Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168662
Xinran Wu, Xiang Chen, Kun Yue
Bayesian network with latent variables (BNLV) plays an important role in the representation of dependence relations and inference of uncertain knowledge with unobserved variables. The variables with large cardinalities in high-dimensional data make it challenging to efficiently learn the large-scaled probability parameters as the conditional probability distributions (CPDs) of BNLV. In this paper, we first propose the multinomial parameter network to parameterize the CPDs w.r.t. latent variables. Then, we extend the M-step of the classic EM algorithm and give the efficient algorithm for parameter learning of BNLV. Experimental results show that our proposed method outperforms some state-of-the-art competitors.
{"title":"Efficient Parameter Learning of Bayesian Network with Latent Variables from High-Dimensional Data","authors":"Xinran Wu, Xiang Chen, Kun Yue","doi":"10.1109/AICAS57966.2023.10168662","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168662","url":null,"abstract":"Bayesian network with latent variables (BNLV) plays an important role in the representation of dependence relations and inference of uncertain knowledge with unobserved variables. The variables with large cardinalities in high-dimensional data make it challenging to efficiently learn the large-scaled probability parameters as the conditional probability distributions (CPDs) of BNLV. In this paper, we first propose the multinomial parameter network to parameterize the CPDs w.r.t. latent variables. Then, we extend the M-step of the classic EM algorithm and give the efficient algorithm for parameter learning of BNLV. Experimental results show that our proposed method outperforms some state-of-the-art competitors.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126950267","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168567
I. Elfadel
In this circuit-theoretic paper, we establish a new result for the global convergence of the waveform relaxation (WR) algorithm in the specific context of analog associative arrays having the Rectified Linear Unit (ReLU) as an activation function. The traditional methods for proving WR convergence on generic analog circuits rely on the use of exponentially weighted norms to control the behavior of the transient waveforms for large simulation intervals. The main contribution of this paper is to show that in the particular case of analog associative ReLU arrays, WR convergence for large simulation intervals does not require exponentially weighted norms and can instead be ascertained using the common norm of uniform convergence. Using the connectivity matrix of the associativity array, a practical criterion for guaranteeing WR convergence is provided.
{"title":"Convergent Waveform Relaxation Schemes for the Transient Analysis of Associative ReLU Arrays","authors":"I. Elfadel","doi":"10.1109/AICAS57966.2023.10168567","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168567","url":null,"abstract":"In this circuit-theoretic paper, we establish a new result for the global convergence of the waveform relaxation (WR) algorithm in the specific context of analog associative arrays having the Rectified Linear Unit (ReLU) as an activation function. The traditional methods for proving WR convergence on generic analog circuits rely on the use of exponentially weighted norms to control the behavior of the transient waveforms for large simulation intervals. The main contribution of this paper is to show that in the particular case of analog associative ReLU arrays, WR convergence for large simulation intervals does not require exponentially weighted norms and can instead be ascertained using the common norm of uniform convergence. Using the connectivity matrix of the associativity array, a practical criterion for guaranteeing WR convergence is provided.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"131 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114521553","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168667
Zhengqing Zhong, Tengxiao Wang, Haibing Wang, Zhihua Zhou, Junxian He, Fang Tang, Xichuan Zhou, Shuangming Yu, Liyuan Liu, N. Wu, Min Tian, Cong Shi
Spiking neural networks (SNNs) and neuromorphic systems have attracted ever increasing interests recently, due to their high computational and energy efficiencies originated from closely imitating the functional mechanism of cerebral cortex, which adopts sparse spikes for information processing. In this work, we present a low-cost real-time face recognition system for potential edge-side intelligent applications. This system is mainly built upon our prior reported MorphBungee neuromorphic chip, which is capable of fast on-chip deep learning for fully-connected (FC) SNN of up to 4 layers, 1K spiking neurons and 256K synapses, under a low power consumption of about 100 mW. Our face recognition system achieves 20-fps and 30-fps image frame rates for real-life human face learning and inference, respectively, and obtains a high face recognition accuracy of 100% among 6 persons. It demonstrates that our face recognition system with the neuromorphic chip is suitable for resource-limited real-time intelligent edge applications.
