检索结果-内蒙古大学图书馆

Mining Spatiotemporal Information for Harmful Algal Bloom Forecasting and Mechanism Interpreting

ACS ES&T WATER 2024年第6期4卷 2608-2618页

作者： Jia, Qimeng Xu, Changqing Jia, Haifeng Velazquez, Carlos Leng, Linyuan Yin, Dingkun Tsinghua Univ Sch Environm Beijing 100084 Peoples R China Beijing Inst Technol Sch Econ & Management Beijing 100081 Peoples R China Minist Ind & Informat Technol Digital Econ & Policy Intelligentizat Key Lab Beijing 100081 Peoples R China Suzhou Univ Sci & Technol Jiangsu Collaborat Innovat Ctr Technol & Mat Water Suzhou 215009 Peoples R China

A multistep spatiotemporal forecasting (MSTF) network is developed through incorporating the graph convolutional network (GCN) and the long short-term memory (LSTM) network within a sequence-to-sequence (seq2seq) framework. The MSTF method can not only extract spatial and temporal information from the input data but also make multistep-ahead and continuous predictions. An MSTF-based harmful algal bloom (HAB) forecasting model is then formulated to predict the chlorophyll-a (Chl-a) concentration of the Dianchi Lake (China). The integrated gradients (IG) method is employed to interpret the trained MSTF model and quantify the attribution of each input dimension to the Chl-a prediction. Results indicate that (i) the coefficient of determination (R-2) of the MSTF model in 24-h-ahead Chl-a prediction reaches 0.926, 28.4% higher than that of the traditional LSTM model;(ii) the ammonia nitrogen (12.3%), the total phosphorus (10.2%), the total nitrogen (9.9%), and the temperature (8.6%) are significant variables for Chl-a prediction;(iii) the spatial information from neighbor lake and river stations plays an important role in the HAB forecasting, with an average contribution of 35.0%;(iv) the proposed MSTF model is also skillful in the 72-h-ahead Chl-a prediction. Results presented highlight the importance of considering both spatial and temporal dependency of monitoring data in HAB forecasting and mechanism interpreting.

关键词： graph convolutional network long short-term memorynetwork harmful algal blooms sequence-to-sequence Dianchi Lake chlorophyll-a integrated gradients

来源：评论

学校读者我要写书评

暂无评论

JOINT END-TO-END SPOKEN LANGUAGE UNDERSTANDING AND AUTOMATIC SPEECH RECOGNITION TRAINING BASED ON UNIFIED SPEECH-TO-TEXT PRE-TRAINING 49

JOINT END-TO-END SPOKEN LANGUAGE UNDERSTANDING AND AUTOMATIC...

引用

49th IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Kim, Eesung Tang, Yun Ki, Taeyeon Neelagiri, Divya Apsingek, Vijendra Raj Samsung Res Amer Mountain View CA 94043 USA

ISBN: (纸本)9798350344868;9798350344851

Modern spoken language understanding (SLU) approaches optimize the system in an end-to-end (E2E) manner. This approach offers two key advantages. Firstly, it helps mitigate error propagation from upstream systems. Secondly, combining various information types and optimizing them towards the same objective is straightforward. In this study, we attempt to build an SLU system by integrating information from two modalities, i.e., speech and text, and concurrently optimizing the associated tasks. We leverage a pre-trained model built with speech and text data and fine-tune it for the E2E SLU tasks. The SLU model is jointly optimized with automatic speech recognition (ASR) and SLU tasks under single-mode and dual-mode schemes. In the single-mode model, ASR and SLU results are predicted sequentially, whereas the dual-mode model predicts either ASR or SLU outputs based on the task tag. Our proposed method demonstrates its superiority through benchmarking against FSC, SLURP, and in-house datasets, exhibiting improved intent accuracy, SLU-F1, and Word Error Rate (WER).

关键词： spoken language understanding automatic speech recognition end-to-end model sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

EM-TTS: Efficiently Trained Low-Resource Mongolian Lightweight Text-to-Speech 27

EM-TTS: Efficiently Trained Low-Resource Mongolian Lightweig...

