End-to-End DNN Reconfiguration for Mobile Edge Computing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The complexity of managing edge computing resources is increased by device mobility in wireless networks, as UEs move between coverage areas with and without edge compute servers, requiring dynamic redirection of application processing.

Innovation Solution

Adapting end-to-end deep neural networks (DNNs) to dynamically reconfigure based on changes in edge compute server participation, optimizing processing for edge or cloud-based computing modes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the wireless network dynamically redirects application processing between edge compute servers and cloud-based services to accommodate device mobility, then the system maintains service availability, but the complexity of managing edge computing resources increases

Engineering Contradiction:
Improveservice availabilityVSAvoidcomplexity of managing edge computing resources
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements dynamic DNN adaptation where the neural network configuration is automatically adjusted based on real-time participation mode changes of edge compute servers. The system transitions from static edge computing configurations to dynamic reconfiguration, allowing the DNN architecture, layers, and parameters to adapt automatically when UEs move between coverage areas with and without ECS support, thereby maintaining service availability while reducing management complexity

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes key parameters of the DNN including architecture, number of layers, and computational parameters based on the participation mode of edge compute servers. When a UE moves from an area with ECS support to without, the system automatically modifies DNN parameters to transition from edge-based processing to cloud-based processing, and vice versa, enabling seamless service continuity without manual intervention

Inventive Principle:
Principle #35Parameter changes

2Loss of time

If the system uses edge compute servers for application processing to reduce data transfer latency, then the network response time improves, but the system requires dynamic redirection when UEs move between coverage areas

Engineering Contradiction:
Improvedata transfer latencyVSAvoiddynamic redirection complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The system enables self-service through automatic DNN reconfiguration triggered by participation mode changes of edge compute servers. When UEs move between coverage areas, the system automatically detects the change in ECS availability and reconfigures the appropriate DNNs without requiring manual intervention or complex redirection protocols, thereby maintaining low latency while simplifying the redirection process

Inventive Principle:
Principle #25Self-service

3Power

If the wireless network incorporates additional computing devices and data storage resources to accommodate increased data usage, then the computing power increases, but the complexity of managing and redirecting application processing increases

Engineering Contradiction:
Improvecomputing powerVSAvoidcomplexity of managing and redirecting application processing
Core Design Contradiction:
PowerVSDevice complexity

Solution Approach 1:

The patent creates a universal DNN framework that can operate in multiple modes (edge-based and cloud-based processing) through a single adaptable architecture. The same DNN infrastructure serves both edge computing scenarios with low latency requirements and cloud computing scenarios with higher latency tolerance, eliminating the need for separate management systems for different computing modes and reducing overall system complexity

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

Data Source

PatentEP4140170B1End-to-end deep neural network adaptation for edge computing
Publication Date: 2025.11.12 GOOGLE LLC
  • EP4140170B1 patent drawingFigure 1
  • EP4140170B1 patent drawingFigure 2
  • EP4140170B1 patent drawingFigure 3

AI summary

Techniques and apparatuses are described for adapting an end-to-end, E2E, machine-learning, ML, configuration for processing communications transferred through an E2E communication. A network entity directs (805) a user equipment (UE) and a base station participating in the E2E communication to implement the E2E communication by forming at least a portion of an E2E deep neural network, DNN, based on a first E2E ML configuration. The network entity determines (815) to update the first E2E ML configuration based on a change in a participation mode of an edge compute server (ECS) in the E2E communication. The network entity identifies (820) a second E2E ML configuration based on the change in participation mode and directs (825) the UE or the base station to update the portion of the E2E DNN using the second E2E ML configuration.