Pub Date : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319491
Zhiyi Gao, A. Doi, Toru Kato, H. Takahashi, Kenji Sakakibara, Tomonori Hosokawa, Masahiro Harada
In this study, we propose a method to automatically remove utility poles and electric parts from 3D measurement data of shopping streets and roads, and be used for the landscape simulation of underground utility poles construction at Suehiro-cho of Miyako City. In this method, utility poles were stably extracted automatically by using plural rectangular volumes, Hough transform, and two-dimensional Gaussian function. The efficiency of processing is improved by five times compared with the interactive method of removing the utility poles and the electric parts by the dialogue processing.
{"title":"Utility pole extraction processing from point cloud data from 3D measurement and its applications","authors":"Zhiyi Gao, A. Doi, Toru Kato, H. Takahashi, Kenji Sakakibara, Tomonori Hosokawa, Masahiro Harada","doi":"10.1109/iCAST51195.2020.9319491","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319491","url":null,"abstract":"In this study, we propose a method to automatically remove utility poles and electric parts from 3D measurement data of shopping streets and roads, and be used for the landscape simulation of underground utility poles construction at Suehiro-cho of Miyako City. In this method, utility poles were stably extracted automatically by using plural rectangular volumes, Hough transform, and two-dimensional Gaussian function. The efficiency of processing is improved by five times compared with the interactive method of removing the utility poles and the electric parts by the dialogue processing.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"66 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123768896","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319477
Xiang Li, B. Wei, H. Wu, Xuzhou Li, Jinyu Cong
Migraine is a brain disease that seriously endangers human health in which migraine without aura accounts for the largest proportion in the clinic and is challenging to diagnose. Currently, the auxiliary diagnosis methods based on functional connectivity analysis combined with machine learning algorithms is an important research domain for migraine without aura. Although a few earlier studies have made significant progress, it is still hard to meet the clinical and research needs. The main reason is that the functional connectivity analysis methods mostly rely on the prior template, which is easily affected by subjective factors and the performance of the classifier, the intelligence and accuracy are still at a low level. In this paper, we propose an intelligent auxiliary diagnosis algorithm for migraine without aura based on improved 3D convolutional neural network dubbed MwoA3D-Net. To avoid the difference results caused by varying prior templates, a group information guided independent component analysis method is employed to obtain the resting state network for training the MwoA3D-Net algorithm. Subsequently, the MwoA3D-Net algorithm is applied to diagnose migraine without aura patients and healthy controls automatically. Several optimization strategies, such as 3D data augmentation and L2 regularization, are introduced to prevent overfitting effectively. Experimental results on a data set of 65 migraine without aura patients and 60 healthy subjects show that MwoA3D-Net has a highly robust performance, with an average diagnostic accuracy of 98.40%. Furthermore, the selected resting-state brain function network has robust identification and can be adopted as potential biomarkers of migraine without aura toward individualized diagnosis.
