Chatbots are computer programs designed to simulate conversation by interacting with a human user. In this paper we present a chatbot framework designed specifically to aid prolonged grief disorder (PGD) sufferers by replicating the techniques performed during cold readings. Our initial framework performed an association rule analysis on transcripts of real-world cold reading performances, in order to generate the required data as used in traditional rules based chatbots. However due to the structure of cold readings the traditional approach was unable to determine a satisfactory set of rules. Therefore, in this paper we discuss the limitations of this approach and subsequently provide a generative solution using sequence-to-sequence modeling with long short-term memory. We demonstrate how our generative chatbot is therefore able to provide appropriate responses to the majority of inputs. However, as inappropriate responses can present a risk to sensitive PGD sufferers we suggest a final iteration of our chatbot which successfully adjusts to account for multi-turn conversations.
{"title":"Applying NLP to Build a Cold Reading Chatbot","authors":"Peter Tracey, Mohamad Saraee, Chris J. Hughes","doi":"10.1145/3459104.3459119","DOIUrl":"https://doi.org/10.1145/3459104.3459119","url":null,"abstract":"Chatbots are computer programs designed to simulate conversation by interacting with a human user. In this paper we present a chatbot framework designed specifically to aid prolonged grief disorder (PGD) sufferers by replicating the techniques performed during cold readings. Our initial framework performed an association rule analysis on transcripts of real-world cold reading performances, in order to generate the required data as used in traditional rules based chatbots. However due to the structure of cold readings the traditional approach was unable to determine a satisfactory set of rules. Therefore, in this paper we discuss the limitations of this approach and subsequently provide a generative solution using sequence-to-sequence modeling with long short-term memory. We demonstrate how our generative chatbot is therefore able to provide appropriate responses to the majority of inputs. However, as inappropriate responses can present a risk to sensitive PGD sufferers we suggest a final iteration of our chatbot which successfully adjusts to account for multi-turn conversations.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129805631","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}
Ayush Pradhan, Eldhose K Joy, Harsha Jawagal, Sundar Prasad Jayaraman
When it comes to information, Enterprises today are seen as a black hole, a mass of it goes in but gets difficult to extract the practical knowledge out of it. An automated system that has the ability to consume this large mass of information and provide specific, knowledgeable, domain-oriented responses back, will go a long way in unlocking the value of this large-scale unstructured information. In a bid to enrich the answering system's accuracy in Machine Reading Comprehension (MRC), we propose a domain-specific Question Answers (QuAns) framework that specifically aims to auto-generate questions from a domain-based document using an improvised Sequence to Sequence (Seq2Seq) technique equipped with Attention and Copy mechanism. The generated questions are conditioned on a set of candidate answers, derived using a combination of heuristic-driven and graph-based techniques. Further, it also leverages the contextual information by pooling strategy to build an automated response system using a deep custom fine-tuned Bidirectional Encoder Representations from Transformers (BERT) framework and retrieving the top-k contexts for a user query. The evaluation of the QuAns architecture is performed in combination with human supervision as at times, the automated metrics like BLEU, Exact Match (EM), F1 score, etc. fail to gauge the diverse semantic and structural aspects of a generated response. Primarily, the proffered ensemble technique has leveraged the augmented domain knowledge to enrich the answering response efficacy and improving the EM and F1 score by 14.86% and 12.76% respectively over Vanilla BERT architecture. To enhance the user experience, the conversational system is equipped with Natural Language Generation (NLG) to present a human-readable response. Our architectural pipeline aims to provide a one-stop solution for the organizations in processing huge volumes of multidisciplinary data by significantly reducing the human introspection and the overhead cost.
