In-Field AI/ML Model Validation via Reference Signals

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current testing procedures for AI/ML-enabled features in wireless communication are inadequate, as they assume static models, fail to account for continuous updates, and struggle with compatibility and interoperability between devices and network equipment, especially in scenarios involving two-sided models where changes can cause incompatibility.

Innovation Solution

Implement a validation process that involves triggering validation requests for AI/ML models in standby mode, transmitting reference signals, and receiving reports to validate functionality within the network or device, using a validation entity to compare results with expected outcomes, allowing for in-field validation and verification of AI/ML models and features without full re-testing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If AI/ML models are updated continuously in the field, then model adaptability and performance improve, but compatibility and interoperability between devices and network equipment deteriorate

Engineering Contradiction:
Improvemodel adaptabilityVSAvoidcompatibility
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where validation results from in-field testing are transmitted back to the network entity. This feedback loop enables the system to verify model updates before full deployment, ensuring compatibility is maintained while allowing continuous improvement. The validation entity uses reference signals to test updated models and provides feedback on whether they meet interoperability requirements.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent performs preliminary validation of AI/ML model updates before they are fully deployed across the network. By conducting in-field validation tests using reference signals and comparing results against expected outcomes, the system ensures that updated models maintain compatibility with network equipment and other devices before widespread implementation.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If comprehensive re-testing is performed for every AI/ML model update, then validation accuracy improves, but validation time and network disruption increase

Engineering Contradiction:
Improvevalidation accuracyVSAvoidvalidation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements partial validation by transmitting specific reference signals designed to test critical functionality rather than performing exhaustive comprehensive testing. This approach validates the most important aspects of model performance and compatibility while significantly reducing validation time and network disruption compared to full re-testing.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The validation process is segmented into specific testable components using reference signals. Instead of treating model validation as a monolithic process, the system divides it into discrete measurements of specific functions and performance metrics, allowing efficient validation of critical aspects without requiring complete re-testing of all model functionalities.

Inventive Principle:
Principle #1Segmentation

3Reliability

If in-field validation is implemented for AI/ML models, then model performance in live conditions is ensured, but system complexity and validation overhead increase

Engineering Contradiction:
Improvemodel performanceVSAvoidvalidation system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces a validation entity as an intermediary component that coordinates the in-field validation process. This intermediary manages the complexity by handling reference signal transmission, result collection, and comparison against expected outcomes, thereby ensuring model performance validation without requiring each device to implement complex validation logic independently.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4447563A1Post-deployment in-field validation of ai/ML enabled features
Publication Date: 2024.10.16 NOKIA TECHNOLOGIES OY
  • EP4447563A1 patent drawingFigure 1
  • EP4447563A1 patent drawingFigure 2
  • EP4447563A1 patent drawingFigure 3

AI summary

An apparatus configured to: determine that a validation of at least one second functionality has been triggered, wherein a first functionality is in an active mode, wherein the at least one second functionality is in a standby mode; transmit a request for the validation of the at least one second functionality; receive a plurality of reference signals; determine a result of the at least one second functionality based, at least partially, on at least one reference signal of the plurality of reference signals; and provide the result of the at least one second functionality for the validation of the at least one second functionality.