Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems
Binarized neural networks are well suited for FPGA accelerators since their fine-grained architecture allows the creation of custom operators to support low-precision arithmetic operations, and the reduction in memory requirements means that all the network parameters can be stored in internal memor...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-10-01
|
| Series: | Electronics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2079-9292/9/11/1765 |
| _version_ | 1797550037813690368 |
|---|---|
| author | Elim Yi Lam Kwan Jose Nunez-Yanez |
| author_facet | Elim Yi Lam Kwan Jose Nunez-Yanez |
| author_sort | Elim Yi Lam Kwan |
| collection | DOAJ |
| description | Binarized neural networks are well suited for FPGA accelerators since their fine-grained architecture allows the creation of custom operators to support low-precision arithmetic operations, and the reduction in memory requirements means that all the network parameters can be stored in internal memory. Although good progress has been made to improve the accuracy of binarized networks, it can be significantly lower than networks where weights and activations have multi-bit precision. In this paper, we address this issue by adaptively choosing the number of frames used during inference, exploiting the high frame rates that binarized neural networks can achieve. We present a novel entropy-based adaptive filtering technique that improves accuracy by varying the system’s processing rate based on the entropy present in the neural network output. We focus on using real data captured with a standard camera rather than using standard datasets that do not realistically represent the artifacts in video stream content. The overall design has been prototyped on the Avnet Zedboard, which achieved 70.4% accuracy with a full processing pipeline from video capture to final classification output, which is 1.9 times better compared to the base static frame rate system. The main feature of the system is that while the classification rate averages a constant 30 fps, the real processing rate is dynamic and varies between 30 and 142 fps, adapting to the complexity of the data. The dynamic processing rate results in better efficiency that simply working at full frame rate while delivering high accuracy. |
| first_indexed | 2024-03-10T15:22:47Z |
| format | Article |
| id | doaj.art-8abd9164f62d4f3ebb8645ff4acd31a6 |
| institution | Directory Open Access Journal |
| issn | 2079-9292 |
| language | English |
| last_indexed | 2024-03-10T15:22:47Z |
| publishDate | 2020-10-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Electronics |
| spelling | doaj.art-8abd9164f62d4f3ebb8645ff4acd31a62023-11-20T18:18:34ZengMDPI AGElectronics2079-92922020-10-01911176510.3390/electronics9111765Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network SystemsElim Yi Lam Kwan0Jose Nunez-Yanez1Department of Electrical and Electronics Engineering, Faculty of Engineering, University of Bristol, Bristol BS8 1TL, UKDepartment of Electrical and Electronics Engineering, Faculty of Engineering, University of Bristol, Bristol BS8 1TL, UKBinarized neural networks are well suited for FPGA accelerators since their fine-grained architecture allows the creation of custom operators to support low-precision arithmetic operations, and the reduction in memory requirements means that all the network parameters can be stored in internal memory. Although good progress has been made to improve the accuracy of binarized networks, it can be significantly lower than networks where weights and activations have multi-bit precision. In this paper, we address this issue by adaptively choosing the number of frames used during inference, exploiting the high frame rates that binarized neural networks can achieve. We present a novel entropy-based adaptive filtering technique that improves accuracy by varying the system’s processing rate based on the entropy present in the neural network output. We focus on using real data captured with a standard camera rather than using standard datasets that do not realistically represent the artifacts in video stream content. The overall design has been prototyped on the Avnet Zedboard, which achieved 70.4% accuracy with a full processing pipeline from video capture to final classification output, which is 1.9 times better compared to the base static frame rate system. The main feature of the system is that while the classification rate averages a constant 30 fps, the real processing rate is dynamic and varies between 30 and 142 fps, adapting to the complexity of the data. The dynamic processing rate results in better efficiency that simply working at full frame rate while delivering high accuracy.https://www.mdpi.com/2079-9292/9/11/1765entropy adaptive filterhigh-accuracy FPGAresource-efficient FPGAreal time object recognitionsoftware accelerationlow precision network |
| spellingShingle | Elim Yi Lam Kwan Jose Nunez-Yanez Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems Electronics entropy adaptive filter high-accuracy FPGA resource-efficient FPGA real time object recognition software acceleration low precision network |
| title | Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems |
| title_full | Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems |
| title_fullStr | Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems |
| title_full_unstemmed | Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems |
| title_short | Entropy-Driven Adaptive Filtering for High-Accuracy and Resource-Efficient FPGA-Based Neural Network Systems |
| title_sort | entropy driven adaptive filtering for high accuracy and resource efficient fpga based neural network systems |
| topic | entropy adaptive filter high-accuracy FPGA resource-efficient FPGA real time object recognition software acceleration low precision network |
| url | https://www.mdpi.com/2079-9292/9/11/1765 |
| work_keys_str_mv | AT elimyilamkwan entropydrivenadaptivefilteringforhighaccuracyandresourceefficientfpgabasedneuralnetworksystems AT josenunezyanez entropydrivenadaptivefilteringforhighaccuracyandresourceefficientfpgabasedneuralnetworksystems |