{"title":"MwoA auxiliary diagnosis using 3D convolutional neural network","authors":"Xiang Li, B. Wei, H. Wu, Xuzhou Li, Jinyu Cong","doi":"10.1109/iCAST51195.2020.9319477","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319477","url":null,"abstract":"Migraine is a brain disease that seriously endangers human health in which migraine without aura accounts for the largest proportion in the clinic and is challenging to diagnose. Currently, the auxiliary diagnosis methods based on functional connectivity analysis combined with machine learning algorithms is an important research domain for migraine without aura. Although a few earlier studies have made significant progress, it is still hard to meet the clinical and research needs. The main reason is that the functional connectivity analysis methods mostly rely on the prior template, which is easily affected by subjective factors and the performance of the classifier, the intelligence and accuracy are still at a low level. In this paper, we propose an intelligent auxiliary diagnosis algorithm for migraine without aura based on improved 3D convolutional neural network dubbed MwoA3D-Net. To avoid the difference results caused by varying prior templates, a group information guided independent component analysis method is employed to obtain the resting state network for training the MwoA3D-Net algorithm. Subsequently, the MwoA3D-Net algorithm is applied to diagnose migraine without aura patients and healthy controls automatically. Several optimization strategies, such as 3D data augmentation and L2 regularization, are introduced to prevent overfitting effectively. Experimental results on a data set of 65 migraine without aura patients and 60 healthy subjects show that MwoA3D-Net has a highly robust performance, with an average diagnostic accuracy of 98.40%. Furthermore, the selected resting-state brain function network has robust identification and can be adopted as potential biomarkers of migraine without aura toward individualized diagnosis.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129587309","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319487
Mohammad Agus Prihandono, R. Harwahyu, R. F. Sari
Incident Management is a part of managing IT services, improving services, and achieving organizational goals. IT incidents can be learned and predicted future incidents. This research compares the factors that cause incidents using initial machine learning techniques such as Random Forest, SVM, Multilayer perceptron, and the latest machine learning techniques such as RNN, LSTM, GRU, to predict IT incidents. Grid search is used to find the optimal parameter combination. 5-fold and 10-fold Cross-validation evaluates the model's optimal performance by dividing the dataset into training data and test data. The results show that the highest accuracy of 98.866% is produced by LSTM machine learning techniques at 5-fold and 10-fold cross-validation. SVM has the lowest accuracy of 97.837% made at 5-fold and 10-fold cross-validation.
{"title":"Performance of Machine Learning Algorithms for IT Incident Management","authors":"Mohammad Agus Prihandono, R. Harwahyu, R. F. Sari","doi":"10.1109/iCAST51195.2020.9319487","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319487","url":null,"abstract":"Incident Management is a part of managing IT services, improving services, and achieving organizational goals. IT incidents can be learned and predicted future incidents. This research compares the factors that cause incidents using initial machine learning techniques such as Random Forest, SVM, Multilayer perceptron, and the latest machine learning techniques such as RNN, LSTM, GRU, to predict IT incidents. Grid search is used to find the optimal parameter combination. 5-fold and 10-fold Cross-validation evaluates the model's optimal performance by dividing the dataset into training data and test data. The results show that the highest accuracy of 98.866% is produced by LSTM machine learning techniques at 5-fold and 10-fold cross-validation. SVM has the lowest accuracy of 97.837% made at 5-fold and 10-fold cross-validation.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128410577","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319492
K. Yamaguchi
SHAP is a measurement based on Shapley values and has been used widely in machine-learning regressions. In the paper, I describe the intrinsic meaning of SHAP values and I propose that the SHAP was a better measurement for the performance evaluation of a company in the same industry, compared with a raw variable value such as ROE. In my regression analysis of company performance, I found that a linear relationship appeared between the target values and the SHAP values of the predictor variables, even when there was no linear relationship between the target values and the raw predictor values. This visualization of the relationships made us notice the intrinsic meaning and potential of SHAP values. In the SHAP calculation process, through each company's characteristics, how effective a predictor value works to increase the target value within the company is evaluated. The utility of the predictor depends on the individual company's characteristics. Because the individual company's characteristics are used as the characteristic function, the linear relationship could be extracted.
{"title":"Intrinsic Meaning of Shapley Values in Regression","authors":"K. Yamaguchi","doi":"10.1109/iCAST51195.2020.9319492","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319492","url":null,"abstract":"SHAP is a measurement based on Shapley values and has been used widely in machine-learning regressions. In the paper, I describe the intrinsic meaning of SHAP values and I propose that the SHAP was a better measurement for the performance evaluation of a company in the same industry, compared with a raw variable value such as ROE. In my regression analysis of company performance, I found that a linear relationship appeared between the target values and the SHAP values of the predictor variables, even when there was no linear relationship between the target values and the raw predictor values. This visualization of the relationships made us notice the intrinsic meaning and potential of SHAP values. In the SHAP calculation process, through each company's characteristics, how effective a predictor value works to increase the target value within the company is evaluated. The utility of the predictor depends on the individual company's characteristics. Because the individual company's characteristics are used as the characteristic function, the linear relationship could be extracted.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122789289","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319488
Mingze Shi, Qiangfu Zhao
This research is mainly about the prediction of the price change in the stock market. Instead of daily change, this paper analyzes the trend of price change for weeks by judging turning points. Deep neural networks will be used as the classifier of true and fake golden crosses to judge the growth trend of price change. Most stocks on the sample list have positive profits after simulated trading of 10 years. Based on the results we may conclude that deep neural networks are helpful to assist users positively for stock investment.
