A new type of number systems (integer coding systems) is used. In the system a set of digits, each of which corresponds to a certain prime number, is used instead of digits corresponding to the powers of a certain integer (for example, ten), All the prime numbers corresponding to different digits are different. Such an encoding of integers corresponds to a discrete signal model, in which the function corresponding to this model takes values in some algebraic ring. The advantage of such an encoding is the independent multiplication of numbers corresponding to different digits, which provides a significant simplification of calculations, including calculation of convolutions of signals presented in a discrete form. It is shown that in this case the convolution operation can be reduced to a situation where the convolution is calculated in Galois fields. In this case, the convolution operations carried out for the signals presented in proposed number system are carried out independently for each digit. A specific algorithm that implements this approach is proposed and its advantages for describing convolutional neural networks are proved. A specific example demonstrating these advantages is considered.
{"title":"Prospects for the use of algebraic rings to describe the operation of convolutional neural networks","authors":"I. Suleimenov, A. Bakirov, Y. Vitulyova","doi":"10.1145/3571560.3571561","DOIUrl":"https://doi.org/10.1145/3571560.3571561","url":null,"abstract":"A new type of number systems (integer coding systems) is used. In the system a set of digits, each of which corresponds to a certain prime number, is used instead of digits corresponding to the powers of a certain integer (for example, ten), All the prime numbers corresponding to different digits are different. Such an encoding of integers corresponds to a discrete signal model, in which the function corresponding to this model takes values in some algebraic ring. The advantage of such an encoding is the independent multiplication of numbers corresponding to different digits, which provides a significant simplification of calculations, including calculation of convolutions of signals presented in a discrete form. It is shown that in this case the convolution operation can be reduced to a situation where the convolution is calculated in Galois fields. In this case, the convolution operations carried out for the signals presented in proposed number system are carried out independently for each digit. A specific algorithm that implements this approach is proposed and its advantages for describing convolutional neural networks are proved. A specific example demonstrating these advantages is considered.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123596644","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}
Deep learning has promoted the development of many areas in computer vision and robotics. However, most of the researches focus on an individual task. In this paper, we design a multi-task robot system based on ROS platform and YOLO network to complete the object detection, positioning, and grasping tasks. In terms of hardware, a heterogeneous computing platform is established to achieve high computing power while reducing energy consumption. In terms of software, an algorithm framework is designed for the multi-task robot system according to the characters the heterogeneous computing platform. Experimental results on real data show that the proposed robot system achieves promising object detection, positioning and grasping performance.
{"title":"A Visual Learning based Robotic Grasping System","authors":"Weijun Guan, Yulan Guo","doi":"10.1145/3571560.3571564","DOIUrl":"https://doi.org/10.1145/3571560.3571564","url":null,"abstract":"Deep learning has promoted the development of many areas in computer vision and robotics. However, most of the researches focus on an individual task. In this paper, we design a multi-task robot system based on ROS platform and YOLO network to complete the object detection, positioning, and grasping tasks. In terms of hardware, a heterogeneous computing platform is established to achieve high computing power while reducing energy consumption. In terms of software, an algorithm framework is designed for the multi-task robot system according to the characters the heterogeneous computing platform. Experimental results on real data show that the proposed robot system achieves promising object detection, positioning and grasping performance.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124730855","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}
Hashing has received more and more attention due to the characteristics of small storage and fast retrieval, especially in the field of biometric computing. However, there are only two types of binary relations, logical 'true' and logical 'false', which can't be distinguished well at the demarcation of these two logical relations in the Hamming space, resulting in ambiguity in the Hamming space neighborhood. Therefore, deep hash network is used to extract palmprint feature and optimize the ambiguity in Hamming space by using mutual information (MI) to obtain a trivialized palmprint hash code. The tri-valued Hamming distance using Kleene logic for matching reduces storage and improves matching speed compared to traditional local feature-based coding methods, and outperforms the binary palmprint deep hash coding. All the experiments are conducted on several contact/contactless palmprint and palm vein libraries, and extensive comparisons are made with several state-of-the-art methods, and the results demonstrate the effectiveness of the proposed scheme.
