Energy supply is one of the most critical challenges of wireless sensor networks(WSNs)and industrial wireless sensor networks(IWSNs).While research on coverage optimization problem(COP)centers on the network’s monito...Energy supply is one of the most critical challenges of wireless sensor networks(WSNs)and industrial wireless sensor networks(IWSNs).While research on coverage optimization problem(COP)centers on the network’s monitoring coverage,this research focuses on the power banks’energy supply coverage.The study of 2-D and 3-D spaces is typical in IWSN,with the realistic environment being more complex with obstacles(i.e.,machines).A 3-D surface is the field of interest(FOI)in this work with the established hybrid power bank deployment model for the energy supply COP optimization of IWSN.The hybrid power bank deployment model is highly adaptive and flexible for new or existing plants already using the IWSN system.The model improves the power supply to a more considerable extent with the least number of power bank deployments.The main innovation in this work is the utilization of a more practical surface model with obstacles and training while improving the convergence speed and quality of the heuristic algorithm.An overall probabilistic coverage rate analysis of every point on the FOI is provided,not limiting the scope to target points or areas.Bresenham’s algorithm is extended from 2-D to 3-D surface to enhance the probabilistic covering model for coverage measurement.A dynamic search strategy(DSS)is proposed to modify the artificial bee colony(ABC)and balance the exploration and exploitation ability for better convergence toward eliminating NP-hard deployment problems.Further,the cellular automata(CA)is utilized to enhance the convergence speed.The case study based on two typical FOI in the IWSN shows that the CA scheme effectively speeds up the optimization process.Comparative experiments are conducted on four benchmark functions to validate the effectiveness of the proposed method.The experimental results show that the proposed algorithm outperforms the ABC and gbest-guided ABC(GABC)algorithms.The results show that the proposed energy coverage optimization method based on the hybrid power bank deployment model generates more accurate results than the results obtained by similar algorithms(i.e.,ABC,GABC).The proposed model is,therefore,effective and efficient for optimization in the IWSN.展开更多
The quantization algorithm compresses the original network by reducing the numerical bit width of the model,which improves the computation speed. Because different layers have different redundancy and sensitivity to d...The quantization algorithm compresses the original network by reducing the numerical bit width of the model,which improves the computation speed. Because different layers have different redundancy and sensitivity to databit width. Reducing the data bit width will result in a loss of accuracy. Therefore, it is difficult to determinethe optimal bit width for different parts of the network with guaranteed accuracy. Mixed precision quantizationcan effectively reduce the amount of computation while keeping the model accuracy basically unchanged. In thispaper, a hardware-aware mixed precision quantization strategy optimal assignment algorithm adapted to low bitwidth is proposed, and reinforcement learning is used to automatically predict the mixed precision that meets theconstraints of hardware resources. In the state-space design, the standard deviation of weights is used to measurethe distribution difference of data, the execution speed feedback of simulated neural network accelerator inferenceis used as the environment to limit the action space of the agent, and the accuracy of the quantization model afterretraining is used as the reward function to guide the agent to carry out deep reinforcement learning training. Theexperimental results show that the proposed method obtains a suitable model layer-by-layer quantization strategyunder the condition that the computational resources are satisfied, and themodel accuracy is effectively improved.The proposed method has strong intelligence and certain universality and has strong application potential in thefield of mixed precision quantization and embedded neural network model deployment.展开更多
随着人工智能(AI)技术的蓬勃发展,深度神经网络(DNN)模型被大规模应用到各类移动端与边缘端。然而,边缘端算力低、内存容量小,且实现模型加速需要深入掌握边缘端硬件知识,这增加了模型的部署难度,也限制了模型的推广应用。因此,基于张...随着人工智能(AI)技术的蓬勃发展,深度神经网络(DNN)模型被大规模应用到各类移动端与边缘端。然而,边缘端算力低、内存容量小,且实现模型加速需要深入掌握边缘端硬件知识,这增加了模型的部署难度,也限制了模型的推广应用。因此,基于张量虚拟机(TVM)提出一种DNN加速与部署方法,从而实现卷积神经网络(CNN)模型在现场可编程门阵列(FPGA)上的加速,并在分心驾驶分类应用场景下验证了所提方法的可行性。通过计算图优化方法减小了模型的访存和计算开销,通过模型量化方法减小了模型尺寸,通过计算图打包方法将卷积计算卸载到FPGA上执行以提高模型推理速度。与微处理器(MPU)相比,所提方法可使ResNet50和ResNet18在MPU+FPGA上的推理时间分别减少88.63%和77.53%;而在AUC(American University in Cairo)数据集上,相较于MPU,两个模型在MPU+FPGA上的top1推理精度仅下降了0.26和0.16个百分点。可见,所提方法可以降低不同模型在FPGA上的部署难度。展开更多
