Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SHA-CNN: Scalable Hierarchical Aware Convolutional Neural Network for Edge AI

Narendra Singh Dhakad, Yuvnish Malhotra, Santosh Kumar Vishvakarma, Kaushik Roy

TL;DR

SHA-CNN addresses scalability and edge deployment challenges in hierarchical CNNs by introducing a training-then-sharing architecture that uses B-CNN weight extraction and shared fully connected layers across hierarchy levels. The approach reduces inference parameters and MACs while maintaining competitive accuracy on MNIST, CIFAR-10, and CIFAR-100, and enables easy addition of new classes. Experimental results show accuracy of 99.34%, 83.35%, and 63.66% at three datasets with about 10% MAC reduction compared to BCNN, validated on a PYNQ Z2 FPGA. This work advances edge AI by delivering scalable, hierarchy-aware classifiers suitable for dynamic datasets and hardware-constrained environments.

Abstract

This paper introduces a Scalable Hierarchical Aware Convolutional Neural Network (SHA-CNN) model architecture for Edge AI applications. The proposed hierarchical CNN model is meticulously crafted to strike a balance between computational efficiency and accuracy, addressing the challenges posed by resource-constrained edge devices. SHA-CNN demonstrates its efficacy by achieving accuracy comparable to state-of-the-art hierarchical models while outperforming baseline models in accuracy metrics. The key innovation lies in the model's hierarchical awareness, enabling it to discern and prioritize relevant features at multiple levels of abstraction. The proposed architecture classifies data in a hierarchical manner, facilitating a nuanced understanding of complex features within the datasets. Moreover, SHA-CNN exhibits a remarkable capacity for scalability, allowing for the seamless incorporation of new classes. This flexibility is particularly advantageous in dynamic environments where the model needs to adapt to evolving datasets and accommodate additional classes without the need for extensive retraining. Testing has been conducted on the PYNQ Z2 FPGA board to validate the proposed model. The results achieved an accuracy of 99.34%, 83.35%, and 63.66% for MNIST, CIFAR-10, and CIFAR-100 datasets, respectively. For CIFAR-100, our proposed architecture performs hierarchical classification with 10% reduced computation while compromising only 0.7% accuracy with the state-of-the-art. The adaptability of SHA-CNN to FPGA architecture underscores its potential for deployment in edge devices, where computational resources are limited. The SHA-CNN framework thus emerges as a promising advancement in the intersection of hierarchical CNNs, scalability, and FPGA-based Edge AI.

SHA-CNN: Scalable Hierarchical Aware Convolutional Neural Network for Edge AI

TL;DR

SHA-CNN addresses scalability and edge deployment challenges in hierarchical CNNs by introducing a training-then-sharing architecture that uses B-CNN weight extraction and shared fully connected layers across hierarchy levels. The approach reduces inference parameters and MACs while maintaining competitive accuracy on MNIST, CIFAR-10, and CIFAR-100, and enables easy addition of new classes. Experimental results show accuracy of 99.34%, 83.35%, and 63.66% at three datasets with about 10% MAC reduction compared to BCNN, validated on a PYNQ Z2 FPGA. This work advances edge AI by delivering scalable, hierarchy-aware classifiers suitable for dynamic datasets and hardware-constrained environments.

Abstract

This paper introduces a Scalable Hierarchical Aware Convolutional Neural Network (SHA-CNN) model architecture for Edge AI applications. The proposed hierarchical CNN model is meticulously crafted to strike a balance between computational efficiency and accuracy, addressing the challenges posed by resource-constrained edge devices. SHA-CNN demonstrates its efficacy by achieving accuracy comparable to state-of-the-art hierarchical models while outperforming baseline models in accuracy metrics. The key innovation lies in the model's hierarchical awareness, enabling it to discern and prioritize relevant features at multiple levels of abstraction. The proposed architecture classifies data in a hierarchical manner, facilitating a nuanced understanding of complex features within the datasets. Moreover, SHA-CNN exhibits a remarkable capacity for scalability, allowing for the seamless incorporation of new classes. This flexibility is particularly advantageous in dynamic environments where the model needs to adapt to evolving datasets and accommodate additional classes without the need for extensive retraining. Testing has been conducted on the PYNQ Z2 FPGA board to validate the proposed model. The results achieved an accuracy of 99.34%, 83.35%, and 63.66% for MNIST, CIFAR-10, and CIFAR-100 datasets, respectively. For CIFAR-100, our proposed architecture performs hierarchical classification with 10% reduced computation while compromising only 0.7% accuracy with the state-of-the-art. The adaptability of SHA-CNN to FPGA architecture underscores its potential for deployment in edge devices, where computational resources are limited. The SHA-CNN framework thus emerges as a promising advancement in the intersection of hierarchical CNNs, scalability, and FPGA-based Edge AI.
Paper Structure (7 sections, 11 figures, 1 table)

This paper contains 7 sections, 11 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Proposed Model Description
  4. B-CNN Training and Weights Extraction
  5. Shared Fully Connected Layers at all levels
  6. Experimental Results
  7. Conclusion

Figures (11)

  • Figure 1: Importance of hierarchical classification.
  • Figure 2: Example of a hierarchical tree for various classes of transportation modes.
  • Figure 3: B-CNN model architecture BCNN. The level increase resulted in the increase of another block and branch with a fully connected layer, leading to more parameters in the network.
  • Figure 4: SHA-CNN: training and weight extraction.
  • Figure 5: SHA-CNN: shared fully connected layers for all branches.
  • ...and 6 more figures