Lightweight AI Computing Device With Context-Routed Neuron Modules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based AI systems require enormous memory and are inflexible, making them difficult to implement in environments with limited hardware space and power, and they are limited to specific datasets, restricting their application flexibility.
Innovation Solution
A lightweight AI computing device with neuron cell modules designated for specific contexts, allowing real-time training and recognition by adjusting neuron cell module radii based on context and class information, using an AI engine controller and minimum distance detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If deep learning-based AI systems are used for processing big data, then AI processing capability is improved, but hardware memory requirements increase enormously
Solution Approach 1:
The patent segments the AI computing system into multiple neuron cell modules, each dedicated to processing specific contexts or data types. This segmentation allows the system to handle complex AI tasks by distributing computations across smaller, specialized units rather than requiring a single large-memory system, thus reducing overall hardware memory requirements while maintaining AI processing capability.
Solution Approach 2:
The patent implements dynamic radius adjustment for neuron cell modules based on context information and class information. The radius of each neuron cell module can be adjusted in real-time during operation, allowing the system to adapt its computational requirements dynamically. This enables the system to process AI tasks with varying complexity using appropriate memory resources at each moment, avoiding the need for enormous fixed memory capacity.
2Power
If deep learning-based AI systems are used for processing big data, then AI processing capability is improved, but device flexibility deteriorates
Solution Approach 1:
The patent creates a universal AI computing device where neuron cell modules can be configured to handle different contexts and applications. Each neuron cell module can be assigned different radii and processed different types of data based on context information, enabling the same hardware architecture to support multiple AI applications and datasets flexibly without requiring hardware changes.
Solution Approach 2:
The dynamic adjustment of neuron cell module radii based on context and class information provides real-time adaptability. The system can reconfigure its computational architecture on-the-fly to match different application requirements, maintaining high AI processing capability while adapting to various use cases. This dynamic reconfiguration enables the device to switch between different AI tasks without physical hardware changes.
3Productivity
If fixed AI hardware is used for specific datasets, then processing efficiency is improved, but adaptability to different applications deteriorates
Solution Approach 1:
The patent implements dynamic radius adjustment for neuron cell modules based on real-time context information and class information. This allows the hardware to reconfigure its processing parameters dynamically when switching between different datasets or applications, maintaining high processing efficiency for each specific task while enabling broad adaptability across different applications without requiring fixed hardware configurations.
Solution Approach 2:
The system changes operational parameters (specifically the radius of neuron cell modules) based on the context and class information of incoming data. By adjusting these parameters dynamically rather than being fixed, the system achieves both efficiency for specific datasets and flexibility for different applications. The parameter changes allow the same hardware to optimize its performance for varying workloads.
Data Source
AI summary
Proposed is a lightweight AI computing device and operating method for various applications. The AI computing device may include an AI engine controller and a plurality of neuron cell modules. The AI engine controller may receive a data set including a feature vector and context information thereof, and transmit the feature vector to a neuron cell module that matches the context information of the feature vector. The neuron cell module may generate distance information between the center value vector of the neuron cell module and the feature vector, and when the distance information is less than or equal to the radius of the neuron cell module, transmit the distance information and class information of the neuron cell module to the AI engine controller. The AI engine controller may perform a learning or discrimination process on the basis of the distance information and the class information.