Ronie C. Bituin, Ronielle B. Antonio, James A. Esquivel
Recruitment industry is better and bigger than ever. There is no denying that technology plays a major role in helping recruiters evolve and adopt with the pace of recruitment on a global scale. With the increasing population, the demand for manpower has been relative to the growth and challenging needs of recruiters; be it online or traditional way of outsourcing. In this study, we propose a combination of angle or similarity and term frequency–inverse document frequency to easily classify prospective job applicants. The results show that the two models are relative to each other, value-wise and harmonic means. Their values are synchronized to a certain extent based on our query. This is helpful because recruiters may save a lot of time in classifying prospective applicants. It can also be concluded that harmonic similarity is viable in combining the two models. As a future work, it is possible to develop a full featured application to be deployed in a production setting.
{"title":"Harmonic Means between TF-IDF and Angle of Similarity to Identify Prospective Applicants in a Recruitment Setting","authors":"Ronie C. Bituin, Ronielle B. Antonio, James A. Esquivel","doi":"10.1145/3446132.3446414","DOIUrl":"https://doi.org/10.1145/3446132.3446414","url":null,"abstract":"Recruitment industry is better and bigger than ever. There is no denying that technology plays a major role in helping recruiters evolve and adopt with the pace of recruitment on a global scale. With the increasing population, the demand for manpower has been relative to the growth and challenging needs of recruiters; be it online or traditional way of outsourcing. In this study, we propose a combination of angle or similarity and term frequency–inverse document frequency to easily classify prospective job applicants. The results show that the two models are relative to each other, value-wise and harmonic means. Their values are synchronized to a certain extent based on our query. This is helpful because recruiters may save a lot of time in classifying prospective applicants. It can also be concluded that harmonic similarity is viable in combining the two models. As a future work, it is possible to develop a full featured application to be deployed in a production setting.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"85 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122255600","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}
Lane detection methods based on convolutional neural network have achieved excellent performance in recent years. Most of them treat lane detection as a semantic segmentation task which judges whether each pixel belongs to a lane. To make full use of the characteristics of lane shape, some researchers proposed to predict the whole lane. In this paper, we propose Lane Detection Combining Details and Integrity (LDCDI) which can explicitly leverage the advantages of both the segmentation-based methods and the regression-based methods. Specifically, we exploit an extra branch with regression-based methods as the auxiliary module after the main module. It not only maintains the advantages of the segmentation-based methods in lane detail segmentation, but also enables the model to have a sufficient understanding of the lane shape. Besides, the auxiliary module only takes part in the training, and there is no extra cost in the prediction. To further improve the quality of lane detection, we introduce a novel direction-sensitive block (DSB) based on ERFNet as the main module, which is more sensitive to the direction information of the image, so as to obtain better performance. Extensive experiments on the CULane dataset can demonstrate that our method outperforms other methods and achieves the state-of-the-art.
{"title":"Lane Detection Combining Details and Integrity: an Advanced Method for Lane Detection","authors":"Xingjian Dai, Jin Xie, J. Qian, Jian Yang","doi":"10.1145/3446132.3446145","DOIUrl":"https://doi.org/10.1145/3446132.3446145","url":null,"abstract":"Lane detection methods based on convolutional neural network have achieved excellent performance in recent years. Most of them treat lane detection as a semantic segmentation task which judges whether each pixel belongs to a lane. To make full use of the characteristics of lane shape, some researchers proposed to predict the whole lane. In this paper, we propose Lane Detection Combining Details and Integrity (LDCDI) which can explicitly leverage the advantages of both the segmentation-based methods and the regression-based methods. Specifically, we exploit an extra branch with regression-based methods as the auxiliary module after the main module. It not only maintains the advantages of the segmentation-based methods in lane detail segmentation, but also enables the model to have a sufficient understanding of the lane shape. Besides, the auxiliary module only takes part in the training, and there is no extra cost in the prediction. To further improve the quality of lane detection, we introduce a novel direction-sensitive block (DSB) based on ERFNet as the main module, which is more sensitive to the direction information of the image, so as to obtain better performance. Extensive experiments on the CULane dataset can demonstrate that our method outperforms other methods and achieves the state-of-the-art.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":" 33","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"113947128","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 neural network text-to-speech (TTS) systems can produce high-quality audio. However, modern TTS systems usually need a sizable of studio-quality pairs as input. In view of the insufficient research on Bahasa Indonesia, available data are usually worse in term of both quality and size. The End-to-End(E2E) TTS systems trained on those corpora are difficult to generate satisfactory speech, especially the prosodic features are not obvious. Therefore, we propose a method to enhance the prosodic features of synthesized speech based on GST-Tacotron2 model, and pre-trained language model with the BERT (Bidirectional Encoder Representation from Transformers) model. The BERT learned from large number of unlabeled text data contains rich linguistic information, which can help TTS systems produce the more obvious prosodic features. The subjective evaluation of our experimental results shows that the proposed method can indeed enhance the rhythm of synthesized speech.
