• The Journal of Korea Robotics Society
• /
• v.18 no.1
• /
• pp.88-92
• /
• 2023

Lithium-ion batteries are actively used in various industrial sites such as field robots, drones, and electric vehicles due to their high energy efficiency, light weight, long life span, and low self-discharge rate. When using a lithium-ion battery in a field, it is important to accurately estimate the SoC (State of Charge) of batteries to prevent damage. In recent years, SoC estimation using data-based artificial neural networks has been in the spotlight, but it has been difficult to deploy in the embedded board environment at the actual site because the computation is heavy and complex. To solve this problem, neural network lightening technologies such as network pruning have recently attracted attention. When pruning a neural network, the performance varies depending on which layer and how much pruning is performed. In this paper, we introduce an optimized pruning technique by improving the existing pruning method, and perform a comparative experiment to analyze the results.

• The Journal of Korea Robotics Society
• /
• v.17 no.2
• /
• pp.133-141
• /
• 2022

This paper introduces model compression algorithms which make a deep neural network smaller and faster for embedded systems. The model compression algorithms can be largely categorized into pruning, quantization and knowledge distillation. In this study, gradual pruning, quantization aware training, and knowledge distillation which learns the activation boundary in the hidden layer of the teacher neural network are integrated. As a large deep neural network is compressed and accelerated by these algorithms, embedded computing boards can run the deep neural network much faster with less memory usage while preserving the reasonable accuracy. To evaluate the performance of the compressed neural networks, we evaluate the size, latency and accuracy of the deep neural network, DenseNet201, for image classification with CIFAR-10 dataset on the NVIDIA Jetson Xavier.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2001.12a
• /
• pp.247-250
• /
• 2001

In this paper, we propose an growing and pruning algorithm to design the adaptive structure of modular wavelet neural network(MWNN) with F-projection and geometric growing criterion. Geometric growing criterion consists of estimated error criterion considering local error and angle criterion which attempts to assign wavelet function that is nearly orthogonal to all other existing wavelet functions. These criteria provide a methodology that a network designer can constructs wavelet neural network according to one's intention. The proposed growing algorithm grows the module and the size of modules. Also, the pruning algorithm eliminates unnecessary node of module or module from constructed MWNN to overcome the problem due to localized characteristic of wavelet neural network which is used to modules of MWNN. We apply the proposed constructing algorithm of the adaptive structure of MWNN to approximation problems of 1-D function and 2-D function, and evaluate the effectiveness of the proposed algorithm.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.429-431
• /
• 2003

The induced learning retardation method involves the temporary inhibition of the artificial neural network’s active units from participating in the error reduction process during training. This stimulates the less active units to contribute significantly to reduce the network error. However, some less active units are not sensitive to stimulation making them almost useless. The network can then be pruned by removing the less active units to make it smaller and more efficient. This study focuses on making the network more efficient and accurate by developing the induced learning retardation and pruning sequence training method. The developed procedure results to faster learning and more accurate artificial neural network for satellite image classification.

• Journal of the Institute of Convergence Signal Processing
• /
• v.22 no.3
• /
• pp.99-103
• /
• 2021

Recently, convolutional neural networks (CNNs) have been widely utilized due to their outstanding performance in various computer vision fields. However, due to their computational-intensive and high memory requirements, it is difficult to deploy CNNs on hardware platforms that have limited resources, such as mobile devices and IoT devices. To address these limitations, a neural network compression research is underway to reduce the size of neural networks while maintaining their performance. This paper proposes a CNN compression technique that dynamically adjusts the thresholds of pruning, one of the neural network compression techniques. Unlike the conventional pruning that experimentally or heuristically sets the thresholds that determine the weights to be pruned, the proposed technique can dynamically find the optimal thresholds that prevent accuracy degradation and output the light-weight neural network in less time. To validate the performance of the proposed technique, the LeNet was trained using the MNIST dataset and the light-weight LeNet could be automatically obtained 1.3 to 3 times faster without loss of accuracy.

