Recently, China Mobile held the "2021 Science and Technology Workers' Day and Science and Technology Week -- Nine Days Artificial Intelligence New Technology Forum", inviting industry experts from China Mobile, outstanding artificial intelligence enterprises, top universities, the international Organization for Standardization to share the latest technological achievements and industry trends in the field of AI basic technology and network intelligence, to discuss the future development trend of AI technology, promote Industry number intelligent transformation. Yang Jian, senior architect of Jiuji DataCanvas, was invited to attend the forum and delivered a wonderful speech on "The new generation of AutoML helps enterprises scale modeling".
In recent years, the application value of machine learning technology is remarkable, and the subsequent bottleneck of data, computing power and talent is increasingly prominent. With the development of artificial intelligence, the demand for data and computing power can be met, but the training of data science talents lags far behind the demand, resulting in problems such as long training cycle and slow speed. Yang Jian, a senior architect, mentioned in his speech that AutoML technology is experiencing explosive growth. Gartner listed AutoML as one of its key strategic technologies in the field of data and analysis as one of its top ten technology trends in 2020.
Facing the new needs of enterprise business, Jiuzhang DataCanvas provides AutoML modeling platform to build the search space of neural network architecture, apply the super-large spatial search algorithm and combine with high performance evaluation strategy to achieve end-to-end automatic machine learning modeling of the whole process, improve the efficiency and experience of modeling, and enable the business to quickly build and use AI capabilities .
Yang noted that AutoML, while growing, faces many challenges, including uneven data, concept drift, model generalization ability and automatic modeling of large-scale data. To solve these problems, in the modeling process, DataCanvas automatic machine learning platform uses the AutoML core automation technology to enable business personnel to complete data cleaning, feature processing, model screening and hyperparameter optimization without checking data sample distribution, and automatically deal with imbalance problems. At the same time, semi-supervised learning is introduced to further improve the generalization ability of the model and eliminate the uncertainty of data training and prediction. The platform supports distributed architecture, provides efficient evaluation strategies, and reduces the time cost of modeling, so as to realize AutoML for large-scale data.
Based on self-developed "hard technology", the Cloud DataCanvas has opened source the international leading AutoML automation model algorithm and architecture, and has many Au such as Hypernets to realize the development of diverse machine learning tasks and scene automation, DeepTables which won the first place in the global top Kaggle competition toML Open source tool. Moreover, according to the published data sets of UCI and Kaggle, the model accuracy of its product DataCanvas AutoML Toolkit(DAT) is higher than that of the most advanced AutoML open source tools.
With the ever-changing business and market demand, Jiuzhang Cloud Ji DataCanvas has realized the scene application in many industries such as finance, communication and government over the years. With continuous innovation, Jiuzhang Cloud Ji adheres to the independent research and development of new technologies and new products, and covers more customers with standardized AI platform products, helping enterprises in various industries to scale modeling and realize the intelligent upgrade of data.