深度神经网络文本到语音(TTS)系统可以产生高质量的音频。然而，现代TTS系统通常需要相当数量的工作室质量对作为输入。鉴于对印尼语的研究不足，现有的数据在质量和规模上通常都较差。在这些语料库上训练的端到端TTS系统很难产生令人满意的语音，尤其是韵律特征不明显。因此，我们提出了一种基于GST-Tacotron2模型和BERT (Bidirectional Encoder Representation from Transformers)模型的预训练语言模型来增强合成语音的韵律特征的方法。BERT从大量未标注的文本数据中学习到丰富的语言信息，可以帮助TTS系统产生更明显的韵律特征。对实验结果的主观评价表明，所提出的方法确实可以提高合成语音的节奏。
{"title":"Enhancing Prosodic Features by Adopting Pre-trained Language Model in Bahasa Indonesia Speech Synthesis","authors":"Lixuan Zhao, Jian Yang, Qinglai Qin","doi":"10.1145/3446132.3446196","DOIUrl":"https://doi.org/10.1145/3446132.3446196","url":null,"abstract":"Deep neural network text-to-speech (TTS) systems can produce high-quality audio. However, modern TTS systems usually need a sizable of studio-quality pairs as input. In view of the insufficient research on Bahasa Indonesia, available data are usually worse in term of both quality and size. The End-to-End(E2E) TTS systems trained on those corpora are difficult to generate satisfactory speech, especially the prosodic features are not obvious. Therefore, we propose a method to enhance the prosodic features of synthesized speech based on GST-Tacotron2 model, and pre-trained language model with the BERT (Bidirectional Encoder Representation from Transformers) model. The BERT learned from large number of unlabeled text data contains rich linguistic information, which can help TTS systems produce the more obvious prosodic features. The subjective evaluation of our experimental results shows that the proposed method can indeed enhance the rhythm of synthesized speech.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124946612","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}
Simultaneously recovering the 3D shape and its surface color from a single image has been a very challenging. In this paper, we substantially improve Soft Rasterizer that is a state-of-the art method for 3D color object reconstruction. The model adopts the structure of the encoder and decoder with a single image as input. Firstly, the features are extracted by the encoder, and then they are simultaneously sent to the shape generator and the color generator to obtain the shape estimate and the corresponding surface color, and finally the final colorful 3D model is rendered by the differentiable renderer. In order to ensure the details of the reconstructed 3D model, this paper introduces an attention mechanism into the encoder to further improve the reconstruction effect. For surface color reconstruction, we propose a combination loss. The experimental results show that compared with the 3D reconstruction network models 3D-R2N2 and OccNet, the intersection-over-union (IOU) increases by 10% and 3% in our model. Compared to the open source project SoftRas_O, the model increases by 3.8% on structural similarity (SSIM) and decreases by 1.2% on mean square error (MSE).
