Distributed AI Model Aggregation for CNN Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of obtaining an optimal artificial intelligence (AI) model, such as a convolutional neural network (CNN), for an AI chip is exacerbated by the large size of the model, which requires significant computing resources due to hundreds of layers and tens of thousands of weights, making it difficult to achieve optimal performance.
Innovation Solution
A distributed system and method are implemented where multiple client devices and AI chips collaborate to determine and refine an AI model through iterative processes, with a host device coordinating to select and update AI models based on performance values, and a global AI model is determined by aggregating optimal models from multiple devices, allowing for efficient training and convergence to a global optimal model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a large-scale CNN model with hundreds of layers and tens of thousands of weights is embedded in an AI chip, then the model size increases to a few megabytes, but the computing time required to obtain an optimal model becomes excessively long
Solution Approach 1:
The patent divides the large-scale CNN model training task into multiple segments distributed across multiple client devices. Each client device trains a portion of the model or different layers independently, then the results are aggregated. This segmentation allows parallel processing of model training, significantly reducing the total computing time required to obtain an optimal model while handling the large model size of a few megabytes.
2Productivity
If multiple client devices and AI chips are used to collaboratively determine and refine AI models through iterative processes, then the convergence to global optimal model is accelerated, but the system complexity increases
Solution Approach 1:
The patent introduces a server as an intermediary that coordinates the distributed model training process across multiple client devices and AI chips. The server manages the iterative refinement process, aggregates results from various clients, and orchestrates the collaboration to converge on a global optimal model. This intermediary simplifies the system complexity by centralizing coordination while maintaining the productivity benefits of distributed processing.
Data Source
AI summary
A system may include multiple client devices and a processing device communicatively coupled to the client devices. Each client device includes an artificial intelligence (AI) chip and is configured to generate an AI model. The processing device may be configured to (i) receive a respective AI model and an associated performance value of the respective AI model from each of the plurality of client devices; (ii) determine an optimal AI model based on the performance values associated with the respective AI models from the plurality of client devices; and (iii) determine a global AI model based on the optimal AI model. The system may load the global AI model into an AI chip of a client device to cause the client device to perform an AI task based on the global AI model in the AI chip. The AI model may include a convolutional neural network.