{"title":"Deep Ternary Hashing Code for Palmprint Retrieval and Recognition","authors":"Qizhou Lin, L. Leng, M. Khan","doi":"10.1145/3571560.3571573","DOIUrl":"https://doi.org/10.1145/3571560.3571573","url":null,"abstract":"Hashing has received more and more attention due to the characteristics of small storage and fast retrieval, especially in the field of biometric computing. However, there are only two types of binary relations, logical 'true' and logical 'false', which can't be distinguished well at the demarcation of these two logical relations in the Hamming space, resulting in ambiguity in the Hamming space neighborhood. Therefore, deep hash network is used to extract palmprint feature and optimize the ambiguity in Hamming space by using mutual information (MI) to obtain a trivialized palmprint hash code. The tri-valued Hamming distance using Kleene logic for matching reduces storage and improves matching speed compared to traditional local feature-based coding methods, and outperforms the binary palmprint deep hash coding. All the experiments are conducted on several contact/contactless palmprint and palm vein libraries, and extensive comparisons are made with several state-of-the-art methods, and the results demonstrate the effectiveness of the proposed scheme.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131101901","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}
Question answering is one of the well-studied tasks in the natural language processing(NLP) community, which aims to secure an answer span from a given document and query. Previous attempts decomposed this task into two subtask, i.e., understanding the semantic information of the given document and query, then finding a reasonable textual span within the document as the corresponding answer. However, one of the major drawbacks of the previous works is lack of extracting sufficient semantics that is buried within the input. To alleviate the issue above, in this paper, we propose a global-local attention flow model to take advantage of the semantic features from different aspects and reduce the redundancy of model encoder. Experimental results on the SQUAD dataset shows that our model outperforms the baseline models, which proves the effectiveness of the proposed method.
{"title":"GLAF: Global-and-Local Attention Flow Model for Question Answering","authors":"Shao-Hua Sun","doi":"10.1145/3571560.3571570","DOIUrl":"https://doi.org/10.1145/3571560.3571570","url":null,"abstract":"Question answering is one of the well-studied tasks in the natural language processing(NLP) community, which aims to secure an answer span from a given document and query. Previous attempts decomposed this task into two subtask, i.e., understanding the semantic information of the given document and query, then finding a reasonable textual span within the document as the corresponding answer. However, one of the major drawbacks of the previous works is lack of extracting sufficient semantics that is buried within the input. To alleviate the issue above, in this paper, we propose a global-local attention flow model to take advantage of the semantic features from different aspects and reduce the redundancy of model encoder. Experimental results on the SQUAD dataset shows that our model outperforms the baseline models, which proves the effectiveness of the proposed method.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132294024","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}
Maintaining the identity while synthesizing the frontal view image is the most critical step in developing a "recognition via generation" framework. To this end, this paper investigates, tests and compares the performance of two deep learning architectures: Light-CNN and FaceNet. The Light-CNN is used to learn a robust feature for face verification tasks that produces a high-level facial identity accuracy over many traditional deep learning models. FaceNet, on the other hand, is a model to maps face images into a compact Euclidean space where distances directly represent a measure of face similarity. In our comparison, we use the TP-GAN model to perform several pre-processing stages. The face features are then extracted from the synthesized face images using Light-CNN and FaceNet as 256- and 128-dimensional representations, respectively. We evaluate the accuracy performances of Light-CNN and FaceNet architectures on Multi-PIE and FEI datasets.
