Convolutional Neural Networks (CNNs) in the Internet-of-Things (IoT)-based applications face stringent constraints, like limited memory capacity and energy resources due to many computations in convolution layers. In order to reduce the computational workload in these layers, this paper proposes a hybrid convolution method in conjunction with a Particle of Swarm Convolution Layer Optimization (PSCLO) algorithm. The hybrid convolution is an approximation that exploits the inherent symmetry of filter termed as symmetry approximation and Winograd algorithm structure termed as tile quantization approximation. PSCLO optimizes the balance between workload reduction and accuracy degradation for each convolution layer by selecting fine-tuned thresholds to control each approximation's intensity. The proposed methods have been evaluated on ImageNet, MNIST, Fashion-MNIST, SVHN, and CIFAR-10 datasets. The proposed techniques achieved ∼5.28x multiplicative workload reduction without significant accuracy degradation (<0.1%) for ImageNet on ResNet-18, which is ∼1.08x less multiplicative workload as compared to state-of-the-art Winograd CNN pruning. For LeNet, ∼3.87x and ∼3.93x was the multiplicative workload reduction for MNIST and Fashion-MNIST datasets. The additive workload reduction was ∼2.5x and ∼2.56x for the respective datasets. There is no significant accuracy loss for MNIST and Fashion-MNIST dataset.

No abstract

Using Spatial Domain Correlation Pattern Recognition (CPR) in Internet-of-Things (IoT)-based applications often faces constraints, like inadequate computational resources and limited memory. To reduce the computation workload of inference due to large spatial-domain CPR filters and convert filter weights into hardware-friendly data-types, this paper introduces the power-of-two (Po2) and dynamic-fixed-point (DFP) quantization techniques for weight compression and the sparsity induction in filters. Weight quantization re-training (WQR), the log-polar, and the inverse log-polar geometric transformations are introduced to reduce quantization error. WQR is a method of retraining the CPR filter, which is presented to recover the accuracy loss. It forces the given quantization scheme by adding the quantization error in the training sample and then re-quantizes the filter to the desired quantization levels which reduce quantization noise. Further, Particle Swarm Optimization (PSO) is used to fine-tune parameters during WQR. Both geometric transforms are applied as pre-processing steps. The Po2 quantization scheme showed better performance close to the performance of full precision, while the DFP quantization showed further closeness to the Receiver Operator Characteristic of full precision for the same bit-length. Overall, spatial-trained filters showed a better compression ratio for Po2 quantization after retraining of the CPR filter. The direct quantization approach achieved a compression ratio of 8 at 4.37× speedup with no accuracy degradation. In contrast, quantization with a log-polar transform is accomplished at a compression ratio of 4 at 1.12× speedup, but, in this case, 16% accuracy of degradation is noticed. Inverse log-polar transform showed a compression ratio of 16 at 8.90× speedup and 6% accuracy degradation. All the mentioned accuracies are reported for a common database.