in appropriate locations.

Inventive Principle:
Principle #3Local quality

2Reliability

If training is performed in on-premises environment, then data security is improved, but computing performance deteriorates

Engineering Contradiction:
Improvedata securityVSAvoidcomputing performance
Core Design Contradiction:
ReliabilityVSPower

Solution Approach 1:

The patent merges on-premises and cloud computing environments into a unified training system. By combining the security advantages of on-premises infrastructure with the performance advantages of cloud resources, the system achieves both data security and high computing performance that neither environment could provide alone.

Inventive Principle:
Principle #5Merging (Combining)

3Ease of operation

If separate management is performed for each site, then local control is improved, but system complexity deteriorates

Engineering Contradiction:
Improvelocal controlVSAvoidsystem complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent implements a universal management platform that can manage both on-premises and cloud computing nodes through a single interface. This multi-functional system eliminates the need for separate management tools for each site, reducing operational complexity while maintaining the ability to control local resources and cloud resources uniformly.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent introduces a central management platform as an intermediary between users and the distributed computing nodes. This mediator handles task allocation, resource management, and coordination across different environments, simplifying the user experience while managing the underlying complexity of the distributed system.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4679270A1Method and user interface for training artificial neural network models in environment including multiple computing nodes
Publication Date: 2026.01.14 AIVEX CO LTD
  • EP4679270A1 patent drawingFigure 1
  • EP4679270A1 patent drawingFigure 2
  • EP4679270A1 patent drawingFigure 3

AI summary

Disclosed is a method training an artificial neural network model performed by a computing device according to an exemplary embodiment of the present disclosure. According to the present disclosure, the computing device identifies a first computing node and a second computing node for training an artificial neural network model, acquires information related to a first task for training the artificial neural network model, based on a user input, and dynamically allocates the first task to at least one of the first computing node or the second computing node, and the first computing node includes one or more local computing resources of the user and the second computing node includes one or more cloud computing resources.