{"title":"Live Demonstration: Face Recognition at The Edge Using Fast On-Chip Deep Learning Neuromorphic Chip","authors":"Zhengqing Zhong, Tengxiao Wang, Haibing Wang, Zhihua Zhou, Junxian He, Fang Tang, Xichuan Zhou, Shuangming Yu, Liyuan Liu, N. Wu, Min Tian, Cong Shi","doi":"10.1109/AICAS57966.2023.10168667","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168667","url":null,"abstract":"Spiking neural networks (SNNs) and neuromorphic systems have attracted ever increasing interests recently, due to their high computational and energy efficiencies originated from closely imitating the functional mechanism of cerebral cortex, which adopts sparse spikes for information processing. In this work, we present a low-cost real-time face recognition system for potential edge-side intelligent applications. This system is mainly built upon our prior reported MorphBungee neuromorphic chip, which is capable of fast on-chip deep learning for fully-connected (FC) SNN of up to 4 layers, 1K spiking neurons and 256K synapses, under a low power consumption of about 100 mW. Our face recognition system achieves 20-fps and 30-fps image frame rates for real-life human face learning and inference, respectively, and obtains a high face recognition accuracy of 100% among 6 persons. It demonstrates that our face recognition system with the neuromorphic chip is suitable for resource-limited real-time intelligent edge applications.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124562046","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168613
Guang Shi, Yirong Kan, Renyuan Zhang
In this paper, the clinical data thru the questionnaire of body constitution (BC) is analyzed by multiple efficient machine learning algorithms for wide use in traditional Chinese medicine (TCM). This research aims at precisely categorizing the BCs from the life-style; offering the health guidance on the life-styles for recovering the so-called "biased" BCs to the healthy status known as the "Gentle BC". The key features of life-style are identified by machine learning (ML). However, the conventional sole ML algorithm for such application, known as random forest (RF), partial least squares (PLS), or least absolute shrinkage and selection operator (LASSO) hardly offers a small set of significant life-style features. In this work, a special scheme of LASSO learning technology is developed for identifying the reasonably few medical features and improve the diagnosis accuracy simultaneously. By pairing each "biased" BC against the gentle BC, the categorization task is conducted with reduced features. Similarly to the federated learning process, the common features among multiple algorithms are refined. From the real clinical data validation, the BC categorization accuracy is 94.6% which is 24.7% higher than the state-of-the-art (SOTA) works; the average key features are reduced to 17 where the best effort of SOTA is 31. Finally, the common key features are summarized among multiple algorithms.
{"title":"Computer-Aided-Prediction of Body Constitution with Efficient Cock-Tail Learning","authors":"Guang Shi, Yirong Kan, Renyuan Zhang","doi":"10.1109/AICAS57966.2023.10168613","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168613","url":null,"abstract":"In this paper, the clinical data thru the questionnaire of body constitution (BC) is analyzed by multiple efficient machine learning algorithms for wide use in traditional Chinese medicine (TCM). This research aims at precisely categorizing the BCs from the life-style; offering the health guidance on the life-styles for recovering the so-called \"biased\" BCs to the healthy status known as the \"Gentle BC\". The key features of life-style are identified by machine learning (ML). However, the conventional sole ML algorithm for such application, known as random forest (RF), partial least squares (PLS), or least absolute shrinkage and selection operator (LASSO) hardly offers a small set of significant life-style features. In this work, a special scheme of LASSO learning technology is developed for identifying the reasonably few medical features and improve the diagnosis accuracy simultaneously. By pairing each \"biased\" BC against the gentle BC, the categorization task is conducted with reduced features. Similarly to the federated learning process, the common features among multiple algorithms are refined. From the real clinical data validation, the BC categorization accuracy is 94.6% which is 24.7% higher than the state-of-the-art (SOTA) works; the average key features are reduced to 17 where the best effort of SOTA is 31. Finally, the common key features are summarized among multiple algorithms.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116118763","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168655
Anupam Golder, A. Raychowdhury
Manufacturing variability demonstrates significant variations in dynamic power consumption profiles during program execution, even if the printed circuit boards (PCB) are identical and the processors execute the same operations on the same data. In this work, we show how this variability can be leveraged to the benefit of manufacturers by utilizing machine learning (ML) based PCB identification. The proposed technique based on power consumption variability achieves 100% accuracy in identifying PCBs from their power consumption traces after training a linear discriminant analysis (LDA) classifier on a collection of 30 identical PCBs for two test sets collected several months apart.
