Processor Inference During FPGA Model Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for processing images using convolutional neural networks on programmable logic devices, such as FPGAs, face challenges in efficiently carrying out inference processing based on new learning models during the time it takes to convert these models into operable logic data.
Innovation Solution
An information processing system comprising a processor, a programmable logic device, a machine learning processing unit, a converter, and a controller, which enables the processor to perform specified processing based on a new learning model while the conversion of the learning model into logic data is in progress.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the new learning model is converted into logic data for the programmable logic device, then the inference processing efficiency is improved, but the time required for conversion causes a delay in using the new learning model
Solution Approach 1:
The system performs preliminary conversion of the new learning model into logic data before the programmable logic device is fully ready to use it. The convertor starts converting the new learning model into logic data immediately when the new learning model is generated, so that the logic data is ready in advance when the programmable logic device completes its rewriting process.
Solution Approach 2:
The convertor acts as an intermediary component that bridges the gap between the machine learning processing unit and the programmable logic device. It converts the new learning model into logic data format that the programmable logic device can use, enabling seamless transition and eliminating the time loss during model updates.
2Adaptability or versatility
If the programmable logic device rewrites logic data to accommodate a new learning model, then the system can process new data patterns, but the rewriting process takes several hours to days
Solution Approach 1:
The conversion of the new learning model into logic data is performed in advance, before the programmable logic device completes its rewriting process. This preliminary action ensures that when the rewriting is complete, the system can immediately use the new learning model without additional delay.
Solution Approach 2:
The system maintains continuous inference processing capability by using the processor to perform inference based on the new learning model during the rewriting period. This ensures that the useful action of processing data continues without interruption while the programmable logic device is being updated.
3Loss of time
If the processor carries out inference based on the new learning model during conversion, then early-stage processing is enabled, but the processor is less efficient than the programmable logic device
Solution Approach 1:
The processor performs inference processing in advance during the conversion period, enabling early-stage processing of data with the new learning model. Although the processor is less efficient than the programmable logic device, this preliminary action eliminates the time loss and ensures continuous processing capability.
Solution Approach 2:
The system dynamically switches between the processor and the programmable logic device for inference processing. During the conversion period, the processor handles inference tasks. Once the conversion is complete and the programmable logic device is ready, the system transitions to using the programmable logic device for more efficient processing.
Data Source
AI summary
An information processing system that carries out a specified processing based on a learning model, comprises: a processor; a programmable logic device that rewrites logic data and reconstitutes a circuit; a machine learning processing unit that carries out machine learning and generates a new learning model for the specified processing; a convertor that converts the new learning model into the logic data that is operable in the programmable logic device; and a controller that enables the processor to carry out the specified processing based on the new learning model while the time the new learning model is converted into the logic data by the convertor.