{"title":"A Framework for Leveraging Contextual Information in Automated Domain Specific Comprehension","authors":"Ayush Pradhan, Eldhose K Joy, Harsha Jawagal, Sundar Prasad Jayaraman","doi":"10.1145/3459104.3459148","DOIUrl":"https://doi.org/10.1145/3459104.3459148","url":null,"abstract":"When it comes to information, Enterprises today are seen as a black hole, a mass of it goes in but gets difficult to extract the practical knowledge out of it. An automated system that has the ability to consume this large mass of information and provide specific, knowledgeable, domain-oriented responses back, will go a long way in unlocking the value of this large-scale unstructured information. In a bid to enrich the answering system's accuracy in Machine Reading Comprehension (MRC), we propose a domain-specific Question Answers (QuAns) framework that specifically aims to auto-generate questions from a domain-based document using an improvised Sequence to Sequence (Seq2Seq) technique equipped with Attention and Copy mechanism. The generated questions are conditioned on a set of candidate answers, derived using a combination of heuristic-driven and graph-based techniques. Further, it also leverages the contextual information by pooling strategy to build an automated response system using a deep custom fine-tuned Bidirectional Encoder Representations from Transformers (BERT) framework and retrieving the top-k contexts for a user query. The evaluation of the QuAns architecture is performed in combination with human supervision as at times, the automated metrics like BLEU, Exact Match (EM), F1 score, etc. fail to gauge the diverse semantic and structural aspects of a generated response. Primarily, the proffered ensemble technique has leveraged the augmented domain knowledge to enrich the answering response efficacy and improving the EM and F1 score by 14.86% and 12.76% respectively over Vanilla BERT architecture. To enhance the user experience, the conversational system is equipped with Natural Language Generation (NLG) to present a human-readable response. Our architectural pipeline aims to provide a one-stop solution for the organizations in processing huge volumes of multidisciplinary data by significantly reducing the human introspection and the overhead cost.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133750690","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}
Massimiliano Botticelli, Robin Hellmann, P. Jochmann, K. Stapf, Erik Schuenemann
New emission regulations, the demand of high power output and high efficiency of Gasoline Direct Injection (GDI) engines led to an intense development of new tools and approaches in the study of combustion processes. The generation of the data in this context through simulations and measurements is however an expensive and time-consuming process. Therefore, novel Machine Learning methods can be applied to support GDI developments. In the current paper, an innovative approach regarding the analysis of GDI related data is proposed. Specifically, Extreme Gradient Boosting Machine is chosen due to its high efficiency and powerful feature analysis coming along during the models training. In addition, a parameter-free, fast and dynamic data-driven model selection method is presented. This includes the genetic algorithm NSGA-II to identify the best set of hyperparameters by means of good generalization and precision. The potential of the proposed method is finally demonstrated on real-world data coming from the GDI development field and public data compared with state-of-the-art approaches.
{"title":"Model Selection for Gasoline Direct Injection Characteristics Using Boosting and Genetic Algorithms","authors":"Massimiliano Botticelli, Robin Hellmann, P. Jochmann, K. Stapf, Erik Schuenemann","doi":"10.1145/3459104.3459145","DOIUrl":"https://doi.org/10.1145/3459104.3459145","url":null,"abstract":"New emission regulations, the demand of high power output and high efficiency of Gasoline Direct Injection (GDI) engines led to an intense development of new tools and approaches in the study of combustion processes. The generation of the data in this context through simulations and measurements is however an expensive and time-consuming process. Therefore, novel Machine Learning methods can be applied to support GDI developments. In the current paper, an innovative approach regarding the analysis of GDI related data is proposed. Specifically, Extreme Gradient Boosting Machine is chosen due to its high efficiency and powerful feature analysis coming along during the models training. In addition, a parameter-free, fast and dynamic data-driven model selection method is presented. This includes the genetic algorithm NSGA-II to identify the best set of hyperparameters by means of good generalization and precision. The potential of the proposed method is finally demonstrated on real-world data coming from the GDI development field and public data compared with state-of-the-art approaches.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130146858","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}
In this article, a simple and effective restoration-based haze-removal approach is proposed. This approach is based on refining the course transmission map further by a novel non-linear diffusion patch method. The robustness of the proposed method is validated using quantitative analysis and is compared with other approaches with standard performance metrics. This technique can handle illumination, preserve edges better and ensures the original color of the image is retained. It can be used in many systems for example in object detection and tracking in order to recognize active traffic participants clearly on the road. Other applications include remote sensing for weather prediction, smart cars for smooth navigation and consumer electronics for fault identification.
{"title":"Single Image Haze Removal Using Dark Channel Saturation Priori Model and Non-linear Diffusion Patch Method","authors":"Samiullah, Anuradha Paspathy","doi":"10.1145/3459104.3459186","DOIUrl":"https://doi.org/10.1145/3459104.3459186","url":null,"abstract":"In this article, a simple and effective restoration-based haze-removal approach is proposed. This approach is based on refining the course transmission map further by a novel non-linear diffusion patch method. The robustness of the proposed method is validated using quantitative analysis and is compared with other approaches with standard performance metrics. This technique can handle illumination, preserve edges better and ensures the original color of the image is retained. It can be used in many systems for example in object detection and tracking in order to recognize active traffic participants clearly on the road. Other applications include remote sensing for weather prediction, smart cars for smooth navigation and consumer electronics for fault identification.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130252803","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}
Mingshu Yao, Xiaofeng Xu, Zhenhe Ju, Bo Qv, Zhongyuan Zheng
The weak signal acquisition such as SF6 is affected by power frequency interference and sensor polarization voltage. If the integrated instrument amplifier is directly used as small signal amplifier, there are some problems, such as large data deviation. In this paper, a redundant monitoring system combining temperature compensation method and multi-stage gain is established, and the soft threshold denoising method is used to filter the interference signal, and the SF6 on-line data monitoring is realized by STM32F4 single chip microcomputer.