{"title":"PCB Identification Based on Machine Learning Utilizing Power Consumption Variability","authors":"Anupam Golder, A. Raychowdhury","doi":"10.1109/AICAS57966.2023.10168655","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168655","url":null,"abstract":"Manufacturing variability demonstrates significant variations in dynamic power consumption profiles during program execution, even if the printed circuit boards (PCB) are identical and the processors execute the same operations on the same data. In this work, we show how this variability can be leveraged to the benefit of manufacturers by utilizing machine learning (ML) based PCB identification. The proposed technique based on power consumption variability achieves 100% accuracy in identifying PCBs from their power consumption traces after training a linear discriminant analysis (LDA) classifier on a collection of 30 identical PCBs for two test sets collected several months apart.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121666619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168575
Hanrui Li, Aijaz H. Lone, Fengshi Tian, Jie Yang, M. Sawan, Nazek El‐Atab
Spintronics-based magnetic tunnel junction (MTJ) devices have shown the ability working as both synapse and spike threshold neurons, which is perfectly suitable with the hardware implementation of spike neural network (SNN). It has the inherent advantage of high energy efficiency with ultra-low operation voltage due to its small nanometric size and low depinning current densities. However, hardware-based SNNs training always suffers a significant performance loss compared with original neural networks due to variations among devices and information deficiency as the weights map with device synaptic conductance. Knowledge distillation is a model compression and acceleration method that enables transferring the learning knowledge from a large machine learning model to a smaller model with minimal loss in performance. In this paper, we propose a novel training scheme based on spike knowledge distillation which helps improve the training performance of spin-based SNN (SSNN) model via transferring knowledge from a large CNN model. We propose novel distillation methodologies and demonstrate the effectiveness of the proposed method with detailed experiments on four datasets. The experimental results indicate that our proposed training scheme consistently improves the performance of SSNN model by a large margin.
{"title":"Novel Knowledge Distillation to Improve Training Accuracy of Spin-based SNN","authors":"Hanrui Li, Aijaz H. Lone, Fengshi Tian, Jie Yang, M. Sawan, Nazek El‐Atab","doi":"10.1109/AICAS57966.2023.10168575","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168575","url":null,"abstract":"Spintronics-based magnetic tunnel junction (MTJ) devices have shown the ability working as both synapse and spike threshold neurons, which is perfectly suitable with the hardware implementation of spike neural network (SNN). It has the inherent advantage of high energy efficiency with ultra-low operation voltage due to its small nanometric size and low depinning current densities. However, hardware-based SNNs training always suffers a significant performance loss compared with original neural networks due to variations among devices and information deficiency as the weights map with device synaptic conductance. Knowledge distillation is a model compression and acceleration method that enables transferring the learning knowledge from a large machine learning model to a smaller model with minimal loss in performance. In this paper, we propose a novel training scheme based on spike knowledge distillation which helps improve the training performance of spin-based SNN (SSNN) model via transferring knowledge from a large CNN model. We propose novel distillation methodologies and demonstrate the effectiveness of the proposed method with detailed experiments on four datasets. The experimental results indicate that our proposed training scheme consistently improves the performance of SSNN model by a large margin.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126632409","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168550
Yanghan Zheng, Zhaofang Li, Kaihang Sun, Kuang Lee, K. Tang
Deep neural networks (DNNs) are widely used in various tasks, such as image classification and speech recognition. When deploying DNN to the edge device, the inputs and weights are usually quantized. And there are obvious patterns in the data distribution. Most data have numerous redundant bits, which reduce the utilization rate of computation resources. We proposed an area-efficient DNN accelerator with an effective bit combination mechanism and a reconfigurable multiplier. Based on the modified Baugh-Wooly multiplier, we proposed a multiplier that can process two 4-bit multiplication operations in one cycle, consuming only 1.57 times the area and 2.31 times the power consumption of a traditional multiplier. Based on the data distribution in DNN, we propose a gating approach for the weights of 0, -1, and 1, resulting in a 34.96% reduction in power consumption. The normalized area efficiency of the proposed DNN accelerator using 40nm CMOS technology is 1.11 to 4.90 times higher than previous works [4] - [7].
