Islands of data, data privacy and data security, cloud computing and artificial intelligence is not open around the large-scale industrial applications in the process of "three big mountains."
The case of "federal study" as a new generation of artificial intelligence algorithms, the data can not local, to achieve common modeling, enhance the effect of AI model to ensure data privacy, data silos and restrictions breakthrough small data, which will undoubtedly It became one of the ways across the "three big mountains". Therefore, as a federal study the world's first industrial-grade open source projects, FATE also by the parties concerned, to join the community of developers building have expressed expectations. (FATE open source community address: https: //github.com/FederatedAI/FATE)
And after the introduction of incentives contributors, FATE ushered in the first place an open source community contributors - Liu Yang from Tencent cloud. Federal learn how to energize Industry Data Security? How privacy practitioners evaluate FATE? Dr. Liu Yang expressed his views in an interview.
Improving Efficiency 70% data calculating acceleration landing enterprise applications
Dr. Liu Yang graduated from the Australian National University, is also a senior fellow at the cloud Tencent, Tencent Aegis sandbox responsible for the privacy protection algorithms section. Liu Yang said that for the sake of their own field practitioners, from the beginning of the year began to pay close attention to the "federal study."
Therefore, FATE into their field of vision, has been focused on Liu Yang and Tencent cloud team. After an in-depth understanding of FATE, Liu Yang believes Tencent Aegis sandbox to create privacy + distributed learning concept, and to solve FATE "data security" "data privacy" "Data Compliance" three major problems and do not seek together, and gradually began to use FATE meet the functional requirements of the Aegis sandbox.
Liu Yang said that after long-term exposure, logistic regression algorithm flow FATE and XGBoost very much agree, therefore FATE have begun to join the open source community building, an optimization proposal - using symmetric affine alternative Paillier password password will enhance the training time 70 % or more, and thus to the same operational state "burden." Future joint venture in the FATE version after the application optimization, can effectively reduce the time cost data operations, enhance the technological competitiveness of enterprises in the AI era.
Industry Data Security imminent before loading the line
AI application scenarios, the traditional multi-party cooperation to data centric consolidation process, there is a serious loss of privacy issues, the crux even become a key obstacle to enterprise scale applications of AI.
Liu Yang in view, the key to breaking the data is still safe solution to the problem, namely the problem of data privacy and compromise the utility. Specifically, in order to secure your data out of the island, must go through certain "masked" Operation: influence by Cryptography effective tool to convert data into gibberish, privacy saved, but the key is in the hands of anyone, great utility data; confused with the noise may be raw data, e.g., differential privacy, the greater the noise, the Privacy guaranteed, but the user to get lower the data play utility. How to seek a compromise between privacy and utility of the road in the middle, it is one of the key issues of data security in circulation.
Future ideal state, any data on the user can flow freely distributed and aggregated data, efficient data mining operations, but no sense of privacy protection bond. In the MPC (Multi-party Computation, multi-party Secure Computing) field, the current industry still in confusion circuit, the trusted computing solutions, although support has a general computing tasks, but it requires additional hardware support, learning higher costs, hinder large-scale applications, while data security is not conducive to the formation of alliances.
The federal study in the federal framework is universal in for each one or each type of machine learning algorithm customized privacy transformation, so their use is tantamount to a classic center-type machine learning models. By contrast, the Federal learn to stabilize on the basis of cost, ensuring ease of use. Liu Yang said that for businesses, federal solutions provide learning more attractive; for industry, more convenient operation will attract developers to invest more in order to advance the construction of secure data Union.
FATE ecological × Tencent cloud data security can be expected in the future
Since the beginning of May this year, FATE and Tencent cloud Aegis sandbox began to conduct business and technical exchanges, the core of the current Aegis sandbox calculation module is provided by FATE. In a platform, the two sides cooperate closely. Liu Yang said in an interview, when using FATE team framework, algorithms, will contribute to effective proposals FATE open source projects, participate in the open source community building.
This with the "mutual support, open source Build" special forms of cooperation, in the well-promoted Aegis polished product and sandbox FATE project, but also to other projects or technical team provides a good model demonstration - - with an open attitude to embrace new technology, not only beneficial in itself, it will also boost the development of the industry.
Liu Yang in the vision of the future can be both deeper level of collaboration in enhancing the impact of technology and services such as landing, for example, co-authored important papers, filed patents and jointly take over external within the actual business, the formation of "academic", "industry" two flowering situation better.
As more and more contributors to join the construction industry FATE theory and application of standards, FATE is bound to usher in a broader perspective. In this regard, Liu Yang said that security take root and grow Aegis sandbox and will accelerate data FATE join forces and build future data security alliance on data silos can be expected.