最新刊期
卷 10 ， 期 5 ， 2024
- 摘要：In recent years, artificial intelligence has been widely applied in multiple fields, and the "pre-training and fine-tuning" of large models (LLMs) has become the latest paradigm of artificial intelligence. Distributed technology exists at every stage of the lifecycle of LLMs, providing support for them. In the data acquisition process, the file system called "SuperFS", was developed to address the storage issue of massive small files, which can meet the requirements of low latency and scalability. In the data preprocessing stage, an efficient big data processing engine called "Chukonu" was developed to address the issue of high overhead in reading data from distributed file systems. In the model training stage, a distributed checkpoint strategy was proposed to address the problem of poor read and write performance of checkpoint files, greatly improving the read and write speed of checkpoint files. In the model inference stage, a high-throughput inference scheme called "FastDecode" and a LLM inference architecture called "Mooncake" were developed to address the challenge posed by KVCache to storage system. The applications of distributed technology enable LLMs to fully utilize computing resources, accelerate training speed, and benefit the development of the field of artificial intelligence.关键词：distributed technology;large language model;massive small files;big data processing engine;checkpoint;KVCache549|1038|0更新时间：2024-09-15
STRATEGY RESEARCH
-
- 摘要：In recent years, large language model have demonstrated outstanding capabilities across multiple natural language processing tasks, greatly aiding the development of intelligent law systems. Existing large language model in the law domain, primarily focuses on fine-tuning foundation model for simple question answering tasks in legal consultation, have not considered other application scenarios in the legal field, such as legal information extraction, judgment prediction, etc. Real world law services are far more complex than dialogues. LawLLM was proposed, a Chinese law intelligent large language model, capable of providing diverse law services for different user groups. Additionally, the application of information extraction on long-context judicial documents in the law domain was explored. The average performance of LawLLM on Lawbench outperforms all the compared large language models, being 0.19% higher than the 175 billion parameter GPT-3.5-Turbo model. LawLLM’s average performance on few-shot tasks on Lawbench is slightly lower than GPT-3.5-Turbo, by 0.02%.关键词：large language model;intelligent law;natural language processing511|216|0更新时间：2024-09-15
- 摘要：Large language model (LLM) has demonstrated significant application potential in the medical field. However, evaluating the performance of LLM in medical scenarios poses a challenge. Existing medical benchmarks, predominantly in the form of multiple-choice questions, struggle to comprehensively and accurately assess LLM's performance in pediatric domains. To address this issue, PeMeBench, the first Chinese pediatric question-answering benchmark, was proposed. Leveraging a dual-perspective evaluation dimensions and referencing diagnostic and treatment guidelines from 10 pediatric disease systems, PeMeBench meticulously categorized pediatric medical question-answering tasks into five subdomains: disease knowledge, treatment plans, medication dosages, disease prevention, and pharmacological effects. It comprised over 10 000 open-ended question-answering items and introduced a multi-grained automated evaluation scheme that integrated entity retrieval with the detection of hallucinated sentences. This approach aimed to provide a comprehensive and precise assessment of LLM's performance in pediatric healthcare, delving into their potential limitations and laying a solid foundation for enhancing the intelligence level of medical services.关键词：pediatric medicine;benchmark testing;large language model;Q&A175|77|0更新时间：2024-09-15
- 摘要：In order to implement the "precision supervision" policy, tobacco companies need to increase the hit rate of tobacco-related cases. Past approaches had lacked research on high-risk parties in tobacco-related cases, which had hindered the improvement of case hit rates. Based on the large amount of historical data stored in tobacco companies, mining accurate warning lists is an effective way to improve case hit rates. After conducting the analysis of high-risk parties features, a gated variable relationship graph convolutional network was proposed to obtain an accurate high-risk parties warning list. Firstly, the gated variable relation graph convolutional network used variable relation graph convolutional network to capture the relationship and key features of the parties. Then, the gated layer was applied to further learn the features. Finally, the learned features were inputted to the Softmax layer to get the classification results, and then an alert list is obtained. Through comparison experiments, the constructed model is proved to be more effective. After a municipal monopoly bureau applied the results of this project, its case hit rate improved from about 0.01% to about 0.5%, which proves that the early warning model can meet the needs of real regulation.关键词：Chinese tobacco industry;cigarette case;high-risk parties recognition;gated layer;graph convolutional network115|82|0更新时间：2024-09-15
