Federated Automatic Machine Learning Pipeline
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated machine learning systems are not natively amenable to automatic machine learning due to data segregation, leading to high complexity and long training times, which can impact party resources and reliability, and pose challenges in integrating with automatic machine learning systems while ensuring data privacy.
Innovation Solution
A method for federated automatic machine learning that distributes a search process across multiple parties to find the best pipeline configuration, with each party evaluating model configurations on local datasets and sending scores back to a server for aggregation, allowing for the generation of a globally-trained model without compromising data security, using a search parameters aggregator to orchestrate the search process and distribute parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If federated machine learning systems use data segregation to ensure data privacy, then data security is improved, but the system complexity and training time increase significantly
Solution Approach 1:
The patent segments the automatic machine learning pipeline into distinct stages (estimator selection, hyper-parameter optimization, feature engineering) and distributes different segments across multiple parties. Each party performs specific pipeline segments locally on their data while a coordinator aggregates results, enabling data segregation without requiring complete isolation of all operations.
Solution Approach 2:
The patent introduces a coordinator as an intermediary component that manages communication between parties, distributes pipeline configurations, collects evaluation results, and aggregates models. This intermediary layer simplifies the complexity of coordinating federated operations by centralizing management functions while maintaining data segregation at individual parties.
2Reliability
If federated machine learning systems maintain data segregation, then data privacy is improved, but training time and computational resources increase
Solution Approach 1:
By dividing the machine learning pipeline into sequential stages and assigning different stages to different parties, the system performs computations in parallel across multiple parties rather than sequentially. This segmentation enables simultaneous progress on different pipeline components while maintaining data privacy through distribution.
Solution Approach 2:
The patent implements preliminary actions by having parties pre-process and evaluate model configurations locally on their data before aggregation. Parties perform preliminary evaluations of pipeline configurations and prepare local models in advance, reducing the overall training time by avoiding redundant computations during the aggregation phase.
3Manufacturing precision
If automatic machine learning optimizes multiple pipeline stages sequentially, then model quality is improved, but computational efficiency decreases
Solution Approach 1:
The patent segments the sequential pipeline optimization process into independent stages that can be performed simultaneously by different parties. Estimator selection, hyper-parameter optimization, and feature engineering are divided into separate tasks that parties execute in parallel, maintaining model quality through comprehensive optimization while improving computational efficiency through concurrent processing.
Solution Approach 2:
The patent applies partial action by having parties perform only the specific pipeline stages relevant to their local data and expertise, rather than requiring all parties to complete the entire sequential pipeline. This allows each party to optimize their portion efficiently while the coordinator integrates results, achieving overall computational efficiency without sacrificing model quality.
Data Source
AI summary
Aspects of the invention include systems and methods configured for federated automatic machine learning. A non-limiting example computer-implemented method includes defining a search process including a model configuration for building an automatic machine learning pipeline definition and distributing the search process across a plurality of parties. Each member of the plurality of parties retains federated data including training data and holdout data. The method includes receiving, from each member of the plurality of parties, an evaluation result of the model configuration against respective holdout data and aggregating the received evaluation results to define aggregated parameters. A new pipeline definition is generated from the aggregated parameters and trained local models received from each member of the plurality of parties are aggregated to define an aggregated model. Each trained local model includes the new pipeline definition.