Feature Extraction for ML Model Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine-learning assisted code drafting technologies require the service provider to have direct access to the client's large and proprietary codebase, leading to burdensome data transfer, privacy risks, and incomplete model training due to incomplete or outdated data.

Innovation Solution

Allowing clients to extract and transmit only a limited set of data features from their codebase to the service provider, enabling the creation of custom machine learning models without the need for the service provider to access the entire codebase, thus reducing network resource usage and enhancing data security.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the service provider has direct access to the client's complete codebase, then the machine learning model can be trained on comprehensive data, but the data transfer burden increases and privacy risks arise

Engineering Contradiction:
Improvemodel training completenessVSAvoiddata transfer volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential features and metadata from the client's codebase rather than transferring the complete codebase. The feature extraction component identifies and extracts relevant characteristics (e.g., code structure, dependencies, configuration) while leaving the proprietary code on the client's system, thus reducing data transfer volume while maintaining model training effectiveness

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The codebase is segmented into essential features and non-essential proprietary content. Only the essential features needed for model training are extracted and transmitted to the service provider, while the rest remains on the client's system. This segmentation allows selective data transfer that balances model training completeness with data privacy and transfer efficiency

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If the service provider has direct access to the client's codebase, then complete models can be trained, but the client's proprietary data is at risk

Engineering Contradiction:
Improvemodel training completenessVSAvoiddata leakage risk
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only the essential features and metadata from the client's codebase rather than transferring the complete codebase. The feature extraction component identifies and extracts relevant characteristics (e.g., code structure, dependencies, configuration) while leaving the proprietary code on the client's system, thus reducing data transfer volume while maintaining model training effectiveness

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The feature extraction component acts as an intermediary that processes the client's proprietary codebase locally and transforms it into anonymized features. This intermediary layer prevents direct exposure of proprietary code to the service provider while still enabling model training on derived features, thus mitigating data leakage risk

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If the client transmits the entire codebase, then comprehensive model training is achieved, but network resources are inefficiently used

Engineering Contradiction:
Improvemodel training completenessVSAvoidnetwork resource consumption
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The patent extracts only the essential features and metadata from the client's codebase rather than transferring the complete codebase. The feature extraction component identifies and extracts relevant characteristics (e.g., code structure, dependencies, configuration) while leaving the proprietary code on the client's system, thus reducing data transfer volume while maintaining model training effectiveness

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of transmitting the entire codebase (excessive action), the system transmits only the necessary features and metadata (partial action) required for model training. This partial transmission approach reduces network resource consumption while still providing sufficient data for effective model training

Inventive Principle:
Principle #16Partial or excessive action

4Ease of manufacture

If the client configures the service provider's service to build their codebase, then feature extraction can be performed, but the configuration process becomes burdensome

Engineering Contradiction:
Improvefeature extraction capabilityVSAvoidconfiguration complexity
Core Design Contradiction:
Ease of manufactureVSEase of operation

Solution Approach 1:

The feature extraction component runs on the client's own system using the client's existing build tools and environment. This self-service approach eliminates the need for the client to configure the service provider's system, as the extraction process leverages the client's own development environment to build and analyze their codebase locally

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3884373B1Extracting program features for assisting software development
Publication Date: 2024.03.13 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3884373B1 patent drawingFigure 1
  • EP3884373B1 patent drawingFigure 2
  • EP3884373B1 patent drawingFigure 3

AI summary

Improving the results and process of machine learning service in computer program development. A clients codebase is accessed. A set of features are extracted from the clients codebase. One or more features from the set of features are then selected. Thereafter, at least one of the selected features is sent to a machine learning service that uses the received feature(s) to build custom model(s) for the client's computer system.