Decentralized AI Model Selection via Virtual Chip Simulation
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 hindered by the large size of the model, requiring significant computing resources due to tens of thousands of weights, making it difficult to achieve efficient loading and performance.
Innovation Solution
A decentralized network system that utilizes multiple processing devices to generate, verify, and update AI models, allowing for the use of both physical and virtual AI chips, where virtual chips simulate operations and enable efficient testing of various models before loading the optimal model into a physical chip for real-time applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the CNN model size is increased to improve AI performance, then the AI capability is improved, but the computing time and resource requirements increase significantly
Solution Approach 1:
The patent segments the AI model training and optimization process across multiple processing devices in a decentralized network. Each device contributes computational resources to train different portions of the CNN model or perform parallel computations, dividing the overall computing task into manageable segments that can be processed simultaneously, thereby reducing total computing time while maintaining model performance.
Solution Approach 2:
The patent implements a nested structure where virtual AI chips are nested within the decentralized network system, which is itself nested within the broader AI model training framework. The virtual chips simulate and validate model portions before full deployment, creating nested layers of computation that optimize resource usage and reduce redundant computing time.
2Reliability
If the CNN model size is increased to improve AI performance, then the AI capability is improved, but the difficulty of obtaining an optimal model increases
Solution Approach 1:
The patent merges the computational capabilities of multiple processing devices into a unified decentralized network system. By combining resources across devices, the system achieves the computational power needed to optimize large CNN models without requiring a single complex centralized system, thereby distributing and simplifying the overall optimization complexity.
Solution Approach 2:
The patent introduces virtual AI chips as intermediary elements between the physical processing devices and the final AI model deployment. These virtual chips serve as simulation environments that validate model portions before full implementation, acting as intermediaries that simplify the optimization process by filtering out suboptimal configurations early in the training process.
3Productivity
If more computing resources are allocated to a single device for model training, then the training speed is improved, but the system cost and energy consumption increase
Solution Approach 1:
The patent makes each processing device in the decentralized network multi-functional, capable of performing various training tasks, validation, and model optimization roles. This universality allows the system to achieve high training speeds by dynamically allocating tasks across multiple devices rather than concentrating all computational power in a single high-energy-consuming device, thereby improving productivity while distributing energy consumption.
Data Source
AI summary
A system may include a decentralized communication network and multiple processing devices on the network. Each processing device may have an artificial intelligence (AI) chip, the device may be configured to generate an AI model, determine the performance value of the AI model on the AI chip, receive a chain from the network where the chain contains a performance measure. If the performance value of the AI model is better than the performance measure, then the processing device may broadcast the AI model to the network for verification. If the AI model is verified by the network, the device may update the chain with the performance value so that the chain can be shared by the multiple processing devices on the network. Any processing device on the network may also verify an AI model broadcasted by any other device. Methods for generating the AI model are also provided.