{"title":"Colorful 3d reconstruction from a single image based on deep learning","authors":"Yuzheng Zhu, Yaping Zhang, Qiaosheng Feng","doi":"10.1145/3446132.3446157","DOIUrl":"https://doi.org/10.1145/3446132.3446157","url":null,"abstract":"Simultaneously recovering the 3D shape and its surface color from a single image has been a very challenging. In this paper, we substantially improve Soft Rasterizer that is a state-of-the art method for 3D color object reconstruction. The model adopts the structure of the encoder and decoder with a single image as input. Firstly, the features are extracted by the encoder, and then they are simultaneously sent to the shape generator and the color generator to obtain the shape estimate and the corresponding surface color, and finally the final colorful 3D model is rendered by the differentiable renderer. In order to ensure the details of the reconstructed 3D model, this paper introduces an attention mechanism into the encoder to further improve the reconstruction effect. For surface color reconstruction, we propose a combination loss. The experimental results show that compared with the 3D reconstruction network models 3D-R2N2 and OccNet, the intersection-over-union (IOU) increases by 10% and 3% in our model. Compared to the open source project SoftRas_O, the model increases by 3.8% on structural similarity (SSIM) and decreases by 1.2% on mean square error (MSE).","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125002107","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 prediction of the barrage emotional change is very important for video playback effect and the analysis of user interest. Currently, some existing method including least squares and BP network for data fitting were used. However, these methods often have "bulging phenomenon", poor applicability to small samples, and low generalization performance. In order to solve these problems, in this paper, we propose a hybrid kernel PSO_LSSVM model based on least squares support vector machine. The fitting performance of the model is mainly determined by the selected kernel function and its parameters. Considering that the local Gaussian radial basis kernel function has strong learning ability but weak generalization ability, while the global polynomial kernel function has strong generalization ability but weak learning ability. We propose to combine the advantages of the two, build a least squares support vector machine model based on hybrid kernels, and cited the particle swarm optimization algorithm to optimize twice to obtain the optimal parameter value of the model. Hence the model can achieve high fitting accuracy, and can also ensure a higher prediction accuracy. So as to obtain the fitting curve of the user's barrage emotion change, we carried out fitting experiments on the emotional data samples obtained from the barrage comment text, and conducted comparison experiments with unimproved least squares support vector machine, BP neural network and other methods. Verifying the effectiveness and generalization of the model in fitting the barrage emotional change curve.
{"title":"Curve fitting of the user barrage emotional change based on the hybrid kernel PSO_LSSVM model","authors":"Fulian Yin, Xiaoli Feng, Fangyuan Ju, Yanyan Wang","doi":"10.1145/3446132.3446138","DOIUrl":"https://doi.org/10.1145/3446132.3446138","url":null,"abstract":"The prediction of the barrage emotional change is very important for video playback effect and the analysis of user interest. Currently, some existing method including least squares and BP network for data fitting were used. However, these methods often have \"bulging phenomenon\", poor applicability to small samples, and low generalization performance. In order to solve these problems, in this paper, we propose a hybrid kernel PSO_LSSVM model based on least squares support vector machine. The fitting performance of the model is mainly determined by the selected kernel function and its parameters. Considering that the local Gaussian radial basis kernel function has strong learning ability but weak generalization ability, while the global polynomial kernel function has strong generalization ability but weak learning ability. We propose to combine the advantages of the two, build a least squares support vector machine model based on hybrid kernels, and cited the particle swarm optimization algorithm to optimize twice to obtain the optimal parameter value of the model. Hence the model can achieve high fitting accuracy, and can also ensure a higher prediction accuracy. So as to obtain the fitting curve of the user's barrage emotion change, we carried out fitting experiments on the emotional data samples obtained from the barrage comment text, and conducted comparison experiments with unimproved least squares support vector machine, BP neural network and other methods. Verifying the effectiveness and generalization of the model in fitting the barrage emotional change curve.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127692314","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}
Previous works use a similar process to prune channels: train, prune, fine-tune. In this paper, we treat channel pruning as a method of network architecture search. Specifically, we limit the search space by adding some conditions on it, and after searching, we only reserve the architecture of the network and train it from scratch. We train the model with augmented sparsity to get a higher ratio of pruning. During pruning, we add a protect threshold to prevent the pruned model from being disconnection. Our process of channel pruning is as follows: train with sparsity, prune, train from scratch. we verified the effectiveness of our method on several models, including VGGNet, ResNet and DenseNet on various datasets. Otherwise, we test our method on different architectures of ResNet and analyze the results on both models.