文摘Energy supply is one of the most critical challenges of wireless sensor networks(WSNs)and industrial wireless sensor networks(IWSNs).While research on coverage optimization problem(COP)centers on the network’s monitoring coverage,this research focuses on the power banks’energy supply coverage.The study of 2-D and 3-D spaces is typical in IWSN,with the realistic environment being more complex with obstacles(i.e.,machines).A 3-D surface is the field of interest(FOI)in this work with the established hybrid power bank deployment model for the energy supply COP optimization of IWSN.The hybrid power bank deployment model is highly adaptive and flexible for new or existing plants already using the IWSN system.The model improves the power supply to a more considerable extent with the least number of power bank deployments.The main innovation in this work is the utilization of a more practical surface model with obstacles and training while improving the convergence speed and quality of the heuristic algorithm.An overall probabilistic coverage rate analysis of every point on the FOI is provided,not limiting the scope to target points or areas.Bresenham’s algorithm is extended from 2-D to 3-D surface to enhance the probabilistic covering model for coverage measurement.A dynamic search strategy(DSS)is proposed to modify the artificial bee colony(ABC)and balance the exploration and exploitation ability for better convergence toward eliminating NP-hard deployment problems.Further,the cellular automata(CA)is utilized to enhance the convergence speed.The case study based on two typical FOI in the IWSN shows that the CA scheme effectively speeds up the optimization process.Comparative experiments are conducted on four benchmark functions to validate the effectiveness of the proposed method.The experimental results show that the proposed algorithm outperforms the ABC and gbest-guided ABC(GABC)algorithms.The results show that the proposed energy coverage optimization method based on the hybrid power bank deployment model generates more accurate results than the results obtained by similar algorithms(i.e.,ABC,GABC).The proposed model is,therefore,effective and efficient for optimization in the IWSN.
文摘The quantization algorithm compresses the original network by reducing the numerical bit width of the model,which improves the computation speed. Because different layers have different redundancy and sensitivity to databit width. Reducing the data bit width will result in a loss of accuracy. Therefore, it is difficult to determinethe optimal bit width for different parts of the network with guaranteed accuracy. Mixed precision quantizationcan effectively reduce the amount of computation while keeping the model accuracy basically unchanged. In thispaper, a hardware-aware mixed precision quantization strategy optimal assignment algorithm adapted to low bitwidth is proposed, and reinforcement learning is used to automatically predict the mixed precision that meets theconstraints of hardware resources. In the state-space design, the standard deviation of weights is used to measurethe distribution difference of data, the execution speed feedback of simulated neural network accelerator inferenceis used as the environment to limit the action space of the agent, and the accuracy of the quantization model afterretraining is used as the reward function to guide the agent to carry out deep reinforcement learning training. Theexperimental results show that the proposed method obtains a suitable model layer-by-layer quantization strategyunder the condition that the computational resources are satisfied, and themodel accuracy is effectively improved.The proposed method has strong intelligence and certain universality and has strong application potential in thefield of mixed precision quantization and embedded neural network model deployment.
文摘随着人工智能(AI)技术的蓬勃发展,深度神经网络(DNN)模型被大规模应用到各类移动端与边缘端。然而,边缘端算力低、内存容量小,且实现模型加速需要深入掌握边缘端硬件知识,这增加了模型的部署难度,也限制了模型的推广应用。因此,基于张量虚拟机(TVM)提出一种DNN加速与部署方法,从而实现卷积神经网络(CNN)模型在现场可编程门阵列(FPGA)上的加速,并在分心驾驶分类应用场景下验证了所提方法的可行性。通过计算图优化方法减小了模型的访存和计算开销,通过模型量化方法减小了模型尺寸,通过计算图打包方法将卷积计算卸载到FPGA上执行以提高模型推理速度。与微处理器(MPU)相比,所提方法可使ResNet50和ResNet18在MPU+FPGA上的推理时间分别减少88.63%和77.53%;而在AUC(American University in Cairo)数据集上,相较于MPU,两个模型在MPU+FPGA上的top1推理精度仅下降了0.26和0.16个百分点。可见,所提方法可以降低不同模型在FPGA上的部署难度。
