Distributed Machine-Learned Model Segmentation for Edge Inference

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine-learned models deployed on edge devices with limited computational resources face challenges in processing complex tasks due to resource constraints, while remote deployment raises privacy concerns and bandwidth issues from transmitting raw sensor data.

Innovation Solution

A method for dynamically distributing and reallocating portions of a machine-learned model across multiple interactive objects based on their resource availability, allowing for efficient processing and inference generation using sensor data from multiple devices while maintaining data privacy and reducing bandwidth usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a machine-learned model is deployed at an edge device, then raw sensor data does not need to be transmitted to a remote computing device, but the edge device has limited computational resources that are inadequate for deploying complex machine-learned models

Engineering Contradiction:

Improvedata privacyVSAvoidcomputational resources

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The machine-learned model is divided into multiple portions that are distributed across different interactive objects in the network. Each interactive object executes a specific portion of the model locally, eliminating the need to transmit raw sensor data to a central remote device while still enabling complex processing capabilities through the distributed model execution.

Inventive Principle:

Principle #1Segmentation

2Productivity

If a machine-learned model is deployed at a remote computing device, then processing capabilities are sufficient, but raw sensor data must be transmitted from the edge device which raises privacy concerns and bandwidth considerations

Engineering Contradiction:

Improveprocessing capabilitiesVSAvoidbandwidth usage

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The model is segmented and distributed across the network of interactive objects, allowing processing to occur locally at each device rather than requiring transmission of raw sensor data to a remote computing device. This maintains sufficient processing capabilities while eliminating bandwidth consumption for data transmission.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The distributed model portions at each interactive object act as intermediaries that process sensor data locally before any potential transmission of processed results. This intermediary processing layer eliminates the need to transmit raw sensor data, addressing both privacy concerns and bandwidth limitations.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If a machine-learned model is deployed at a remote computing device, then processing capabilities are sufficient, but transmitting sensor data from the edge device raises privacy concerns

Engineering Contradiction:

Improveprocessing capabilitiesVSAvoiddata privacy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

By segmenting the model and distributing it across interactive objects, the system enables complex processing capabilities to be executed locally at the edge devices rather than requiring transmission of sensor data to a remote device. This segmentation approach maintains processing sufficiency while keeping data processing local, thereby addressing privacy concerns.

Inventive Principle:

Principle #1Segmentation

4Productivity

If edge devices execute complex machine-learned models, then processing capabilities are sufficient, but power consumption increases which may be insufficient to support large processing operations

Engineering Contradiction:

Improveprocessing capabilitiesVSAvoidpower consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The model is divided into portions that are distributed across multiple interactive objects, allowing processing capabilities to be aggregated across the network without requiring any single edge device to execute the entire complex model. This segmentation enables sufficient processing capabilities while distributing power consumption across multiple devices, preventing any single device from consuming excessive power.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20230061808A1Distributed Machine-Learned Models Across Networks of Interactive Objects

Publication Date: 2023.03.02 GOOGLE LLC

AI summary

A set of interactive objects can implement a machine-learned model for monitoring an activity while communicatively coupled over one or more networks. The machine-learned model can be configured to generate data indicative of at least one inference associated with the activity based at least in part on sensor data associated with two or more interactive objects of the set of interactive objects. The computing system can determine for each interactive object a respective portion of the machine-learned model for execution by the interactive object during at least a portion of the activity. The computing system can generate for each interactive object configuration data indicative of the respective portion of the machine-learned model for execution by the interactive object during the portion of the activity. The computing system can communicate the configuration data indicative of the respective portion of the machine-learned model for execution by to each interactive object.