{"title":"Network Slimming with Augmented Sparse Training and Optimized Pruning","authors":"Ziliang Guo, Xueming Li","doi":"10.1145/3446132.3446159","DOIUrl":"https://doi.org/10.1145/3446132.3446159","url":null,"abstract":"Previous works use a similar process to prune channels: train, prune, fine-tune. In this paper, we treat channel pruning as a method of network architecture search. Specifically, we limit the search space by adding some conditions on it, and after searching, we only reserve the architecture of the network and train it from scratch. We train the model with augmented sparsity to get a higher ratio of pruning. During pruning, we add a protect threshold to prevent the pruned model from being disconnection. Our process of channel pruning is as follows: train with sparsity, prune, train from scratch. we verified the effectiveness of our method on several models, including VGGNet, ResNet and DenseNet on various datasets. Otherwise, we test our method on different architectures of ResNet and analyze the results on both models.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115606928","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}
Chuang Wang, Song Li, Wenfeng Wei, Shijie Li, Jiayi Liu
Under the background of great changes in military science and technology theory, in order to solve the problem of massive high-dimensional situation knowledge processing in the process of battlefield situation assessment.The current dimensionality reduction methods often ignore the influence of algorithm complexity and model representation ability on dimensionality reduction when solving the massive dimensionality reduction problem of high-dimensional situation knowledge. In order to balance this problem, this paper proposes a situation knowledge dimension reduction method based on Sparse Autoencoder, which has a good performance in achieving dimension reduction of high-dimensional situation information and obtaining its abstract feature representation.
{"title":"A dimension reduction method of situation knowledge based on Sparse Autoencoder","authors":"Chuang Wang, Song Li, Wenfeng Wei, Shijie Li, Jiayi Liu","doi":"10.1145/3446132.3446151","DOIUrl":"https://doi.org/10.1145/3446132.3446151","url":null,"abstract":"Under the background of great changes in military science and technology theory, in order to solve the problem of massive high-dimensional situation knowledge processing in the process of battlefield situation assessment.The current dimensionality reduction methods often ignore the influence of algorithm complexity and model representation ability on dimensionality reduction when solving the massive dimensionality reduction problem of high-dimensional situation knowledge. In order to balance this problem, this paper proposes a situation knowledge dimension reduction method based on Sparse Autoencoder, which has a good performance in achieving dimension reduction of high-dimensional situation information and obtaining its abstract feature representation.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"69 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117155007","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 Multi-constrained Vehicle Routing Problem (MCVRP) is an extension of the basic vehicle routing problem (VRP). There may be more than one constraint, and the distribution cost is not only related to the routing decision, but also to the transportation volume of vehicles. This paper describes and analyzes the MCVRP, then builds an integer programming model of the multi-constraint vehicle routing problem. For this problem model, the best algorithm for solving multi-constrained vehicle routing problems is based on genetic algorithm (GA). To overcome the shortcomings of traditional GA, an improved adaptive GA for MCVRP optimization is proposed. Finally, a simulation experiment was performed on the actual data set to verify the effectiveness of the model and algorithm.