引用

27th International Conference on Computer Supported Cooperative Work in Design (CSCWD)

作者： Liang, Ziqi Shi, Haoxiang Wang, Jiawei Lu, Keda Univ Sci & Technol China Hefei Peoples R China Hefei Univ Technol Hefei Peoples R China

ISBN: (纸本)9798350349184;9798350349191

Recently, deep learning-based Text-to-Speech (TTS) systems have achieved high-quality speech synthesis results. Recurrent neural networks have become a standard modeling technique for sequential data in TTS systems and are widely used. However, training a TTS model which includes RNN components requires powerful GPU performance and takes a long time. In contrast, CNN-based sequence synthesis techniques can significantly reduce the parameters and training time of a TTS model while guaranteeing a certain performance due to their high parallelism, which alleviate these economic costs of training. In this paper, we propose a lightweight TTS system based on deep convolutional neural networks, which is a two-stage training end-to-end TTS model and does not employ any recurrent units. Our model consists of two stages: Text2Spectrum and SSRN. The former is used to encode phonemes into a coarse mel spectrogram and the latter is used to synthesize the complete spectrum from the coarse mel spectrogram. Meanwhile, we improve the robustness of our model by a series of data augmentations, such as noise suppression, time warping, frequency masking and time masking, for solving the low resource mongolian problem. Experiments show that our model can reduce the training time and parameters while ensuring the quality and naturalness of the synthesized speech compared to using mainstream TTS models. Our method uses NCMMSC2022-MTTSC Challenge dataset for validation, which significantly reduces training time while maintaining a certain accuracy.

关键词： Text-to-Speech sequence-to-sequence Efficiently Lightweight

来源：评论

学校读者我要写书评

暂无评论

Class Incremental Learning for Character String Recognition 18th

Class Incremental Learning for Character String Recognition

引用

18th International Conference on Document Analysis and Recognition (ICDAR)

作者： Hu, Yijie Zhang, Yan-Ming Huang, Kaizhu Wang, Qiu-Feng Xian Jiaotong Liverpool Univ Suzhou Peoples R China Chinese Acad Sci Inst Automat Beijing Peoples R China Duke Kunshan Univ Suzhou Peoples R China

ISBN: (纸本)9783031705489;9783031705496

Character string recognition (CSR) has drawn much attention for document intelligence, but its performance is limited by the pre-defined character set without the ability to recognize new characters. To overcome this issue, class incremental learning (CIL) can be adopted where the model learns from new data instances incrementally over time. However, it is challenging to directly apply existing CIL methods in CSR because CSR is a typical sequence recognition problem. Without accurate alignment, the recognition error of new characters will affect the recognition of other characters in the same sequence. Moreover, the new characters are usually much fewer than the old ones, resulting in a data imbalance issue for learning new classes. To tackle the misalignment issue, we decouple the learning of feature alignment and classifiers during the incremental process in CSR. To handle the data imbalance issue, we propose a Prototype Incremental Learning framework for CSR, namely PIL-CSR. In the PIL-CSR framework, we propose a prototype-centered loss (PCL) to aid the model in facilitating better class separation, and we further propose a prototype separation and feature alignment (PSFA) strategy, allowing the model to adapt and learn new classes seamlessly. Finally, we collect a CSR dataset to evaluate CIL performance (***/tambourine666/Doc-CIL). Experimental results demonstrate the effectiveness of our proposed sequence CIL method, obtaining a significant improvement in both line-level and character-level accuracy.

关键词： Character String Recognition Class Incremental Learning sequence-to-sequence OCR

来源：评论

学校读者我要写书评

暂无评论

De Novo Design of Target-Specific Ligands Using BERT-Pretrained Transformer 6th

De Novo Design of Target-Specific Ligands Using BERT-Pretrai...

引用

6th Chinese Conference on Pattern Recognition and Computer Vision (PRCV)

作者： Zheng, Yangkun Lu, Fengqing Zou, Jiajun Hua, Haoyu Lu, Xiaoli Min, Xiaoping Xiamen Univ China Sch Informat Dept Comp Sci & Technol 422 Siming South Rd Xiamen 361005 Peoples R China Xiamen Univ China Sch Informat Dept Artificial Intelligence 422 Siming South Rd Xiamen 361005 Peoples R China Xiamen Univ China Informat & Networking Ctr 422 Siming South Rd Xiamen 361005 Peoples R China

ISBN: (纸本)9789819985487;9789819985494

The principal goal of drug design is to find ligand molecules that exhibit affinity to a given target protein. In recent years, deep generative methods have shown their promise in de novo drug design. However, most of these methods design molecules based on target-specific ligand datasets instead of targets' features and fail to design drugs against novel target proteins that barely have active ligand datasets. A fast and relatively accurate evaluation method is needed to evaluate algorithms capable of generating large numbers of molecules. In this work, we treat target-specific de novo drug design as a sequence-to-sequence generation task and propose a Transformer architecture that compensates for the lack of training data with a BERT pretraining approach to generate protein sequence-conditioned Target Ligand Molecules SMILES. First, we pre-train two self-attention blocks of Transformer on the large-scale amino acid sequence dataset and molecular SMILES dataset, respectively, to capture the feature representation of the target. Then we fine-tune the Transformer's encoder-decoder mutual attention block on the protein-ligand complex dataset to learn conditional generation using autoregressive supervised learning. The individual results do not demonstrate the effect of the generative algorithm, so we propose to evaluate the model by calculating the affinity distribution of the molecules. We also evaluate our method by designing ligands against three well-studied proteins. Furthermore, our model proposes molecules with binding affinities exceeding certain FDA-approved drugs in docking experiments.

