Google’s decentralized strategy speculation — Federated computation on the way to Web3
Since Google released Private Compute Services in October, it has increasingly clearly outlined a technical strategy for personal privacy data protection. As a leading international company in cloud computing, we believe that privacy-protected cloud services will be Google’s next strategic goal that will be one of google solution for Web3 world.
In the current and next computing era, personal privacy data will be the most sensitive issue in countries around the world. Internet giants will invest heavily in this field to avoid increasing legal risks. Personal privacy data protection requires a full-stack privacy protection solution, that is, a closed-loop system of data circulation, which involves data calculation (storage, processing), management, and transactions. Only solutions that cover these links can truly protect personal privacy. Therefore, this article analyzes Google’s data cloud strategic plan from the perspective of full-stack privacy protection.
Distributed Computation for Big Data
Federated learning and analysis is Google’s core privacy computing infrastructure, and it is the latest solution that takes into account versatility, performance, and security. At the PPAI (1) conference in February 2021, Google introduced the latest federated learning and federated analysis tutorials (2). For the first time, federated learning was divided into two categories.
Cross-device (Cross-device) and cross-data set (Cross-silo) federated learning, the processes and scenarios of the two schemes are different. In our opinion, the main difference is that the cross-device solution is used to protect personal privacy, and the cross-data set solution is used to protect corporate confidential data. Speaking of this, everyone will definitely think of the famous solution “federal learning” in B2B territory, which is a typical cross-silo solution. What is more gratifying is that it is not only officially classified by the industry, but also officially included in the federated learning open topic (3) in March 2021 to promote development. Although this article will not discuss in depth, we still want to congratulate FATE for cross-silo Open source projects for federated learning.
In the cross-device solution roadmap, Google’s federated learning framework becomes a separate component Private-Compute Core and Private-Compute Services, which can be downloaded in the Play Store, in the new Android12.
The latest GPhone Pixel 6 is equipped with Google’s latest SoC, an integrated CPU and GPU chip, which can greatly enhance the machine learning computing capabilities of mobile phones and establish high-performance platform nodes for joint computing. We will continue to release technical analysis on hardware computing enhancements in the future.
With the application of Private Compute Services, Core and third-party applications can be connected, enabling Private-Compute to serve the entire Android ecosystem. The above figure shows that the so-called Trusty OS is a generalized framework based on TrustZone, which should guarantee the integrity of the Core. Here is a mention of Alita’s solution, which is also based on TrustZone, but limited to the computing power of the three parties. We use TrustZone for consensus (PoTA) computation.
The above figure(7) shows the relationship between Core and Services. Core does not have network access capabilities and is used to ensure privacy and security, and is unified by Services.
If you want to play to the value of data, you must make the data manageable, generally using a metadata system. In order to cooperate with the federated learning service to have a high-quality, large-scale data set, Google released the incubation project FLoC (4) in 2021, a management system for Chrome browsing behavior metadata, which can group users according to user browsing behaviors , As well as confidential classification and inspection to ensure that the user’s interest is classified correctly and no privacy is leaked. These data will be read by the federated learning framework. On the surface, the cookie format is unified in the field of online advertising, and the deep-level strategy should be to provide data sets for privacy protection for federated computing. We believe that more data sets will be provided in more fields in the future.
There are also related projects that have planned a more comprehensive and open metadata management system in advance. For details, see DataDEX’s DBGraph (5)
The value of data needs to flow, and the traditional way is through direct data market transactions. For cloud computing vendors such as Google, AWS, and Azure, they will not provide a direct data market, but will use SaaS and PaaS platforms. In this form, we become a data cloud service, which directly provides cloud services by combining data computation and algorithms with data. Can greatly enhance the efficiency and security of privacy-reserved data utilization.
The current federated computation is still being tested in Google’s internal projects. It is expected that in the future, the federated computation power and data sets in the field like FLoC will be opened to the outside world through products such as Data Studio, BigQuery and even Adsense to achieve the purpose of data value transactions.
Here to mention the trading method of data value in the blockchain. DataDEX first proposed the DeFi in Data Economic (6) program last year, which is to trade data in the decentralized trading market and use the AMM automatic market-making mechanism to establish compliance. Flexible pricing model of market supply and demand. This new solution focusing on the data economy has led the trend of decentralized data transactions. In addition, unlike other data markets, DataDEX’s solution uses FLoC-like ideas to build large-scale high-quality data sets, rather than a flea market for data.
As mentioned above, Google’s strategy has been laid out in various fields of data computation, management, and exchange. Then, we have made some predictions on how it will develop in the future.
The current federated learning requires a centralized scheduling service. In the future, if computing is provided on the cloud to provide public services, it will definitely face centralized compliance, credibility, and technical scale issues. Therefore, long-term development will provide decentralized Computing network. This goal is also included in the open agenda (3) as a research and development goal.
The fully decentralized federated learning is based on the P2P communication mode of the WAN on the network, which is similar to the blockchain network. Let’s make a bold guess here, is Google in the future blockchain still Google? This possibility is increasing. . . . .
However, Alita-Network has already practiced decentralized federated computation before the giants, and strives to use blockchain technology to make up for the credibility and incentive network of this decentralized computing.
On the other hand, Google will continue to public from the federated series of services and gradually develop towards general federated computation, which is also a necessary stage for its data cloud service goals to be achieved.
In terms of business model, Google can be regarded as a data-driven advertising company in which any business revolution must start with data. Therefore, it’s reasonable to believe that the decentralization plan of Google’s federated computation layout has officially moved closer to Web3. Because in the past, Web2, as Google’s business interface layer, has been built as a search engine based on data distributed computing. In the Web3 era, Google still wants to repeat this process. Although they have tried in the financial-centric Web3 applications in the early days (8), the Android 12 and Federated Computation plans have revealed that they are rapidly developing compute and data centric of Web3 era.
1, PPAI. “The Second AAAI Workshop on Privacy-Preserving Artificial Intelligence (PPAI-21).” https://ppai21.github.io/.
2, FC. “Privacy and Federated Learning: Principles, Techniques and Emerging Frontiers.” https://ppai21.github.io/files/BM_KB_PK-slides.pdf.
3, AO-FL. “Advances and Open Problems in Federated Learning.” https://arxiv.org/abs/1912.04977.
4, FLoC. “Federated Learning of Cohorts.” https://en.wikipedia.org/wiki/Federated_Learning_of_Cohorts.
5, DataGraph. “Meta Data Solution of Open Dataset for Personal Privacy Data.” https://smallpdf.com/file#s=4e2c6d34-7503-41fa-9b9a-e63dd84c9428.
6, DataEconomic. “DeFi in Data Economy: Decentralized Data Exchange based on privacy-preserving computation network.” https://gravity-link-team.medium.com/defi-in-data-economy-decentralized-data-exchange-based-on-privacy-preserving-computation-network-c6b25c88ea1f.
7, PrivateComputeCore. “The Google Pixel 6 and Pixel 6 Pro have a new Private Compute Core, and here’s what we know about it.” https://www.xda-developers.com/private-compute-core/.
8, GoogleWeb3. “Google Takes Giant Step Towards Powering Blockchain-Based Web 3.” https://www.forbes.com/sites/michaeldelcastillo/2021/09/14/google-takes-giant-step-to-powering-web-3-with-dapper-labs-nft-deal/?sh=4d581d2639ca.