ML Model Segmentation for Dynamic Resource Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed Machine Learning solutions face challenges in rapidly adapting to changing requirements and priorities, as well as security concerns, due to the inability to modify ML model parameters or structure without complete retraining, which is resource-intensive and insecure.
Innovation Solution
The introduction of functional building blocks that can combine to execute ML models, allowing for real-time modification of the model structure without retraining, and the use of resource nodes to execute and reassign these blocks, enabling continuous training and improved security through dynamic resource allocation and secure model sharing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an ML model is trained and deployed, then the model can perform its task, but the model parameters and structure cannot be changed without complete retraining which is time-consuming and resource-intensive
Solution Approach 1:
The patent segments the ML model into multiple independent functional parts or components that can be individually modified and reconfigured. This segmentation allows specific model parts to be updated without requiring complete retraining of the entire model, thereby reducing retraining time and resource consumption while maintaining model adaptability.
Solution Approach 2:
The patent implements dynamic model architecture where the model structure and parameters can be changed during operation. This dynamic capability enables the system to adapt to new requirements by modifying individual model components on-demand without static constraints, eliminating the need for complete retraining when adaptations are required.
2Power
If additional computing resources are dedicated to an ML task by adding hidden layers to the Neural Network, then the model capacity increases, but the model must be completely retrained from scratch
Solution Approach 1:
The patent divides the neural network into modular functional parts that can be independently configured. When additional computing power is needed, new functional parts can be added to the existing model architecture without disrupting or requiring retraining of previously trained components, thus maintaining model availability while increasing capacity.
Solution Approach 2:
The patent enables continuous model operation where the model remains available and functional throughout the process of adding new layers or capacity. The segmented architecture allows new computational resources to be integrated without stopping model service, ensuring continuous useful action and avoiding downtime associated with complete retraining.
3Ease of operation
If the complete ML model is distributed to all local nodes in federated learning, then the model can be executed locally, but security is compromised as a third party needs only compromise one local node to obtain the model structure
Solution Approach 1:
The patent segments the ML model into multiple functional parts and distributes these segments across different nodes in a distributed architecture. This segmentation ensures that no single node contains the complete model structure, thereby mitigating security risks. Even if one node is compromised, the attacker obtains only a fragment of the model rather than the complete structure.
Solution Approach 2:
The patent introduces an intermediary layer or orchestration mechanism that manages the distributed model parts. This intermediary coordinates the execution of segmented model components across multiple nodes without requiring any single node to hold the complete model, thereby maintaining both local execution capability and enhanced security through distributed fragment storage.
Data Source
AI summary
A computer implemented method is disclosed for facilitating execution of a Machine Learning (ML) model by a system of resource nodes, the ML model comprising a plurality of functional model parts. The method, performed by a resource node of the system, comprises generating a placement map for the ML model, wherein the placement map specifies, for each of the functional model parts, a mapping between the functional model part and at least one resource node of the system that is to execute the functional model part. The method further comprises identifying, from the placement map, a functional model part that is to be executed by the resource node, and executing the identified functional model part.