{"title":"A 40nm area-efficient Effective-bit-combination-based DNN accelerator with the reconfigurable multiplier","authors":"Yanghan Zheng, Zhaofang Li, Kaihang Sun, Kuang Lee, K. Tang","doi":"10.1109/AICAS57966.2023.10168550","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168550","url":null,"abstract":"Deep neural networks (DNNs) are widely used in various tasks, such as image classification and speech recognition. When deploying DNN to the edge device, the inputs and weights are usually quantized. And there are obvious patterns in the data distribution. Most data have numerous redundant bits, which reduce the utilization rate of computation resources. We proposed an area-efficient DNN accelerator with an effective bit combination mechanism and a reconfigurable multiplier. Based on the modified Baugh-Wooly multiplier, we proposed a multiplier that can process two 4-bit multiplication operations in one cycle, consuming only 1.57 times the area and 2.31 times the power consumption of a traditional multiplier. Based on the data distribution in DNN, we propose a gating approach for the weights of 0, -1, and 1, resulting in a 34.96% reduction in power consumption. The normalized area efficiency of the proposed DNN accelerator using 40nm CMOS technology is 1.11 to 4.90 times higher than previous works [4] - [7].","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"98 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117164178","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168650
W. Zhang, Junfeng Chang, Zizhao Peng, Lei Chen, F. An
This demonstration showcases a framework of visual quantification for image enhancement where multivariate Gaussian (MVG) models are trained to assess image visibility. The visibility of an image is depicted by statistical features such as the contrast energy of the gray channel, yellow-blue channel, and red-green channel, average saturation, and gradients. The predicted visibility scores are then applied to define adaptive histogram equalization clip parameters for image enhancement. Finally, the hardware architecture is implemented on an FPGA to demonstrate the results for real-time image enhancement.
{"title":"Live Demonstration: Supervised-learning-based Visual Quantification for Image Enhancement","authors":"W. Zhang, Junfeng Chang, Zizhao Peng, Lei Chen, F. An","doi":"10.1109/AICAS57966.2023.10168650","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168650","url":null,"abstract":"This demonstration showcases a framework of visual quantification for image enhancement where multivariate Gaussian (MVG) models are trained to assess image visibility. The visibility of an image is depicted by statistical features such as the contrast energy of the gray channel, yellow-blue channel, and red-green channel, average saturation, and gradients. The predicted visibility scores are then applied to define adaptive histogram equalization clip parameters for image enhancement. Finally, the hardware architecture is implemented on an FPGA to demonstrate the results for real-time image enhancement.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121920768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168670
Mustafa Matar, Tian Xia, Kimberly Huguenard, D. Huston, S. Wshah
Anomaly detection is a widely utilized technique in the field of wireless sensor networks (WSNs) data stream analysis, aimed at identifying unusual events or anomalies in an early stage. However, the constraints of WSN applications pose a significant challenge in achieving effective and efficient anomaly detection. In this work, we proposes a new multi-head attention-based Bi-LSTM approach for anomaly detection in multivariate time-series. Rather than modeling the time series of individual sensor independently, the proposed approach models the time series of multiple sensors concurrently, taking into account potential latent interactions among them, thereby enhancing the accuracy of anomaly detection. The proposed approach does not require labeled data and can be directly applied in real-world scenarios where labeling a large stream of data from heterogeneous sensors is both difficult and time-consuming. Finally, empirical evaluations using a real-world WSN demonstrate effectiveness and robustness of the proposed approach, outperforming traditional deep learning approaches.