{"title":"Multi-constrained Vehicle Routing Problem Solution based on Adaptive Genetic Algorithm","authors":"Wen-Qing Fan","doi":"10.1145/3446132.3446401","DOIUrl":"https://doi.org/10.1145/3446132.3446401","url":null,"abstract":"The Multi-constrained Vehicle Routing Problem (MCVRP) is an extension of the basic vehicle routing problem (VRP). There may be more than one constraint, and the distribution cost is not only related to the routing decision, but also to the transportation volume of vehicles. This paper describes and analyzes the MCVRP, then builds an integer programming model of the multi-constraint vehicle routing problem. For this problem model, the best algorithm for solving multi-constrained vehicle routing problems is based on genetic algorithm (GA). To overcome the shortcomings of traditional GA, an improved adaptive GA for MCVRP optimization is proposed. Finally, a simulation experiment was performed on the actual data set to verify the effectiveness of the model and algorithm.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114984310","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}
With the development of military intelligence, uav has become one of the important means of intelligence acquisition in modern warfare. As the main way of UAV reconnaissance, image reconnaissance is playing an increasingly important role in the mission. At present, in the process of uav image reconnaissance, there are still some problems, such as unclear fog image and inability to reflect the overall situation. Aiming at these two kinds of problems, this paper reviews several mainstream algorithms of image processing. Then the algorithm is compared and analyzed based on the characteristics of uav reconnaissance image. Finally, the application prospect of image processing algorithm in improving uav reconnaissance efficiency is prospected.
{"title":"Research on the application of image processing in improving the reconnaissance efficiency of UAV","authors":"Xiang-hui Shen, Xiaoyang Liu, Pengfei Jiao","doi":"10.1145/3446132.3446198","DOIUrl":"https://doi.org/10.1145/3446132.3446198","url":null,"abstract":"With the development of military intelligence, uav has become one of the important means of intelligence acquisition in modern warfare. As the main way of UAV reconnaissance, image reconnaissance is playing an increasingly important role in the mission. At present, in the process of uav image reconnaissance, there are still some problems, such as unclear fog image and inability to reflect the overall situation. Aiming at these two kinds of problems, this paper reviews several mainstream algorithms of image processing. Then the algorithm is compared and analyzed based on the characteristics of uav reconnaissance image. Finally, the application prospect of image processing algorithm in improving uav reconnaissance efficiency is prospected.","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"75 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122018885","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}
Pre-trained Transformers(GPT) are showed great performance in natural language generation task. This model was trained in a self-supervised manner on a large amount of text data crawled from the WEB. Such a dataset has not the highest quality, many sentences are prone to errors such as typos or grammar mistakes. As a result, text generated by GPTs consists of a lot of grammar incorrect sentences. While Transformers is also showed great performance in translation tasks, we propose the conception when a model can handle a generation and a translation task at the same time. But we propose a specific type of translation, in our method Transformer is training to translate a sentence with grammar errors to the same sentences without errors. In the full case, an incorrectly generated sentence can be corrected by the extended version of the same model, we call this type of model Transgenerator. We applied several experiments to estimate a generative power of Transgenerator based on GPT-2 architecture and the proposed method outperformed original GPT-2 model on the range of tasks
{"title":"Transgenerators","authors":"Arip Asadulaev, Gideon Stein, A. Filchenkov","doi":"10.1145/3446132.3446417","DOIUrl":"https://doi.org/10.1145/3446132.3446417","url":null,"abstract":"Pre-trained Transformers(GPT) are showed great performance in natural language generation task. This model was trained in a self-supervised manner on a large amount of text data crawled from the WEB. Such a dataset has not the highest quality, many sentences are prone to errors such as typos or grammar mistakes. As a result, text generated by GPTs consists of a lot of grammar incorrect sentences. While Transformers is also showed great performance in translation tasks, we propose the conception when a model can handle a generation and a translation task at the same time. But we propose a specific type of translation, in our method Transformer is training to translate a sentence with grammar errors to the same sentences without errors. In the full case, an incorrectly generated sentence can be corrected by the extended version of the same model, we call this type of model Transgenerator. We applied several experiments to estimate a generative power of Transgenerator based on GPT-2 architecture and the proposed method outperformed original GPT-2 model on the range of tasks","PeriodicalId":125388,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial Intelligence","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126669592","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：