{"title":"Stock Market Trend Prediction and Investment Strategy by Deep Neural Networks","authors":"Mingze Shi, Qiangfu Zhao","doi":"10.1109/iCAST51195.2020.9319488","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319488","url":null,"abstract":"This research is mainly about the prediction of the price change in the stock market. Instead of daily change, this paper analyzes the trend of price change for weeks by judging turning points. Deep neural networks will be used as the classifier of true and fake golden crosses to judge the growth trend of price change. Most stocks on the sample list have positive profits after simulated trading of 10 years. Based on the results we may conclude that deep neural networks are helpful to assist users positively for stock investment.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128290346","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319473
Xiao Zeng, Lin Qi, Tong Yi, Tong Liu
Ocean wave spectrum are mainly estimated from wave surface records, and traditional methods involve physically-based modeling of wave dynamics and frequency analysis. In this paper, we employ machine learning strategy and propose a Seq2Seq (sequence to sequence) model which connects the encoder and decoder with an attention mechanism. This model can both effectively predict wave spectrum and be easily implemented. Experiments on numerical simulations show the feasibility of the proposed model in wave spectrum estimation and the accuracy comparing with traditional methods.
海浪谱的估计主要来自于波面记录，传统的方法包括基于物理的波浪动力学建模和频率分析。在本文中，我们采用机器学习策略并提出了一个Seq2Seq (sequence to sequence)模型，该模型通过注意机制将编码器和解码器连接起来。该模型既能有效地预测波浪谱，又易于实现。数值模拟实验证明了该模型在波浪谱估计方面的可行性和与传统方法相比的准确性。
{"title":"A Sequence-to-Sequence Model Based on Attention Mechanism for Wave Spectrum Prediction","authors":"Xiao Zeng, Lin Qi, Tong Yi, Tong Liu","doi":"10.1109/iCAST51195.2020.9319473","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319473","url":null,"abstract":"Ocean wave spectrum are mainly estimated from wave surface records, and traditional methods involve physically-based modeling of wave dynamics and frequency analysis. In this paper, we employ machine learning strategy and propose a Seq2Seq (sequence to sequence) model which connects the encoder and decoder with an attention mechanism. This model can both effectively predict wave spectrum and be easily implemented. Experiments on numerical simulations show the feasibility of the proposed model in wave spectrum estimation and the accuracy comparing with traditional methods.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116648506","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319486
Xiang Li, B. Wei, Tianyang Li, N. Zhang
Migraine without aura (MwoA) is the most typical migraine disease in the clinic, which is endangered to human health and challenging to diagnose. Developing the auxiliary diagnosis algorithms of MwoA based on functional connectivity (FC) changes from resting-state functional magnetic resonance imaging (rs-fMRI) is an important research domain. However, existing auxiliary diagnostic methods mainly adopt a seed-based correlation method to extract FC, which are easily affected by subjective factors. Moreover, those methods neglect the relationship between changes in FC and disease duration. In this paper, we report a weakly supervised learning method aiming to tackle those issues. We propose a resting-state brain network-based 3D deep multiple instance learning with spatial attention mechanism (R3D-DMILSAM) framework, where the patient-level label is allocated to the rs-fMRI data that view as multiple instances of a bag. R3D-DMILSAM uses the group information guided independent component analysis (GIG-ICA) to generate the subject-specific resting-state brain networks (RSNs). After that, the designed spatial attention-based 3D deep multiple instance learning (SA3D-DMIL) is trained to perform the diagnosis of MwoA. SA3D-DMIL can automatically generate several semantic deep instances and discovers abnormal RSNs using spatial attention mechanism. Extensive experimental results on the MwoA dataset show that R3D-DMILSAM achieves an overall accuracy of 88.80% and AUC of 94.70%. The visual network obtains high weight, which could be used as a potential biomarker for individualized diagnosis of MwoA.
