• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.5
• /
• pp.63-68
• /
• 2022

Autonomous driving technology has recently made great progress with the introduction of deep reinforcement learning. In order to effectively use deep reinforcement learning, it is important to select the appropriate activation function. In the meantime, many activation functions have been presented, but they show different performance depending on the environment to be applied. This paper compares and evaluates the performance of 12 activation functions to see which activation functions are effective when using reinforcement learning to learn autonomous driving on highways. To this end, a performance evaluation method was presented and the average reward value of each activation function was compared. As a result, when using GELU, the highest average reward could be obtained, and SiLU showed the lowest performance. The average reward difference between the two activation functions was 20%.

• International Journal of Internet, Broadcasting and Communication
• /
• v.14 no.1
• /
• pp.136-141
• /
• 2022

A very complex task in deep learning such as image classification must be solved with the help of neural networks and activation functions. The backpropagation algorithm advances backward from the output layer towards the input layer, the gradients often get smaller and smaller and approach zero which eventually leaves the weights of the initial or lower layers nearly unchanged, as a result, the gradient descent never converges to the optimum. We propose a two-factor non-saturating activation functions known as Bea-Mish for machine learning applications in deep neural networks. Our method uses two factors, beta (𝛽) and alpha (𝛼), to normalize the area below the boundary in the Mish activation function and we regard these elements as Bea. Bea-Mish provide a clear understanding of the behaviors and conditions governing this regularization term can lead to a more principled approach for constructing better performing activation functions. We evaluate Bea-Mish results against Mish and Swish activation functions in various models and data sets. Empirical results show that our approach (Bea-Mish) outperforms native Mish using SqueezeNet backbone with an average precision (AP50val) of 2.51% in CIFAR-10 and top-1accuracy in ResNet-50 on ImageNet-1k. shows an improvement of 1.20%.

• 한국정보디스플레이학회:학술대회논문집
• /
• 2003.07a
• /
• pp.907-909
• /
• 2003

We report a novel method of activation-annealing, named as induction annealing (IA). IA is realized by applying alternating electric field induced by alternatingmagnetic filed applied to the sample. We observed the enhanced kinetics of dopant activation by using IA.

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 1999.05a
• /
• pp.473-476
• /
• 1999

PMOS capacitors with Ce-policide electrode were fabricated by the SADS method to study the effects of activation condition on the C-V characteristics. For the activation temperature of $600^{\circ}C$ , the capacitor using CoSi$_2$ formed from Co/Ti bilayer as diffusion source showed excellent C-V properties and the increase in V$_{th}$ with the increasing activation time. But impurties into the oxide.e.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.9
• /
• pp.371-380
• /
• 2022

Convolutional neural networks are widely used to manipulate data arranged in a grid, such as images. A general convolutional neural network consists of a convolutional layers and a fully connected layers, and each layer contains a nonlinear activation functions. This paper proposes a combined parametric activation function to improve the performance of convolutional neural networks. The combined parametric activation function is created by adding the parametric activation functions to which parameters that convert the scale and location of the activation function are applied. Various nonlinear intervals can be created according to parameters that convert multiple scales and locations, and parameters can be learned in the direction of minimizing the loss function calculated by the given input data. As a result of testing the performance of the convolutional neural network using the combined parametric activation function on the MNIST, Fashion MNIST, CIFAR10 and CIFAR100 classification problems, it was confirmed that it had better performance than other activation functions.

• Transactions of the Korean hydrogen and new energy society
• /
• v.34 no.2
• /
• pp.178-186
• /
• 2023

Three activation methods (constant voltage, current cycling, and hydrogen pumping) were applied to investigate the effects on the performance of the membrane electrode assembly (MEA) loaded with PtCo/C catalyst. The current cycling protocol took the shortest time to activate the MEA, while the performance after activation was the worst among the all activation methods. The constant voltage method took a moderate activation time and exhibited the best performance after activation. The hydrogen pumping protocol took the longest time to activate the MEA with moderate performance after activation. According to the distribution of relaxation time analysis, the improved performance after the activation mainly comes from the decrease of charge transfer resistance rather than the ionic resistance in the cathode catalyst layer, which suggests that the existence of water on the electrode is the key factor for activation.

