Parameters Compressing in Deep Learning

He, Shiming; Li, Zhuozhou; Tang, Yangning; Liao, Zhuofan; Li, Feng; Lim, Se-Jung

doi:10.32604/cmc.2020.06130

Cited by 74 publications

(54 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These feature maps will be fused with the feature maps extracted by Resnet for classification. From these works [2,13,30], we find that the additional modules used at the inference stage bring about a performance improvement, but also bring about more parameters and complexity [33][34][35], which may be unfavorable for deployment on devices with limited memory and computational resources [36]. Therefore, it is necessary to explore how to improve the classification performance of the model without increasing the parameters.…”

Section: Motivationmentioning

confidence: 99%

Training Convolutional Neural Networks with Multi-Size Images and Triplet Loss for Remote Sensing Scene Classification

Zhang

Wang

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

Many remote sensing scene classification algorithms improve their classification accuracy by additional modules, which increases the parameters and computing overhead of the model at the inference stage. In this paper, we explore how to improve the classification accuracy of the model without adding modules at the inference stage. First, we propose a network training strategy of training with multi-size images. Then, we introduce more supervision information by triplet loss and design a branch for the triplet loss. In addition, dropout is introduced between the feature extractor and the classifier to avoid over-fitting. These modules only work at the training stage and will not bring about the increase in model parameters at the inference stage. We use Resnet18 as the baseline and add the three modules to the baseline. We perform experiments on three datasets: AID, NWPU-RESISC45, and OPTIMAL. Experimental results show that our model combined with the three modules is more competitive than many existing classification algorithms. In addition, ablation experiments on OPTIMAL show that dropout, triplet loss, and training with multi-size images improve the overall accuracy of the model on the test set by 0.53%, 0.38%, and 0.7%, respectively. The combination of the three modules improves the overall accuracy of the model by 1.61%. It can be seen that the three modules can improve the classification accuracy of the model without increasing model parameters at the inference stage, and training with multi-size images brings a greater gain in accuracy than the other two modules, but the combination of the three modules will be better.

show abstract

Section: Motivationmentioning

confidence: 99%

Training Convolutional Neural Networks with Multi-Size Images and Triplet Loss for Remote Sensing Scene Classification

Zhang

Wang

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…In [9], a method for constructing a state-space model of a building is presented, which can be used to predict indoor temperature, humidity, and thermal comfort to control the indoor environment with MPC. With the rapid development of machine learning [10,11] and data mining technology [12,13], especially deep learning [14][15][16] and reinforcement learning [8,17], the HVAC of a building structure will become smarter based on historical data such as temperature, humidity, and energy consumption.…”

Section: Related Workmentioning

confidence: 99%

Modeling and Controlling of Temperature and Humidity in Building Heating, Ventilating, and Air Conditioning System Using Model Predictive Control

et al. 2019

View full text Add to dashboard Cite

Nowadays, by huge improvements in industrial control and the necessity of efficient energy consumption for buildings, unified managing systems are established to monitor and control mechanical equipment and energy usage. One of the main portions of the building management system (BMS) is the cooling and heating equipment called heating and ventilation and air-conditioning (HVAC). Based on temperature slow dynamic and presented uncertainty in modeling, a model predictive control (MPC) strategy to track both temperature and humidity is proposed in this study. The main goal of this study is to provide a framework to describe temperature and humidity elements required for dynamic modeling. Following that, by utilizing a predictive approach, a control strategy is obtained, which optimizes the tracking error of two interactional channel and performs the effort control by minimizing the optimization index. Other articles have mostly only had control over the temperature variable, but in our article, we tried to study the equations of temperature and humidity as well as their interference and according to the ASHRAE standard, both temperature and humidity controls must be accurate. The humidity was the novelty in our article. Simulation results proved the effectiveness of the proposed approach compared to the conventional proportional-integral controller. Evidently, the key idea behind the control objective is providing the comfort condition while consuming the least possible energy.

show abstract

“…This efficiency decreases to 20% on the ground using the emergency power unit and also with noise pollution and gaseous emissions [1]. Now in the era of artificial intelligence, deep learning has achieved great success [2][3][4], because it requires very little engineering by hand, so it can easily use the increase of available computation and data. In order to monitor and track [5][6][7] the surrounding objects in real time, the intelligent algorithm used in aircraft will consume a lot of power.…”

Section: Motivationmentioning

confidence: 99%

Effect of Load Priority Modeling on the Size of Fuel Cell as an Emergency Power Unit in a More-Electric Aircraft

Salehpour

Radmanesh

Rostami³

et al. 2019

Applied Sciences

View full text Add to dashboard Cite

The proton exchange membrane fuel cell as a green power source is a suitable replacement of the engine mounted generators in the emergency power unit of a more-electric aircraft. Most existing energy management methods for operation of fuel cells in the more-electric aircraft refer to the hydrogen consumption minimization. But due to the increasing number of electrical components and hence electrical demand in the aircraft, demand-side management should be considered in these methods. In order to determine the effect of demand-side management on the fuel cell operation and size, an efficient load priority model is presented and integrated into an optimization framework. The proposed optimization framework is formulated as mixed-integer quadratic programming using Karush–Kuhn–Tucker optimality condition and is solved by CPLEX optimization tool. The Boeing 787 electrical distribution system is considered as a single-bus case study to evaluate the performance of the proposed optimization framework. Numerical results show that the size of fuel cell as an emergency power unit resource depends on the type and importance of the system’s loads in different emergency conditions. Also, with an efficient priority model, both hydrogen consumption and load shedding can be decreased.

show abstract

Parameters Compressing in Deep Learning

Cited by 74 publications

References 0 publications

Training Convolutional Neural Networks with Multi-Size Images and Triplet Loss for Remote Sensing Scene Classification

Training Convolutional Neural Networks with Multi-Size Images and Triplet Loss for Remote Sensing Scene Classification

Modeling and Controlling of Temperature and Humidity in Building Heating, Ventilating, and Air Conditioning System Using Model Predictive Control

Effect of Load Priority Modeling on the Size of Fuel Cell as an Emergency Power Unit in a More-Electric Aircraft

Contact Info

Product

Resources

About