关键词： Molecular Generative Model Deep Learning sequence-to-sequence Unsupervised Learning Pretraining

来源：评论

学校读者我要写书评

暂无评论

Idiomatizing Python Source Code Using Different Recurrent Architectures

Idiomatizing Python Source Code Using Different Recurrent Ar...

引用

Intelligent Systems Conference

作者： Szalontai, Balazs Marton, Tamas Kukucska, Akos Pinter, Balazs Gregorics, Tibor Eotvos Lorand Univ Budapest Hungary

ISBN: (纸本)9783031664304;9783031664311

Recurrent encoder-decoder architectures have been shown to be capable of solving a wide range of sequence-to-sequence problems, such as translating text, generating summary and transforming source code. In this paper, we investigate the capabilities of recurrent encoder-decoder architectures for refactoring nonidiomatic Python code. We propose multiple possible improvements to the base recurrent architecture and investigate their impact on model performance. In order to also examine the effect that such modifications might have on each other, we train 72 different encoder-decoder architectures on the task of Python code idiomatizing. The best performing architectures are thoroughly evaluated on a real-world dataset, containing programs that were written by students. This procedure necessitates extensive manual evaluation by independent annotators. The results show that the proposed improvements can greatly increase the F1-score of the method compared to the base approach.

关键词： Deep learning Recurrent sequence-to-sequence Encoder-decoder Python Idiomatizing

来源：评论

学校读者我要写书评

暂无评论

English-to-Bangla Language Translation: A Comparative Study of GPT-3 and Attention-Based Encoder Decoder 1

引用

9th International Congress on Information and Communication Technology (ICICT)

作者： Kaysar, Md Nadim Khaleque, Taskin Biswas, Ripon Mazumder, Mahjabin Namira Hossain, Elias World Univ Bangladesh Dept Comp Sci & Engn Dhaka Bangladesh Shahjalal Univ Sci & Technol Dept Comp Sci & Engn Sylhet Bangladesh Bangladesh Univ Business & Technol Dept Comp Sci & Engn Dhaka Bangladesh Daffodil Int Univ Dept Comp Sci & Engn Birulia Bangladesh

ISBN: (数字)9789819735624

ISBN: (纸本)9789819735617;9789819735624

This study is designed as a sequence-to-sequence neural network and GPT-3 for English-to-Bangla machine translation. Here, we have analyzed the results in different ways in this study: The first one is embedding-based analysis with Pearson and Spearman correlation and a maximum of 77% similarity between actual and expected sentences has been achieved, the second one is BLEU score, and the third is the human analysis and a comparatively acceptable score has been obtained. We also obtained a 0.85 BLEU score in this study, and we found 93% accuracy of the generated translation output by using the human-based subjective score.

关键词： English-to-Bangla translation GPT-3 sequence-to-sequence LSTM Bangla machine translation

来源：评论

学校读者我要写书评

暂无评论

Contextual Temporal Feature Modeling Based Distributed Photovoltaic Abnormal Power Data Repair Method

Contextual Temporal Feature Modeling Based Distributed Photo...

引用

International Conference on Smart Energy Systems and Technologies (SEST) - Driving the Advances for Future Electrification

作者： Li, Yanran Liu, Hanghang Huang, Chen Zhen, Zhao Wang, Fei Li, Li State Grid Shandong Elect Power Co Jinan Power Supply Co Jinan 250001 Peoples R China State Grid Shandong Elect Power Co Power Dispatching & Control Ctr Jinan 250013 Peoples R China North China Elect Power Univ Dept Elect Engn Baoding 071003 Peoples R China