{"title":"MwoA auxiliary diagnosis via RSN-based 3D deep multiple instance learning with spatial attention mechanism","authors":"Xiang Li, B. Wei, Tianyang Li, N. Zhang","doi":"10.1109/iCAST51195.2020.9319486","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319486","url":null,"abstract":"Migraine without aura (MwoA) is the most typical migraine disease in the clinic, which is endangered to human health and challenging to diagnose. Developing the auxiliary diagnosis algorithms of MwoA based on functional connectivity (FC) changes from resting-state functional magnetic resonance imaging (rs-fMRI) is an important research domain. However, existing auxiliary diagnostic methods mainly adopt a seed-based correlation method to extract FC, which are easily affected by subjective factors. Moreover, those methods neglect the relationship between changes in FC and disease duration. In this paper, we report a weakly supervised learning method aiming to tackle those issues. We propose a resting-state brain network-based 3D deep multiple instance learning with spatial attention mechanism (R3D-DMILSAM) framework, where the patient-level label is allocated to the rs-fMRI data that view as multiple instances of a bag. R3D-DMILSAM uses the group information guided independent component analysis (GIG-ICA) to generate the subject-specific resting-state brain networks (RSNs). After that, the designed spatial attention-based 3D deep multiple instance learning (SA3D-DMIL) is trained to perform the diagnosis of MwoA. SA3D-DMIL can automatically generate several semantic deep instances and discovers abnormal RSNs using spatial attention mechanism. Extensive experimental results on the MwoA dataset show that R3D-DMILSAM achieves an overall accuracy of 88.80% and AUC of 94.70%. The visual network obtains high weight, which could be used as a potential biomarker for individualized diagnosis of MwoA.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"193 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124297795","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319483
Jiahao Xu, Jian Li, Shu Zhang, Cui Xie, Junyu Dong
The skeleton analysis introduced by Kinect has been an efficient way for interaction. Skeleton analysis interaction is more intuitive and aligns better with human natural behaviors compared with traditional approaches. However, skeleton analysis often has the problem of producing conflict gesture identifications if two interaction movements are similar. Additionally, it always mistakenly recognizes some unconscious or intentional body movements as positive gestures. To this end, we proposed a new interaction method enhanced by both vision algorithms and deep learning. An improved residual neural network is employed to recognize gestures which are then used for distinguishing similar body movements. A combined human-computer interaction scheme is proposed which includes three main components: (a) a hand shape segmentation approach enhanced by skin color detection and skeleton joint tracking, (b) the deep learning augmented detection for changes of gestures and (c) a deep learning-based gesture command recognition for robot control. Experiments are conducted using the proposed method for robot interaction. The results demonstrate that unconscious body movements can be accurately identified. Similar body movements can also be distinguished robustly. The proposed method can run in real-time with competitive performance.
