ML Stack Smart Engine Resource Management

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Machine learning processes on wireless communication devices consume resources such as battery power, processor load, and memory, leading to performance issues and user experience degradation, and managing native libraries for multiple machine learning services is complex and storage-intensive.

Innovation Solution

A machine learning stack with a smart engine that manages resource usage by selecting appropriate machine learning models and scheduling processes based on device status, modularizing services, and using masked user data for secure authentication, allowing local, remote, or hybrid processing options.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If machine learning processes are executed on wireless communication devices, then pattern recognition and decision-making capabilities are improved, but device resources such as battery power, processor load, and memory are depleted

Engineering Contradiction:

Improvemachine learning decision-making capabilityVSAvoidbattery power consumption

Core Design Contradiction:

Extent of automationVSUse of energy by moving object

Solution Approach 1:

The system segments machine learning workloads between the wireless communication device and external servers. The device executes only essential local inference tasks while offloading computationally intensive training and complex analysis to remote servers, thereby maintaining automation capabilities while reducing local energy consumption and resource usage.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If multiple machine learning services are deployed on wireless communication devices, then service functionality and versatility are improved, but device complexity and storage requirements increase

Engineering Contradiction:

Improvemachine learning service functionalityVSAvoidnative library management complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system implements a universal machine learning stack that provides multi-functional capabilities through a single integrated framework. This stack includes common native libraries and execution environments that can serve multiple different machine learning services, reducing the need for separate implementations and minimizing overall device complexity while maintaining service versatility.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Speed

If machine learning computations are performed locally on wireless communication devices, then response time and user experience are improved, but device performance and stability deteriorate due to resource constraints

Engineering Contradiction:

Improvemachine learning inference speedVSAvoiddevice performance stability

Core Design Contradiction:

SpeedVSReliability

Solution Approach 1:

The system dynamically adjusts the balance between local and remote machine learning execution based on real-time device conditions such as available memory, processor load, and battery status. When resources are sufficient, more computations are performed locally for faster response; when resources are constrained, the system offloads tasks to maintain stability, thereby optimizing both speed and reliability adaptively.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS11513583B2Methods and systems for managing machine learning involving mobile devices

Publication Date: 2022.11.29 VERIZON PATENT & LICENSING INC

AI summary

A computer device may include a memory configured to store instructions and a processor configured to execute the instructions to determine a device status associated with the wireless communication device and determine that a machine learning process is to be performed based on the determined device status. The processor may be further configured to execute the instructions to select a machine learning model based on the determined device status; select one or more data inputs based on the determined device status; and perform the machine learning process using the selected machine learning model and the selected one or more data inputs.