{"title":"Comparative analysis of the Light-CNN and FaceNet methods for identifying and maintaining human faces","authors":"Huang Yea-Shuan, Mahmood Alhlffee","doi":"10.1145/3571560.3571575","DOIUrl":"https://doi.org/10.1145/3571560.3571575","url":null,"abstract":"Maintaining the identity while synthesizing the frontal view image is the most critical step in developing a \"recognition via generation\" framework. To this end, this paper investigates, tests and compares the performance of two deep learning architectures: Light-CNN and FaceNet. The Light-CNN is used to learn a robust feature for face verification tasks that produces a high-level facial identity accuracy over many traditional deep learning models. FaceNet, on the other hand, is a model to maps face images into a compact Euclidean space where distances directly represent a measure of face similarity. In our comparison, we use the TP-GAN model to perform several pre-processing stages. The face features are then extracted from the synthesized face images using Light-CNN and FaceNet as 256- and 128-dimensional representations, respectively. We evaluate the accuracy performances of Light-CNN and FaceNet architectures on Multi-PIE and FEI datasets.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124995188","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}
We introduce an online psychological counseling chatbot that uses the Task-Oriented Dialogue (TOD) system and the open-domain dialogue (ODD) system to communicate and recommend content as an emotion recognition result. If you use the TOD system, which is a dialogue system for a specific purpose, and the ODD system, which is a system that conducts dialogues without a purpose, you can conduct conversations more naturally. In this paper, the TOD system is used in inducing emotional conversation and providing content according to emotion, and the ODD system is used in the process of emotional conversation. This helps people conduct more natural conversations, and finally helps recommend content through emotional analysis.
{"title":"Online Psychological Counseling Chatbot for Seniors","authors":"Byeong-Ryong Kim, Eunji Kim, J. Rhee","doi":"10.1145/3571560.3571583","DOIUrl":"https://doi.org/10.1145/3571560.3571583","url":null,"abstract":"We introduce an online psychological counseling chatbot that uses the Task-Oriented Dialogue (TOD) system and the open-domain dialogue (ODD) system to communicate and recommend content as an emotion recognition result. If you use the TOD system, which is a dialogue system for a specific purpose, and the ODD system, which is a system that conducts dialogues without a purpose, you can conduct conversations more naturally. In this paper, the TOD system is used in inducing emotional conversation and providing content according to emotion, and the ODD system is used in the process of emotional conversation. This helps people conduct more natural conversations, and finally helps recommend content through emotional analysis.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129486233","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}
The type of infrastructure and selection of its materials is one of the principal factors that must be considered. Due to its usual large quantifications on projects, it directly affects the environment and communities where it belonged. And collectively, the future of our world. As a strong, versatile, durable, sustainable, and environmentally beneficial material, bamboo and its derivatives are frequently utilized since the early times; the Philippines is fortunate to have an abundance of it across the country. The mechanical properties of one of the local R&D-prioritized and market-prominent bamboo specie, the Bambusa blumeana, are remarkable and well-known to be an excellent material for many structural elements. But to fully utilize it, reinforcements may be required, just like with any other ligneous and organic materials. Extensions in its compression strength along the grain may be accomplished from its 50.83 MPa average strength by confinement-reinforcing it with the promising, adaptable, and strong Carbon-fiber-reinforced polymer (CFRP). The Artificial Neural Network (ANN) model involving CFRP's confinement reinforcement thickness, edges that constitutes the compression area, moisture content, temperature, and density of Laminated Veneer Bamboo (LVB) was established using the Levenberg-Marquardt (LM) algorithm as the training algorithm (TA) and hyperbolic tangent sigmoid as the transfer function (TF). The relationship of the variables to the composite section's ultimate compressive strength, was indirectly proportional, except for density, and was further checked the influence using Garson's algorithm (GA). In addition, the results were verified using additional physical experimentation and Finite Element (FE) simulations, while the ANN model was compared to other prediction modelling techniques, by which the FE simulation proved to be an effective complement to the physical testing and the ANN prediction model performed the best. The results also reconfirmed other literature on engineered bamboo studies; and the failure of the CFRP-LVB composite section was found to be a combination of isolated partial failures of the LVB core as the cross-sections become larger, while full crushing was observed on smaller cross-sections.
