Parallel AI Model Training via Greedy Client Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in obtaining an optimal artificial intelligence (AI) model for AI chips, particularly convolutional neural networks (CNNs), which are large and require significant computing time due to their size, making it difficult to achieve efficient loading and processing within embedded hardware.
Innovation Solution
The solution involves a parallel configuration method where multiple AI devices or chips work together to train and update AI models in parallel, using a host device to manage training data and configuration parameters, and client devices to execute AI tasks, allowing for efficient distribution of computation and optimization of AI models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the size of the CNN is increased to improve AI task performance, then the model capability is improved, but the computing time required for training increases significantly
Solution Approach 1:
The patent divides the training process into multiple parallel configurations, where different subsets of the CNN model are trained simultaneously across multiple devices. This segmentation allows the large-scale CNN training to be broken down into smaller, parallelizable tasks that can be executed concurrently, reducing overall training time while maintaining model performance.
Solution Approach 2:
The patent introduces a parallel dimension to the training process by distributing training across multiple devices configured in parallel. This transforms the single-sequence training approach into a multi-dimensional parallel execution model, enabling simultaneous training of different model components or data subsets across the parallel configuration.
2Productivity
If the computing power is increased to reduce training time, then the training speed is improved, but the device complexity increases
Solution Approach 1:
Instead of concentrating all computing power in a single complex device, the patent segments the computing workload across multiple devices with moderate computational capabilities. This distribution approach achieves high training speed through parallel execution while keeping individual device complexity manageable.
Solution Approach 2:
The patent combines multiple parallel computing devices to achieve the equivalent computational power of a single high-end device. By merging the computational resources of multiple simpler devices through parallel coordination, the system achieves high training speed without requiring any single device to have excessive complexity.
Data Source
AI summary
A system may include multiple client devices and a processing device communicatively coupled to the client devices. One or more client devices may implement a greedy approach in searching for an optimal artificial intelligence (AI) model. For example, a client device may use a training dataset to perform an AI task, and update its AI model. The client device may verify the performance of the AI task and determine whether to accept or reject its updated AI model. Upon rejection, the client device may repeat updating its AI model until the updated AI model is accepted, or until a stopping criteria is met. The processing device may be configured to update the initial AI models based on the accepted updated AI models obtained in the multiple client device. Training data for each of the client devices may contain a subset shuffled from a larger training dataset.