ISBN: (纸本)9798350386509;9798350386493

Distributed photovoltaic (DPV) abnormal power data repair can provide high quality data for the research of power forecasting. However, there is a lack of DPV abnormal power data repair methods suitable for scenarios without measured meteorological data, and the repair accuracy of the existing methods is limited for longer abnormal duration. To address these issues, a contextual temporal feature modeling based DPV abnormal power data repair method is proposed in this paper, through which the power repair accuracy can be improved in situations with longer abnormal duration and no available measured meteorological data by merely using the normal power data of the target DPV site itself. Firstly, a sliding window strategy is applied to the historical power data of the target DPV site to generate the sample set used for model training, validation, and test. Secondly, a Bidirectional Gated Recurrent Unit (BGRU) based contextual encoding module containing a preceding encoder and a succeeding encoder is constructed to extract the temporal features of the normal sequences in the context of the target sequence. Finally, a Unidirectional Gated Recurrent Unit (UGRU) based decoding module is constructed to mine the temporal dependency between different time indexes and model the nonlinear mapping relationship between the target power sequence and its contextual normal sequences. Comparative experiments on real-world dataset have testified that the validity and progressiveness of the proposed DPV abnormal power data repair mothed.

关键词： distributed photovoltaics abnormal power data repair sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

Applications that Use Recurrent Neural Networks and Their Impact on People's Lives 1st

Applications that Use Recurrent Neural Networks and Their Im...

引用

21st International Conference on Informatics in Economy (IE) - Education, Research and Business Technologies

作者： Cimpeanu, Ionut-Alexandru Bucharest Univ Econ Studies Fac Cybernet Stat & Econ Informat Bucharest Romania

ISBN: (纸本)9789811967573;9789811967559;9789811967542

Human intelligence has no limits and constantly impresses with innovative solutions that increase in complexity, applicability, and efficiency. Over time, a lot of smart applications have been substantially improved by including artificial neural networks. Thus, they came to perform tasks that only humans could accomplish. Other smart applications made decades ago are the starting point for other more complex smart solutions in which researchers have invested time, money, courage, patience, desire for progress and knowledge, and research. In this paper we highlighted the importance of using recurrent neural networks, we exemplified the use of recurring neural networks in two applications in different fields of activity, and we listed the benefits that can be obtained by using applications, the limits of artificial neural networks, the need for human intervention to obtain desired results. My work is inspired by the recent development in neural translation using the attention mechanism in order to learn long sequences and improve the performance of translation from one language to another. Using this approach, I set out to provide arguments to highlight the effectiveness of implementing solutions based on artificial neural networks, but also the fact that in all intelligent solutions based on this network and which have been created so far, the human factor is indispensable.

关键词： Google neuronal machine translation Long short-term memory Recurrent neural networks sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

SleepTransformer: Automatic Sleep Staging With Interpretability and Uncertainty Quantification

引用

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING 2022年第8期69卷 2456-2467页

作者： Huy Phan Mikkelsen, Kaare Chen, Oliver Y. Koch, Philipp Mertins, Alfred De Vos, Maarten Queen Mary Univ London Sch Elect Engn & Comp Sci London E1 4NS England Alan Turing Inst London NW1 2DB England Aarhus Univ Dept Elect & Comp Engn Aarhus Denmark Univ Bristol Sch Econ Finance & Management Bristol Avon England Univ Lubeck Inst Signal Proc Lubeck Germany German Res Ctr Artificial Intelligence DFKI Kaiserslautern Germany Katholieke Univ Leuven Dept Elect Engn Leuven Belgium Katholieke Univ Leuven Dept Dev & Regenerat Leuven Belgium

Background: Black-box skepticism is one of the main hindrances impeding deep-learning-based automatic sleep scoring from being used in clinical environments. Methods: Towards interpretability, this work proposes a sequence-to-sequence sleep-staging model, namely SleepTransformer. It is based on the transformer backbone and offers interpretability of the model's decisions at both the epoch and sequence level. We further propose a simple yet efficient method to quantify uncertainty in the model's decisions. The method, which is based on entropy, can serve as a metric for deferring low-confidence epochs to a human expert for further inspection. Results: Making sense of the transformer's self-attention scores for interpretability, at the epoch level, the attention scores are encoded as a heat map to highlight sleep-relevant features captured from the input EEG signal. At the sequence level, the attention scores are visualized as the influence of different neighboring epochs in an input sequence (i.e. the context) to recognition of a target epoch, mimicking the way manual scoring is done by human experts. Conclusion: Additionally, we demonstrate that SleepTransformer performs on par with existing methods on two databases of different sizes. Significance: Equipped with interpretability and the ability of uncertainty quantification, SleepTransformer holds promise for being integrated into clinical settings.

关键词： Sleep Transformers Brain modeling Uncertainty Databases Task analysis Electroencephalography Automatic sleep staging deep neural network interpretability sequence-to-sequence transformer uncertainty estimation

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：