{"title":"Carbon-Fiber-reinforced Polymer as Confinement Reinforcement to Maximize Compressive Strength of Engineered Bamboo: An Artificial Neural Network Model","authors":"W. E. Silva, D. Silva","doi":"10.1145/3571560.3571569","DOIUrl":"https://doi.org/10.1145/3571560.3571569","url":null,"abstract":"The type of infrastructure and selection of its materials is one of the principal factors that must be considered. Due to its usual large quantifications on projects, it directly affects the environment and communities where it belonged. And collectively, the future of our world. As a strong, versatile, durable, sustainable, and environmentally beneficial material, bamboo and its derivatives are frequently utilized since the early times; the Philippines is fortunate to have an abundance of it across the country. The mechanical properties of one of the local R&D-prioritized and market-prominent bamboo specie, the Bambusa blumeana, are remarkable and well-known to be an excellent material for many structural elements. But to fully utilize it, reinforcements may be required, just like with any other ligneous and organic materials. Extensions in its compression strength along the grain may be accomplished from its 50.83 MPa average strength by confinement-reinforcing it with the promising, adaptable, and strong Carbon-fiber-reinforced polymer (CFRP). The Artificial Neural Network (ANN) model involving CFRP's confinement reinforcement thickness, edges that constitutes the compression area, moisture content, temperature, and density of Laminated Veneer Bamboo (LVB) was established using the Levenberg-Marquardt (LM) algorithm as the training algorithm (TA) and hyperbolic tangent sigmoid as the transfer function (TF). The relationship of the variables to the composite section's ultimate compressive strength, was indirectly proportional, except for density, and was further checked the influence using Garson's algorithm (GA). In addition, the results were verified using additional physical experimentation and Finite Element (FE) simulations, while the ANN model was compared to other prediction modelling techniques, by which the FE simulation proved to be an effective complement to the physical testing and the ANN prediction model performed the best. The results also reconfirmed other literature on engineered bamboo studies; and the failure of the CFRP-LVB composite section was found to be a combination of isolated partial failures of the LVB core as the cross-sections become larger, while full crushing was observed on smaller cross-sections.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129159115","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}
Mohamed Imam, Karim Baïna, Youness Tabii, I. Benzakour, Youssef Adlaoui, El Mostafa Ressami, E. Abdelwahed
Underground prospecting operations are often characterized by critical safety issues mainly due to poor visibility and blind spots around large vehicles and equipment. This can result in vehicle-to-vehicle collisions, as well as vehicle-to-pedestrian or structural-element collisions, resulting in accidents. In this article, we discuss an anti-collision system for pedestrian identification in deep mines under the premise that we are looking to prevent collisions with moving machinery. This study presents the findings from testing an image processing module and sensory system based on deep learnig in the context of "smart connected mine" project.
{"title":"Anti-Collision System for Accident Prevention in Underground Mines using Computer Vision","authors":"Mohamed Imam, Karim Baïna, Youness Tabii, I. Benzakour, Youssef Adlaoui, El Mostafa Ressami, E. Abdelwahed","doi":"10.1145/3571560.3571574","DOIUrl":"https://doi.org/10.1145/3571560.3571574","url":null,"abstract":"Underground prospecting operations are often characterized by critical safety issues mainly due to poor visibility and blind spots around large vehicles and equipment. This can result in vehicle-to-vehicle collisions, as well as vehicle-to-pedestrian or structural-element collisions, resulting in accidents. In this article, we discuss an anti-collision system for pedestrian identification in deep mines under the premise that we are looking to prevent collisions with moving machinery. This study presents the findings from testing an image processing module and sensory system based on deep learnig in the context of \"smart connected mine\" project.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133018255","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}
Mohamad Faiz Dzulkalnine, M. S. Mohamad, Yee Wen Choon, Muhammad Akmal bin Remli, Hany Alashwal
Metabolic engineering for biomass production using microorganisms’ cell has received considerable attention in recent years. This is due to the biomass products being extensively used in the field of food additives, supplements, pharmaceuticals, and polymer materials. In this paper, ethanol production in Escherichia coli (E. coli) is the desired product. Sugarcane and corn are often used to produce ethanol. However, one of the problems to produce adequate amounts of ethanol is that large areas are needed to plant sugarcane and corn. Furthermore, the amount of time for the process of dry milling and wet milling is high, which are 40 to 50 hours and 24 to 48 hours, respectively. The wet laboratory is also having limitation on the production of ethanol in microorganisms because the amount of the ethanol produced is not satisfying. Hence, a lot of metabolic engineering techniques is introduced to enhance the production of ethanol in E. coli, such as gene knockout strategy, but the production is yet to meet the demand. Therefore, this paper proposes a hybrid algorithm of Particle Swarm Optimization with the Artificial Bee Colony algorithm (PSOABC) to identify the optimal set of gene knockout strategy to improve the ethanol production in E. coli. A list of genes to knockout, production of the desired product, and growth rate are presented in this paper. PSOABC has shown better performance in terms of production, growth rate and accuracy.