COLUMN: APPLICATIONS DRIVEN BY DATA AND COMPUTING POWER
- 摘要：As a significant research area in the field of speech technology, speech synthesis is dedicated to converting text into speech. With the rapid development of deep learning technology, the objective of speech synthesis has evolved beyond merely producing "understandable" audio. The incorporation of emotion often enhances the expressiveness of synthesized speech. Consequently, emotional speech synthesis aims to combine speech with different emotions and regulate these emotions to generate flexible and precise emotional speech. Starting from several key issues in emotional speech synthesis, this paper summarizes and analyzes the development based on emotion transfer, emotion intensity control and emotion mixing in recent years, and introduces the relevant data sets and evaluation indicators of emotion speech synthesis. Finally, the emotional speech synthesis is prospected.关键词：emotional speech synthesis;emotion transfer;emotion intensity;deep learning328|110|0更新时间：2024-09-15
- 摘要：In the interdisciplinary field of modern computer vision and natural language processing, digital talking facial generation technology has become an increasingly important research topic. Digital facial generation technology focuses on generating realistic facial images based on predetermined text or audio sequences. In recent years, deep learning methods such as convolutional neural networks, generative adversarial networks, and neural rendering fields have been used for digital talking face generation, which shows significant research and application value. These methods have not only attract widespread attention from the academic community, but also have been applied in industry to solve specific problems in image processing and computer vision. Although some progress has been made, the practical application of these technologies still faces many challenges. Comprehensively review and evaluate the specific implementation of deep learning methods in the generation of digital talking face to identify the advantages and disadvantages of existing methods, explore common problems that need to be solved, and highlight open issues that still require further research. In addition, currently available datasets from a statistical perspective were listed, evaluated and compared so that researchers can more easily choose datasets that meet their needs.关键词：digital talking face generation;virtual human;audio-driven173|83|0更新时间：2024-09-15
- 摘要：Behavioral representations of the patients with depression reflect the clinical features and condition of the patients, therefore it is beneficial for disease diagnosis. However, in the construction of current depression lexicon, the correlation between the behavioral features and the condition of patients in depression texts is overlooked, resulting in incompleteness of the lexicon information. To address this problem, a domain lexicon construction, oriented to behavioral features in depression. was proposed which aimed to extend the domain lexicon's coverage of emotional expressions. Firstly, the seed word sets of sentiment and behavior were constructed by the TF-IDF algorithm respectively, the word set of sentiment was obtained by calculating PMI similarity between the seed word set of sentiment and the existing sentiment lexicon Secondly, the seed words of behavioral were labeled based on correspondence between behavioral features and the condition of patients, and further inputted into WoBERT with depression texts to separately generate dynamic word vectors. In addition, the candidate word set was acquired by calculating the similarity between the seed word set of behavioral and depression texts In addition,based on the similarity between words, the semantic graph was constructed to obtain the word set of behavioral features by label propagation algorithm. Finally, the emoticons with negative emotions on Weibo were collected to build the word set of emoticons. The word set of sentiment, the word set of behavioral features and the word set of emoticons were integrated into the Chinese Depression Domain Lexicon. Experimental results show that the constructed lexicon can improve the effect of depression text classification.关键词：depression;domain lexicon;behavioral feature;WoBERT;label propagation algorithm143|90|0更新时间：2024-09-15
- 摘要：Disk, as a crucial storage medium, can result in significant data loss if it malfunctions, causing immeasurable losses for individuals and businesses. Existing models for predicting disk failures have problems such as imbalanced disk data samples and underutilization of the temporal characteristics of the data. In this study, we focused on real disk data provided by the Backblaze cloud storage company and proposed a disk failure prediction model that combines a convolutional neural network (CNN) with a bidirectional gated recurrent unit(BiGRU) network, incorporating an attention mechanism.In terms of data preprocessing, we employed negative sampling and a focal loss function to balance positive and negative samples. Subsequently, we utilized CNN for feature extraction and combined it with BiGRU to effectively handle temporal data. The integration of an attention mechanism enables the model to quickly capture more critical feature informations. The selected features were then trained with the input data into the model. Compared to other fault prediction models, the proposed model in this paper demonstrates a performance improvement of 1% to 7% on four evaluation indicators, such as precision. This provides a robust support for enhancing disk storage reliability.关键词：attention mechanism;disk failure prediction;convolutional neural network;bidirectional gated recurrent unit;focal loss function163|226|0更新时间：2024-09-15