• Journal of Environmental Science International
• /
• v.25 no.2
• /
• pp.281-288
• /
• 2016

After analyzing of heating value of four kinds of RDF, the RDF-D has the highest heating value, was chosen to be mixed with carbonized sludge by different ratio. The 85%:15% ratio, which has the highest efficiency, was analyzed with thermogravimetric and pyrolysis kinetics. Applying of Kissinger method, activation energy was obtained from slope which is calculated from relation of ln(${\beta}/T^2{_m}$) and $1/T_m$. The kinetic parameters obtained from Kissinger method were 46.06 kJ/mol of RDF, 55.99 kJ/mol of carbonized sludge and 40.68 kJ/mol of mixture of RDF and carbonized sludge. The mixture of RDF and carbonized sludge has the lowest activation energy and frequency factor, during thermal decomposition reaction it has the slowest reaction rate and needs the lowest energy. Although activation energy with pyrolysis of RDF was irregularly scattered, it showed that activation energy was stabilized by co-pyrolysis of RDF and additives(Carbonized Sludge).

• Journal of the Korean Vacuum Society
• /
• v.17 no.6
• /
• pp.544-548
• /
• 2008

At the semiconductor industry, Photoresist(PR) strip progress has high cost and time consuming process. Accordingly, many research group have been focused on the shortening of the PR strip progress. But the replacements of newly developed materials rather than normally used strip have accompanied by cost consumption. Therefore, we suggested the Plasma Liquid-Vapor Activation (PLVA) method of general PR strip solution for saving the PR strip time and the high strip rate of PR residue. The PLVA method was very effective for PR strip progress. Also, the ion damaged PR(high dose implanted photoresist: HDI-PR) was almost impossible to strip. However, it was very difficult to characterize the change of chemical composition of HDI-PR between with and without PLVA method. Thus, physical properties of HDI-PR surface with and without PLVA method were measured by using the nano-indenter system.

• Journal of KIISE:Software and Applications
• /
• v.31 no.4
• /
• pp.505-515
• /
• 2004

In a newspaper, 5W1H information is the most fundamental and important element for writing and understanding articles. Focusing on such a relation between a newspaper article and the 5W1H, we propose a summarization method based on the activation degree of 5W1H. To overcome problems of the lead-based and the title-based methods, both of which are known to be the most effective in newspaper summarization, sufficient 5W1H information is extracted from both a title and a lead sentence. Moreover, for each sentence, its weight is computed by considering various factors, such as activation degree of 5W1H, the number of 5W1H categories, and its length and position. These factors make a great contribution to the selection of more important sentences, and thus to the improvement of readability of the summarized texts. In an experimental evaluation, the proposed method achieved a precision of 74.7% outperforming the lead-based method. In sum, our 5W1H approach was shown to be promising for automatic summarization of newspaper articles.

• The Korean Journal of Rheology
• /
• v.10 no.4
• /
• pp.227-233
• /
• 1998

The effects of 1 mol% N-benzylpyrazinium hexafluoroantimonate(BPH) as a thermal latent initiator and blend compositions composed of cycloaliphatic and DGEBA epoxies were investigated in the rheological properties and cure kinetics. Latent properties were performed by measurement of the conversion as a function of reaction time using isothermal DSC at $150^{\circ}C$ and $50^{\circ}C$ Rheological properties of the blend systems were investigated in terms of isothermal experiments using a rheometer. The gelation time was obtained from the evaluation of storage modulus (G'), loss modulus (G") and damping factor (tan$\delta$)). Cross-linking activation energy ($E_c$) was also determined from the Arrhenius equation based on gel time and curing temperature. As a result, the gel time and cross-linking activation energy increased with increasing DGEBA composition. The cure activation energies ($E_a$) were obtained by Kissinger method using dynamic DSC thermograms. In this work, the cure activation energy decreased with increasing CAE concentration, which might be resulted from the short repeat units, simple side-groups and viscosity of reaction media.edia.