{"title":"Optimizing Ethanol Production in Escherichia Coli Using a Hybrid of Particle Swarm Optimization and Artificial Bee Colony","authors":"Mohamad Faiz Dzulkalnine, M. S. Mohamad, Yee Wen Choon, Muhammad Akmal bin Remli, Hany Alashwal","doi":"10.1145/3571560.3571581","DOIUrl":"https://doi.org/10.1145/3571560.3571581","url":null,"abstract":"Metabolic engineering for biomass production using microorganisms’ cell has received considerable attention in recent years. This is due to the biomass products being extensively used in the field of food additives, supplements, pharmaceuticals, and polymer materials. In this paper, ethanol production in Escherichia coli (E. coli) is the desired product. Sugarcane and corn are often used to produce ethanol. However, one of the problems to produce adequate amounts of ethanol is that large areas are needed to plant sugarcane and corn. Furthermore, the amount of time for the process of dry milling and wet milling is high, which are 40 to 50 hours and 24 to 48 hours, respectively. The wet laboratory is also having limitation on the production of ethanol in microorganisms because the amount of the ethanol produced is not satisfying. Hence, a lot of metabolic engineering techniques is introduced to enhance the production of ethanol in E. coli, such as gene knockout strategy, but the production is yet to meet the demand. Therefore, this paper proposes a hybrid algorithm of Particle Swarm Optimization with the Artificial Bee Colony algorithm (PSOABC) to identify the optimal set of gene knockout strategy to improve the ethanol production in E. coli. A list of genes to knockout, production of the desired product, and growth rate are presented in this paper. PSOABC has shown better performance in terms of production, growth rate and accuracy.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123732003","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}
Junhyeong Park, Geonsik Youn, Bohan Yoon, Byeonghun Kim, J. Rhee
Emotion Recognition (ER) is an essential research area of natural language processing that can be applied to various fields. Texts in the fields of health care, marketing, and psychological counseling take various forms, and it is very important from a business point of view to find the emotions inherent in these texts. Recently, ER using text embeddings generated through a pre-trained language model with a large corpus was performed. However, since the embeddings are generalized to various domains, there is a limitation to directly using them for ER. In this study, to overcome the limitation, we propose a method that modifies generalized embeddings to emotional embeddings by performing proxy-based metric learning. In the proposed method, we fine-tuned the pre-trained language model by using proxy-anchor loss so that embeddings represent emotion appropriately. Previous studies only added linear classifiers. But, it is possible to capture emotional relationships between data by using proxy-based metric learning. In this study, we conducted ER experiments with benchmark datasets. The experimental result shows that the proposed method achieves better performance than the baseline and creates emotion-specific embeddings.
{"title":"Proxy-based Metric Learning for Emotion Recognition","authors":"Junhyeong Park, Geonsik Youn, Bohan Yoon, Byeonghun Kim, J. Rhee","doi":"10.1145/3571560.3571578","DOIUrl":"https://doi.org/10.1145/3571560.3571578","url":null,"abstract":"Emotion Recognition (ER) is an essential research area of natural language processing that can be applied to various fields. Texts in the fields of health care, marketing, and psychological counseling take various forms, and it is very important from a business point of view to find the emotions inherent in these texts. Recently, ER using text embeddings generated through a pre-trained language model with a large corpus was performed. However, since the embeddings are generalized to various domains, there is a limitation to directly using them for ER. In this study, to overcome the limitation, we propose a method that modifies generalized embeddings to emotional embeddings by performing proxy-based metric learning. In the proposed method, we fine-tuned the pre-trained language model by using proxy-anchor loss so that embeddings represent emotion appropriately. Previous studies only added linear classifiers. But, it is possible to capture emotional relationships between data by using proxy-based metric learning. In this study, we conducted ER experiments with benchmark datasets. The experimental result shows that the proposed method achieves better performance than the baseline and creates emotion-specific embeddings.","PeriodicalId":143909,"journal":{"name":"Proceedings of the 6th International Conference on Advances in Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131202512","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：