{"title":"Research and Application of SF6 Small Signal Detection System Based on Soft Threshold Denoising Method","authors":"Mingshu Yao, Xiaofeng Xu, Zhenhe Ju, Bo Qv, Zhongyuan Zheng","doi":"10.1145/3459104.3459131","DOIUrl":"https://doi.org/10.1145/3459104.3459131","url":null,"abstract":"The weak signal acquisition such as SF6 is affected by power frequency interference and sensor polarization voltage. If the integrated instrument amplifier is directly used as small signal amplifier, there are some problems, such as large data deviation. In this paper, a redundant monitoring system combining temperature compensation method and multi-stage gain is established, and the soft threshold denoising method is used to filter the interference signal, and the SF6 on-line data monitoring is realized by STM32F4 single chip microcomputer.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"198 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116479831","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}
Yan Zhang, Hongmei Zhang, Ji Zhang, Liangyuan Li, Ziyao Zheng
The security and stability of the power grid can ensure the stable balance of the power under the normal actual operation condition. It is an important requirement to guarantee the rapid development of national economy. With the increase of the complexity of the power grid structure, the higher requirements for the stability of the grid are put forward. This paper presents a power grid stability prediction model based on Bi-directional long short-term memory network (BiLSTM) with attention mechanism, which can learn the function of different stability features and the relationship between features. Firstly, the pre-processing power grid stability features are transformed into three-dimensional vector matrix input into the BiLSTM network. The multi-layer neural network layer is used to extract the deep-seated stability information.Then, the attention layer is used to allocate the corresponding weight to the extracted stable features. Finally, through the full connection layer, it can be transformed into a one-dimensional vector, which can be used to extract the stability features represents whether the grid is stable or not. Through the analysis of the results of the public 2018 uci data set, our experimental results are better than other methods, and the effect is more significant after the attention mechanism is added.
{"title":"Power Grid Stability Prediction Model Based on BiLSTM with Attention","authors":"Yan Zhang, Hongmei Zhang, Ji Zhang, Liangyuan Li, Ziyao Zheng","doi":"10.1145/3459104.3459160","DOIUrl":"https://doi.org/10.1145/3459104.3459160","url":null,"abstract":"The security and stability of the power grid can ensure the stable balance of the power under the normal actual operation condition. It is an important requirement to guarantee the rapid development of national economy. With the increase of the complexity of the power grid structure, the higher requirements for the stability of the grid are put forward. This paper presents a power grid stability prediction model based on Bi-directional long short-term memory network (BiLSTM) with attention mechanism, which can learn the function of different stability features and the relationship between features. Firstly, the pre-processing power grid stability features are transformed into three-dimensional vector matrix input into the BiLSTM network. The multi-layer neural network layer is used to extract the deep-seated stability information.Then, the attention layer is used to allocate the corresponding weight to the extracted stable features. Finally, through the full connection layer, it can be transformed into a one-dimensional vector, which can be used to extract the stability features represents whether the grid is stable or not. Through the analysis of the results of the public 2018 uci data set, our experimental results are better than other methods, and the effect is more significant after the attention mechanism is added.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132208696","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}
Di Li, Lei Sun, Wei Chen, B. Ai, Qi Wang, Zhenguo Du, Xiao Han
With rapid adoption of smartphones, context detection is becoming increasingly important to enable new and sophisticated context-aware mobile apps and provide better communication services. In this paper, we propose an Long Short Term Memory (LSTM) based indoor/outdoor/underground detection system for smartphone scene detection with low energy consumption. The proposed system is first compared with other deep learning methods including fully connected network (FC), standard LSTM network and Gated Recurrent Unit (GRU) based models. and then with traditional machine learning methods including K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR) and Random Forest (RF). Experimental results show that our proposed system is superiors in identifying indoor/outdoor/underground scene using only ultra-low power sensors. We collect real data at different periods and locations using multiple mobile devices. The required sensors are common in all types of smartphones, implying high compatibility and availability of the system.
