ML Model Feature Transformation via Intermediary Layer

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional software development platforms face challenges in integrating machine learning (ML) models due to complexities in feature identification, data transformation, and inconsistent feature selection between training and deployment, leading to slower adoption of ML technologies in applications.

Innovation Solution

A ML development platform that automates the training, deployment, and updating of ML models by providing tools for feature identification, data transformation, and consistent feature engineering across server and client sides, allowing for easier integration and maintenance of ML functionality within applications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If conventional software development platforms are used for ML integration, then developers can identify ML model types and determine features, but the platforms fail to provide access to requisite training data and require manual data collection and transformation

Engineering Contradiction:

Improveease of ML model integrationVSAvoiddata collection and transformation complexity

Core Design Contradiction:

Ease of manufactureVSDevice complexity

Solution Approach 1:

The patent introduces a data transformation layer that acts as an intermediary between raw collected data and ML model features. This layer automatically transforms collected data into the required feature format, eliminating the need for developers to manually construct transformation algorithms while maintaining data privacy and security.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The platform enables self-service data transformation where the system automatically identifies, retrieves, and transforms data into required features without developer intervention. The platform serves itself by providing training data access and feature transformation capabilities as built-in functionalities.

Inventive Principle:

Principle #25Self-service

2Adaptability or versatility

If developers manually construct algorithms to transform collected data into features, then they can create custom feature representations, but the process is time-consuming and error-prone

Engineering Contradiction:

Improvecustom feature transformation capabilityVSAvoidtime for feature transformation

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The platform performs preliminary data transformation actions by pre-processing and transforming collected data into required feature formats before the developer needs to use them. This eliminates the need for developers to spend time on manual transformation algorithms while maintaining the ability to handle custom transformation requirements.

Inventive Principle:

Principle #10Preliminary action

3Stability of the object's composition

If features are constructed with fixed definitions in program code, then feature consistency can be maintained, but changes to feature sets require code changes and validation deployment

Engineering Contradiction:

Improvefeature consistency between training and deploymentVSAvoidcode maintenance complexity

Core Design Contradiction:

Stability of the object's compositionVSDevice complexity

Solution Approach 1:

The patent segments the feature definition from the program code by using a separate feature store that maintains feature definitions independently. This allows feature changes to be made in the feature store without requiring code changes, while maintaining consistency between training and deployment through the centralized feature store.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The platform uses copying by creating a centralized feature store that serves as a reference copy for all feature definitions. This copy can be updated independently of the program code, allowing feature changes to be propagated without code modifications while maintaining consistency across training and deployment environments.

Inventive Principle:

Principle #26Copying

4Speed

If ML models are deployed with features from log files, then real-time prediction is possible, but the features must be transformed to match model input requirements

Engineering Contradiction:

Improveprediction speedVSAvoidfeature transformation complexity

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The feature store acts as an intermediary between log file data and ML model inputs, automatically transforming features from log file format to model-compatible format. This enables real-time prediction by pre-transforming features on demand without requiring complex transformation algorithms in the deployment code.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20230137131A1Computing system for training, deploying, executing, and updating machine learning models

Publication Date: 2023.05.04 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

A server computing device generates training data based upon an identifier for a device, a timestamp, and a label received from a developer computing device. The server computing device trains a computer-implemented machine learning (ML) model based upon the training data. The server computing device also generates client configuration data for the ML model that specifies transformations that are to be applied to values in order to generate input values for the ML model. The server computing device deploys ML assets to client computing devices, the ML assets comprising the ML model and the client configuration data. The client computing devices execute the ML model using input values derived via transformations of (local) values produced by the client computing devices and transmit telemetry data to the server computing device. The server computing device updates the ML assets based upon the telemetry data.