{"title":"Skeleton Guided Conflict-Free Hand Gesture Recognition for Robot Control","authors":"Jiahao Xu, Jian Li, Shu Zhang, Cui Xie, Junyu Dong","doi":"10.1109/iCAST51195.2020.9319483","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319483","url":null,"abstract":"The skeleton analysis introduced by Kinect has been an efficient way for interaction. Skeleton analysis interaction is more intuitive and aligns better with human natural behaviors compared with traditional approaches. However, skeleton analysis often has the problem of producing conflict gesture identifications if two interaction movements are similar. Additionally, it always mistakenly recognizes some unconscious or intentional body movements as positive gestures. To this end, we proposed a new interaction method enhanced by both vision algorithms and deep learning. An improved residual neural network is employed to recognize gestures which are then used for distinguishing similar body movements. A combined human-computer interaction scheme is proposed which includes three main components: (a) a hand shape segmentation approach enhanced by skin color detection and skeleton joint tracking, (b) the deep learning augmented detection for changes of gestures and (c) a deep learning-based gesture command recognition for robot control. Experiments are conducted using the proposed method for robot interaction. The results demonstrate that unconscious body movements can be accurately identified. Similar body movements can also be distinguished robustly. The proposed method can run in real-time with competitive performance.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114467789","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319495
Zhihao Zhu, Jian Yang, Xianglong Wang, Guanqi Qi, Chuang Wu, H. Fan, Lin Qi, Junyu Dong
The laser line rotating-scan system has been a popular method for 3D reconstruction. But the rotating axis of the scanner usually does not coincide with the camera origin point, which makes this technique a bit more complex. This paper proposes a rotation axis calibration method to solve the problem for more accurate 3D imaging. The calibration parameters are defined as the external parameters (rotation and translation) between the camera coordinate system and the world coordinate system of the rotation axis. With the rotation axis calibration, 3D results of a single frame are transformed into the world coordinate system of the rotation axis. Finally, we realize multi-frame splicing only with the rotation angle provided by the scanner. Experiments on real data show our method is basically correct and practical.
{"title":"Rotation Axis Calibration of Laser Line Rotating-Scan System for 3D Reconstruction","authors":"Zhihao Zhu, Jian Yang, Xianglong Wang, Guanqi Qi, Chuang Wu, H. Fan, Lin Qi, Junyu Dong","doi":"10.1109/iCAST51195.2020.9319495","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319495","url":null,"abstract":"The laser line rotating-scan system has been a popular method for 3D reconstruction. But the rotating axis of the scanner usually does not coincide with the camera origin point, which makes this technique a bit more complex. This paper proposes a rotation axis calibration method to solve the problem for more accurate 3D imaging. The calibration parameters are defined as the external parameters (rotation and translation) between the camera coordinate system and the world coordinate system of the rotation axis. With the rotation axis calibration, 3D results of a single frame are transformed into the world coordinate system of the rotation axis. Finally, we realize multi-frame splicing only with the rotation angle provided by the scanner. Experiments on real data show our method is basically correct and practical.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123909504","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 : 2020-12-07DOI: 10.1109/iCAST51195.2020.9319476
Yosuke Sato, Jianwei Zhang
In recent years, investment strategies on financial markets using deep learning have attracted a significant amount of research attention. The objective of these studies is to obtain investment action that has a low risk and increases profit. On the other hand, Distributional Reinforcement Learning (DRL) expands the action-value function to a discrete distribution in reinforcement learning, which can control risk. However, DRL has not yet been used to learn investment action. In this study, we construct a low-risk investment trading model using DRL. This model is backtested on Nikkei 225 dataset and compared with Deep Q Network (DQN). We evaluate performance in terms of final asset amounts, their standard deviation, and the Sharpe ratio. The experimental results show that the proposed DRL-based method can learn low-risk actions with increasing profit, outperforming the compared method DQN.
{"title":"Modeling Low-risk Actions from Multivariate Time Series Data Using Distributional Reinforcement Learning","authors":"Yosuke Sato, Jianwei Zhang","doi":"10.1109/iCAST51195.2020.9319476","DOIUrl":"https://doi.org/10.1109/iCAST51195.2020.9319476","url":null,"abstract":"In recent years, investment strategies on financial markets using deep learning have attracted a significant amount of research attention. The objective of these studies is to obtain investment action that has a low risk and increases profit. On the other hand, Distributional Reinforcement Learning (DRL) expands the action-value function to a discrete distribution in reinforcement learning, which can control risk. However, DRL has not yet been used to learn investment action. In this study, we construct a low-risk investment trading model using DRL. This model is backtested on Nikkei 225 dataset and compared with Deep Q Network (DQN). We evaluate performance in terms of final asset amounts, their standard deviation, and the Sharpe ratio. The experimental results show that the proposed DRL-based method can learn low-risk actions with increasing profit, outperforming the compared method DQN.","PeriodicalId":212570,"journal":{"name":"2020 11th International Conference on Awareness Science and Technology (iCAST)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133515891","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：