Neural Intent Translation for Adaptive Network Slice Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing network slice selection methods struggle to accurately translate applications' intents due to differences in vocabulary sets between applications and networks, leading to inefficient network resource allocation and limited scalability.

Innovation Solution

Utilizing neural networks on both mobile devices and network nodes to translate application intents into network slice selection requests using different vocabulary sets, with reinforcement learning for feedback-based protocol development.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional network slice selection methods are used, then network resource allocation is performed, but translation accuracy of application intents deteriorates due to vocabulary set differences

Engineering Contradiction:
Improveintent translation accuracyVSAvoidvocabulary set compatibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent introduces neural network models as intermediary translators between applications and network nodes. These models learn to map application-specific vocabulary sets to network-specific vocabulary sets through training, enabling accurate intent translation without requiring direct compatibility between different vocabulary systems. The neural networks act as adaptive mediators that bridge the semantic gap between application intents and network slice selection criteria.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms the static vocabulary mapping problem into a dynamic parameter optimization problem. By training neural network models with learnable parameters (weights and biases), the system adapts the translation process to match specific application requirements and network conditions. The parameters are adjusted during training to minimize translation errors and improve intent interpretation accuracy across different vocabulary sets.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If pre-defined communication protocols are used, then communication between applications and networks is established, but scalability deteriorates when new application types emerge

Engineering Contradiction:
Improveprotocol compatibilityVSAvoidsystem scalability
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent replaces static pre-defined protocols with dynamic neural network-based translation systems. The neural networks are trained on specific application data and adapt their translation behavior based on learned patterns, enabling the system to handle new application types without requiring pre-defined protocol specifications. This dynamic adaptation mechanism allows the network to evolve and support emerging applications while maintaining communication effectiveness.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system enables applications to serve themselves by learning their own intent representations through neural networks. Each application can be trained to translate its specific vocabulary to network vocabulary, allowing new applications to integrate autonomously without requiring manual protocol definition or configuration. This self-service capability dramatically improves scalability by eliminating the need for centralized protocol management.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If neural network translation is implemented, then intent translation accuracy improves, but computational complexity increases

Engineering Contradiction:
Improveintent translation accuracyVSAvoidneural network implementation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the translation system into separate neural network components: application-side models for intent encoding and network-side models for slice selection decoding. This segmentation allows each component to be optimized independently, reducing the computational burden on any single device. The modular architecture enables distributed training and deployment, making the complex translation task more manageable and scalable across different system components.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4583566A1Method, apparatus and computer program
Publication Date: 2025.07.09 NOKIA SOLUTIONS & NETWORKS OY
  • EP4583566A1 patent drawingFigure 1
  • EP4583566A1 patent drawingFigure 2
  • EP4583566A1 patent drawingFigure 3

AI summary

There is provided a method performed by a mobile device, the method comprising: receiving, from an application running on the mobile device, an input defining an intent of the application, the intent being expressed using a first vocabulary set; generating, using a first neural network and based on the intent, a message for requesting a network slice selection for the application, the message being expressed using a second vocabulary set; sending the generated message to a network node; receiving, from the network node, information indicating a network slice selected for use by the application; determining whether the selected network slice satisfies the intent; and updating the first neural network based on the result of the determining.