Machine Learning Model Architecture Adaptation for Heterogeneous Infrastructure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to efficiently operate in dynamic environments with varying infrastructure and feature sets, requiring labor-intensive adaptation and resulting in suboptimal performance due to limitations in current distributed learning techniques like federated learning and split learning.
Innovation Solution
A method where a leader computing device receives weights and model architecture information from worker nodes to determine a common portion of the ML model usable by all nodes, generating an updated global model that can adapt to changes in the environment, facilitating compatibility and dynamic adaptation across heterogeneous models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML models are trained for specific tasks using infrastructure data, then the models achieve high performance on specific tasks, but the models require labor-intensive adaptation for every execution environment and are not infrastructure agnostic
Solution Approach 1:
The patent implements infrastructure-agnostic ML models that can execute across diverse environments without task-specific retraining. The system uses a unified model architecture with dynamic feature adaptation, allowing a single model to serve multiple infrastructure types (cloud, edge, IoT devices) while maintaining high performance through environment-aware feature selection rather than task-specific customization
2Adaptability or versatility
If ML models are trained locally at worker nodes using locally available data, then the models adapt to local conditions, but the models vary between worker nodes and require extensive communication and coordination
Solution Approach 1:
The patent segments the ML model into a unified global architecture maintained at worker nodes, with local feature adaptation capabilities. Each worker node maintains the complete model structure and trains locally on available data, but the segmentation allows independent operation without requiring coordination for model structure synchronization, reducing system complexity while preserving local adaptability
3Ease of operation
If the same ML model architecture is used across all worker nodes, then the models are easier to manage and deploy, but the models cannot effectively handle dynamic environments with varying infrastructure and feature sets
Solution Approach 1:
The patent implements dynamic feature adaptation within a unified model architecture. The model structure remains consistent across all worker nodes for ease of deployment, but the feature selection and weighting are dynamically adjusted based on local infrastructure characteristics and available measurement tools, allowing the same architecture to adapt to diverse environments including cloud, edge, and IoT deployments
Data Source
AI summary
Methods and leader computing devices for developing machine-learning models. A method comprises receiving, at a leader computing device from each of a plurality of worker computing devices, weights and model architecture information for part of a trained ML model. The method further comprises determining, at the leader computing device, a common portion of the parts of trained ML models that is useable by all of the plurality of worker computing devices, and generating, at the leader computing device, an updated common portion of the ML model using the common portion of the parts of trained ML models and the weights and model architecture information from each of the plurality of worker computing devices. The method further comprises initiating transmission of the updated common portion of the ML model to the worker computing devices.