STUDY
- 摘要：As the main carrier for government agencies to handle documents, services, and meetings, official documents play the role of information recording, exchange, transmission and instruction release in the construction of digital government. The establishment of an official document labeling system is an important means to promote the high-quality handling of government documents and improve the administrative efficiency of the government. Aiming at the official document label, a method for constructing an official document labeling system based on digital government construction was proposed, and a government official document labeling system with the core of business scenario drive and user demand orientation was developed and designed. Through the design, generation and management innovation of government official document labels, this system aims to solve the problems of low efficiency, scattered content and cumbersome processes of government official document management, so as to improve the efficiency of government official document management and operation. At the same time, the innovative application of official document labels in different scenarios was expounded, and the application cases of official document labels based on government agencies was given, which provides important support for promoting the high-quality development of digital government.关键词：digital government;official document labeling;application scenario;smart government affair197|89|0更新时间：2024-09-15
APPLICATION
- 摘要：Data as a factor of production (i.e. data factors) is the key momentum for boosting digital economy. Turning public data into data factors needs the synergy between technology and mechanisms. In this paper, we categorize public data into two types and define the four main parties participating in the generation of public data factors. Dealing with these two types of data and four parties, we propose a data-right operational system, which is comprised of a data-right allocation regulation and a set of data-right authorization mechanisms. While ensuring the security of the public data and the compliance with data-right regulations, the system can protect the rights of each parties, enabling them to use data more effectively and efficiently, and thus unlock the power of public data.关键词：data factors;public data;data-right operational mechanism;data-right allocation regulation;data-right authorization mechanism182|248|0更新时间：2024-09-15
-
Research on the difficulties and countermeasures of data quality management in smart governance AI导读
摘要：Data quality is the foundation for implementing smart governance. Based on a case study of the construction of smart governance projects in City D and data collection from different channels, this article aims to analyze the challenges and their reseasons for data quality management and propose the countermeasures from the perspective of data users. The challenges are inadequate management systems and mechanisms, mismatched data quality evaluation indicators, and obstructed information transmission during collaboration. The reasons include the particularity of quality management requirements in project construction, the difficulties in matching data resources with usage needs, and consensus on quality management difficult to support management execution because of the inertia of thoughts. The following suggestions are brought forth: enhancing the degree of consensus on quality management concepts, upgrading the dynamic ability of quality evaluation methods ,and optimizing the refinement granularity of data collaboration strategies.关键词：smart governance;data quality management;data empowerment170|111|0更新时间：2024-09-15 - 摘要：Large language models based on massive data training bring the possibility of generalized artificial intelligence, but also bring new risks and challenges to data privacy protection. This paper analyzes the risks of data privacy protection in the whole process of large language model, argues the new ethical difficulties faced by the principle of informed consent and the principle of "justification and necessity" of data collection, and explores the possible solution frameworks and paths, as well as the ethical difficulties that may still exist in the practice.关键词：large language model;generative artificial intelligence;data privacy;informed consent;data liability248|118|0更新时间：2024-09-15
FORUM
- Postal code：100079
- Tel：（010）53879208、53859533 Email：bdr@bjxintong.com.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：273110 Visits Today：16
0