{"title":"LSTM Based Scene Detection with Smartphones","authors":"Di Li, Lei Sun, Wei Chen, B. Ai, Qi Wang, Zhenguo Du, Xiao Han","doi":"10.1145/3459104.3459139","DOIUrl":"https://doi.org/10.1145/3459104.3459139","url":null,"abstract":"With rapid adoption of smartphones, context detection is becoming increasingly important to enable new and sophisticated context-aware mobile apps and provide better communication services. In this paper, we propose an Long Short Term Memory (LSTM) based indoor/outdoor/underground detection system for smartphone scene detection with low energy consumption. The proposed system is first compared with other deep learning methods including fully connected network (FC), standard LSTM network and Gated Recurrent Unit (GRU) based models. and then with traditional machine learning methods including K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR) and Random Forest (RF). Experimental results show that our proposed system is superiors in identifying indoor/outdoor/underground scene using only ultra-low power sensors. We collect real data at different periods and locations using multiple mobile devices. The required sensors are common in all types of smartphones, implying high compatibility and availability of the system.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133059997","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}
Recently, the amount of data is rapidly increasing with the continuous development of computation and communication capabilities. So, it has been actively studied for the effective data analysis schemes of the large amounts of data on MapReduce which supports efficient parallel data processing for large-scale data. Among various queries for analysing data, k nearest neighbour (kNN) join query, which aims to combine the k nearest neighbours of each point of dataset R with those from another dataset S, has been considered typical. However, existing kNN join schemes on MapReduce require high computation cost for constructing and managing index structures. To solve the problems, we propose a kNN-join query processing algorithm on MapReduce for analysing large-scale data. First, our algorithm can reduce the overhead for constructing the index structure by using the seed-based dynamic partitioning. Second, it can reduce the computational overhead to find candidate partitions by using the average distance between a pair of neighbouring seeds. We show that our algorithm outperforms the existing scheme in terms of the query processing time.
{"title":"kNN-join Query Processing Algorithm on Mapreduce for Large Amounts of Data","authors":"Hyunjo Lee, Jae-Woo Chang, Cheol-Joo Chae","doi":"10.1145/3459104.3459192","DOIUrl":"https://doi.org/10.1145/3459104.3459192","url":null,"abstract":"Recently, the amount of data is rapidly increasing with the continuous development of computation and communication capabilities. So, it has been actively studied for the effective data analysis schemes of the large amounts of data on MapReduce which supports efficient parallel data processing for large-scale data. Among various queries for analysing data, k nearest neighbour (kNN) join query, which aims to combine the k nearest neighbours of each point of dataset R with those from another dataset S, has been considered typical. However, existing kNN join schemes on MapReduce require high computation cost for constructing and managing index structures. To solve the problems, we propose a kNN-join query processing algorithm on MapReduce for analysing large-scale data. First, our algorithm can reduce the overhead for constructing the index structure by using the seed-based dynamic partitioning. Second, it can reduce the computational overhead to find candidate partitions by using the average distance between a pair of neighbouring seeds. We show that our algorithm outperforms the existing scheme in terms of the query processing time.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"92 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121979876","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}
Besides the goal of reducing driving tasks, modern longitudinal control systems also aim to improve fuel efficiency and driver comfort. Most of the vehicles use Adaptive Cruise Control (ACC) systems that track constant set speeds and set headways which makes the trajectory of the vehicle in headway mode highly dependent on the trajectory of a preceding vehicle. Hence, this might lead to increased consumptions in dense traffic situations or when the leader has a less careful driving style. In this work, a method based on Deep Reinforcement Learning (DRL) is presented that finds a control strategy by estimating an intelligent variable set speed based on the system state. Additional control objectives, such as minimizing consumption, are considered explicitly through the feedback in a reward function. A DRL framework is set up that enables the training of a neural set speed estimator for vehicle longitudinal control in a simulative environment. The Deep Deterministic Policy Gradient algorithm is used for the training of the agent. Training is carried out on a simple test track to teach the basic concepts of the control objective to the DRL agent. The learned behavior is then examined in a more complex, stochastic microscopic traffic simulation of the city center of Darmstadt and is compared to a conventional ACC algorithm. The analysis shows that the DRL controller is capable of finding fuel efficient trajectories which are less dependent on the preceding vehicle and is able to generalize to more complex traffic environments, but still shows some unexpected behavior in certain situations. The combination of DRL and conventional models to build up on the existing engineering knowledge is therefore expected to yield promising results in the future.