• Journal of Convergence for Information Technology
• /
• v.10 no.8
• /
• pp.23-34
• /
• 2020

Recently, the artificial intelligence deep learning field has been hard to commercialize due to the high computing power and the price problem of computing resources. In this paper, we apply a double pruning techniques to evaluate the performance of the in-depth neural network and various datasets. Double pruning combines basic Network-slimming and Parameter-prunning. Our proposed technique has the advantage of reducing the parameters that are not important to the existing learning and improving the speed without compromising the learning accuracy. After training various datasets, the pruning ratio was increased to reduce the size of the model.We confirmed that MobileNet-V3 showed the highest performance as a result of NetScore performance analysis. We confirmed that the performance after pruning was the highest in MobileNet-V3 consisting of depthwise seperable convolution neural networks in the Cifar 10 dataset, and VGGNet and ResNet in traditional convolutional neural networks also increased significantly.

• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.484-484
• /
• 2000

The problem of determining the proper size of an neural network is recognized to be crucial, especially for its practical implications in such important issues as learning and generalization. Unfortunately, it usually is not obvious what size is best: a system that is too snail will not be able to learn the data while one that is just big enough may learn the slowly and be very sensitive to initial conditions and learning parameters. One popular technique is commonly known as pruning and consists of training a larger than necessary network and then removing unnecessary weights/nodes. In this paper, a new pruning method is developed, based on the penalty-term methods. This method makes the neural network good for the generalization and reduces the retraining time after pruning weights/nodes.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.1
• /
• pp.141-144
• /
• 2020

Convolutional neural networks (CNNs) show high performance in computer vision tasks including object detection, but a lot of weight storage and computation is required. In this paper, a pruning scheme is applied to CNNs for object detection, which can remove much amount of weights with a negligible performance degradation. Contrary to the previous ones, the pruning scheme applied in this paper considers the base accelerator architecture. With the consideration, the pruned CNNs can be efficiently performed on an ASIC or FPGA accelerator. Even with the constrained pruning, the resulting CNN shows a negligible degradation of detection performance, less-than-1% point degradation of mAP on VOD0712 test set. With the proposed scheme, CNNs can be applied to objection dtection efficiently.

• Journal of KIISE:Software and Applications
• /
• v.29 no.5
• /
• pp.289-294
• /
• 2002

It is known that restructuring feed-forward neural network affects generalization capability and efficiency of the network. In this paper, we introduce a new approach to restructure a neural network using abstraction of the hidden knowledge that the network has teamed. This method involves extracting local rules from non-input nodes and aggregation of the rules into global rule base. The extracted local rules are used for pruning unnecessary connections of local nodes and the aggregation eliminates any possible redundancies arid inconsistencies among local rule-based structures. Final network is generated by the global rule-based structure. Complexity of the final network is much reduced, compared to a fully-connected neural network and generalization capability is improved. Empirical results are also shown.

• Smart Structures and Systems
• /
• v.32 no.3
• /
• pp.179-193
• /
• 2023

This study explores an alternative to the existing centralized process for data anomaly detection in modern Internet of Things (IoT)-based structural health monitoring (SHM) systems. An edge intelligence framework is proposed for the early detection and classification of various data anomalies facilitating quality enhancement of acquired data before transmitting to a central system. State-of-the-art deep neural network pruning techniques are investigated and compared aiming to significantly reduce the network size so that it can run efficiently on resource-constrained edge devices such as wireless smart sensors. Further, depthwise separable convolution (DSC) is invoked, the integration of which with advanced structural pruning methods exhibited superior compression capability. Last but not least, quantization-aware training (QAT) is adopted for faster processing and lower memory and power consumption. The proposed edge intelligence framework will eventually lead to reduced network overload and latency. This will enable intelligent self-adaptation strategies to be employed to timely deal with a faulty sensor, minimizing the wasteful use of power, memory, and other resources in wireless smart sensors, increasing efficiency, and reducing maintenance costs for modern smart SHM systems. This study presents a theoretical foundation for the proposed framework, the validation of which through actual field trials is a scope for future work.