{"title":"Multi-Head Attention based Bi-LSTM for Anomaly Detection in Multivariate Time-Series of WSN","authors":"Mustafa Matar, Tian Xia, Kimberly Huguenard, D. Huston, S. Wshah","doi":"10.1109/AICAS57966.2023.10168670","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168670","url":null,"abstract":"Anomaly detection is a widely utilized technique in the field of wireless sensor networks (WSNs) data stream analysis, aimed at identifying unusual events or anomalies in an early stage. However, the constraints of WSN applications pose a significant challenge in achieving effective and efficient anomaly detection. In this work, we proposes a new multi-head attention-based Bi-LSTM approach for anomaly detection in multivariate time-series. Rather than modeling the time series of individual sensor independently, the proposed approach models the time series of multiple sensors concurrently, taking into account potential latent interactions among them, thereby enhancing the accuracy of anomaly detection. The proposed approach does not require labeled data and can be directly applied in real-world scenarios where labeling a large stream of data from heterogeneous sensors is both difficult and time-consuming. Finally, empirical evaluations using a real-world WSN demonstrate effectiveness and robustness of the proposed approach, outperforming traditional deep learning approaches.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"50 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131063422","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-11DOI: 10.1109/AICAS57966.2023.10168633
Mohammad Hasan Ahmadilivani, Mahdi Taheri, J. Raik, M. Daneshtalab, M. Jenihhin
The superior performance of Deep Neural Networks (DNNs) has led to their application in various aspects of human life. Safety-critical applications are no exception and impose rigorous reliability requirements on DNNs. Quantized Neural Networks (QNNs) have emerged to tackle the complexity of DNN accelerators, however, they are more prone to reliability issues.In this paper, a recent analytical resilience assessment method is adapted for QNNs to identify critical neurons based on a Neuron Vulnerability Factor (NVF). Thereafter, a novel method for splitting the critical neurons is proposed that enables the design of a Lightweight Correction Unit (LCU) in the accelerator without redesigning its computational part.The method is validated by experiments on different QNNs and datasets. The results demonstrate that the proposed method for correcting the faults has a twice smaller overhead than a selective Triple Modular Redundancy (TMR) while achieving a similar level of fault resiliency.
{"title":"Enhancing Fault Resilience of QNNs by Selective Neuron Splitting","authors":"Mohammad Hasan Ahmadilivani, Mahdi Taheri, J. Raik, M. Daneshtalab, M. Jenihhin","doi":"10.1109/AICAS57966.2023.10168633","DOIUrl":"https://doi.org/10.1109/AICAS57966.2023.10168633","url":null,"abstract":"The superior performance of Deep Neural Networks (DNNs) has led to their application in various aspects of human life. Safety-critical applications are no exception and impose rigorous reliability requirements on DNNs. Quantized Neural Networks (QNNs) have emerged to tackle the complexity of DNN accelerators, however, they are more prone to reliability issues.In this paper, a recent analytical resilience assessment method is adapted for QNNs to identify critical neurons based on a Neuron Vulnerability Factor (NVF). Thereafter, a novel method for splitting the critical neurons is proposed that enables the design of a Lightweight Correction Unit (LCU) in the accelerator without redesigning its computational part.The method is validated by experiments on different QNNs and datasets. The results demonstrate that the proposed method for correcting the faults has a twice smaller overhead than a selective Triple Modular Redundancy (TMR) while achieving a similar level of fault resiliency.","PeriodicalId":296649,"journal":{"name":"2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132951417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：