{"title":"Intelligent Set Speed Estimation for Vehicle Longitudinal Control with Deep Reinforcement Learning","authors":"Tobias Eichenlaub, S. Rinderknecht","doi":"10.1145/3459104.3459123","DOIUrl":"https://doi.org/10.1145/3459104.3459123","url":null,"abstract":"Besides the goal of reducing driving tasks, modern longitudinal control systems also aim to improve fuel efficiency and driver comfort. Most of the vehicles use Adaptive Cruise Control (ACC) systems that track constant set speeds and set headways which makes the trajectory of the vehicle in headway mode highly dependent on the trajectory of a preceding vehicle. Hence, this might lead to increased consumptions in dense traffic situations or when the leader has a less careful driving style. In this work, a method based on Deep Reinforcement Learning (DRL) is presented that finds a control strategy by estimating an intelligent variable set speed based on the system state. Additional control objectives, such as minimizing consumption, are considered explicitly through the feedback in a reward function. A DRL framework is set up that enables the training of a neural set speed estimator for vehicle longitudinal control in a simulative environment. The Deep Deterministic Policy Gradient algorithm is used for the training of the agent. Training is carried out on a simple test track to teach the basic concepts of the control objective to the DRL agent. The learned behavior is then examined in a more complex, stochastic microscopic traffic simulation of the city center of Darmstadt and is compared to a conventional ACC algorithm. The analysis shows that the DRL controller is capable of finding fuel efficient trajectories which are less dependent on the preceding vehicle and is able to generalize to more complex traffic environments, but still shows some unexpected behavior in certain situations. The combination of DRL and conventional models to build up on the existing engineering knowledge is therefore expected to yield promising results in the future.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123191420","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}
Masashi Sugimoto, Ryunosuke Uchida, S. Tsuzuki, Hitoshi Sori, H. Inoue, K. Kurashige, S. Urushihara
The Reinforcement Learning (RL) had been attracting attention for a long time that because it can be easily applied to real robots. On the other hand, in Q-Learning one of RL methods, since it contains the Q-table and grind environment is updated, especially, a large amount of Q-tables are required to express continuous “states,” such as smooth movements of the robot arm. Moreover, there was a disadvantage that calculation could not be performed real-time in case of amount of states and actions. The Deep Q-Network (DQN), on the other hand, uses convolutional neural network to estimate the Q-value itself, so that it can obtain an approximate function of the Q-value. From this characteristic of calculation that ignoring the amount of discrete states, this method has attracted attention, in recent. However, it seems to the following of multitasking and moving goal point that Q-Learning was not good at has been inherited by DQN. In this paper, the authors have improvements the multi-purpose execution of DQN by changing the exploration ratio as known as epsilon dynamically, has been tried. As the verification experiment, in the actual environment, a mobile crawler that mounting the NVIDIA Jetson NX and 2D LiDAR with the improvements DQN has been applied, to verify the object tracking ability, as a moving target position. As the result, the authors have confirmed that the improve its weak point.
{"title":"An Experimental Study for Tracking Ability of Deep Q-Network under the Multi-Objective Behaviour using a Mobile Robot with LiDAR","authors":"Masashi Sugimoto, Ryunosuke Uchida, S. Tsuzuki, Hitoshi Sori, H. Inoue, K. Kurashige, S. Urushihara","doi":"10.1145/3459104.3459120","DOIUrl":"https://doi.org/10.1145/3459104.3459120","url":null,"abstract":"The Reinforcement Learning (RL) had been attracting attention for a long time that because it can be easily applied to real robots. On the other hand, in Q-Learning one of RL methods, since it contains the Q-table and grind environment is updated, especially, a large amount of Q-tables are required to express continuous “states,” such as smooth movements of the robot arm. Moreover, there was a disadvantage that calculation could not be performed real-time in case of amount of states and actions. The Deep Q-Network (DQN), on the other hand, uses convolutional neural network to estimate the Q-value itself, so that it can obtain an approximate function of the Q-value. From this characteristic of calculation that ignoring the amount of discrete states, this method has attracted attention, in recent. However, it seems to the following of multitasking and moving goal point that Q-Learning was not good at has been inherited by DQN. In this paper, the authors have improvements the multi-purpose execution of DQN by changing the exploration ratio as known as epsilon dynamically, has been tried. As the verification experiment, in the actual environment, a mobile crawler that mounting the NVIDIA Jetson NX and 2D LiDAR with the improvements DQN has been applied, to verify the object tracking ability, as a moving target position. As the result, the authors have confirmed that the improve its weak point.","PeriodicalId":142284,"journal":{"name":"2021 International Symposium on Electrical, Electronics and Information Engineering","volume":"92 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115222960","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：