AI is difficult to land in the enterprise market in the form of APIs due to multiple factors. By building a PaaS-based AI platform that supports the "Auto Machine Learning" feature, Sky Cloud Big Data tries to empower AI for customers, making it as easy as reading a book for enterprises to acquire machine intelligence.

Transcript of the presentation.

There are many hot topics discussed about AI nowadays, such as face recognition, chatbots, human-computer interaction scenarios, etc. These are application scenarios rather than the core of AI. The core supporting technology for AI is the technology framework built around Machine Learning (ML).

In mature financial institutions, especially in departments such as risk control and strategy, there has actually been a focus and introduction of the use of AI, ML. So, it's not that AI has entered finance only with the rise of this AI boom, but it has been at play in risk and pricing for more than a decade before that.

To share a case study with you, a joint-stock bank's app, which has over 600+ personalization models behind it, involves the construction of numerous ML models. With so many ML models in use, we should be focusing more on the ability to use AI at scale, rather than limiting ourselves to the scenarios in which it is used. At the Global AI Technology Conference a few days ago, the head of Google Tensorflow also addressed the issue of how to use AI at scale.

AI has become the hot spot of the moment, the future development of AI will focus more on the scale production capacity

Throughout the evolution of AI, we find that the current technology companies have started to shift towards AI. As Google CEO Sundar Pichai proposed a shift in strategy, Google shifted from Moblie First to AI first and started investing more in machine learning.

Over the past 15 years, we have leveraged the convenience of Mobile to make commercialization cost effective and efficient. In the next 15 years, the technology revolution driven by technology companies will focus more on the use of technology in the direction of AI, which will lead to further upgrades in business models. In the financial sector, the biggest challenge encountered with AI-centric technology change is how to land AI into business processes.

When we serve our bank clients, we cut through at the system level. By looking at how to implement machine learning at the system level, we found that systems built with machine learning as a core technology are very similar to the role of Android in the Mobile industry. And so we have found our niche, which is to build the Android of AI.

Over a decade ago, in Mobile, it took a big company to implement very simple application systems such as Tetris. This is because different phones have different operating systems and require the development of different apps. In the following years, the emergence of Android has solved this problem very well. The Mobile industry generally adopts Android as the smart phone operating system, and the development of mobile phone APP is no longer the patent of big companies, and the team of fresh graduates can also develop their own APP based on Android.

In the AI space, we find the same process to be true. At the very beginning of the interface with the client, the entire process from data entry, data cleaning, to building the model took several months.

We believe that the use of a process-oriented approach will reduce the time spent on building processes and thus increase productivity. With increased productivity, we can then find more production capacity at scale to solve more problems. An overall virtuous circle. Based on this logic, we have done practice in several joint-stock banks and the results are remarkable.

Thus, AI is not an elite technology in the ivory tower anymore. AI can be a practical help in business processes, expressing some complex phenomena that were previously difficult to describe in a quantitative way. For example, the decision engine, which has received more attention in the industry, is a process-driven automated system that abstracts clear rules and processes from complex financial operations.

And today we are faced with a data-driven automated system, how to re-express complex business, transform traditional processes and improve productivity with scaled AI applications to save time. This is the future trend of AI development and the current business logic of SkyCloud.

Empowering enterprise customers with AI PaaSification and reducing reliance on data scientists

By doing AI empowerment for our clients, we've found more than just helping them save time, we've helped them uncover more scenarios. These scenarios are not defined by us, but rather, after the client is empowered, their own team is able to apply them to create and improve the business processes corresponding to each scenario, such as lending and approval process improvements. After the constant generation of large amounts of data through empowering customers, followed by customer business processes, standardized data, which is extremely important for machine learning, started to emerge. How to use such data to enable rapid model building for customers places demands on the online modeling capabilities of the system platform, and these demands are beginning to be focused on the business side.

We have positioned ourselves in the role of empowerment. We want to empower our customers by taking the work of data scientists and business elites in the industry and making it possible with machine learning technology. This process of implementation can be well described as "demystifying", that is, removing the mystery from the industry and replacing expensive business elites like data scientists with platforms. The West is known as the democratization of AI.

A platform that can replace data scientists is the distributed data science platform launched by SkyCloud: MaximAI

There are three core drivers of AI: A (Algorithm), B (Bigdata), and C (Cloud). The convergence of these three is necessary for the commercial rollout and implementation of this wave of AI at scale. However, ABC can be difficult to APIize when it is geared to the enterprise market.

In the past, more microservicization was pushed in the application-side market, which means that services such as ID card OCR scanning and text information refinement are realized by calling API interfaces. Through API interface calls, the above AI services can also be used by ordinary business people with weak data usage skills.

And when facing the enterprise market, the data and algorithm choices for building models are different, plus many enterprise customers tend to private cloud service model, then to do better AI empowerment for customers, our understanding is to achieve through the PaaS of AI.

Build a Sky Cloud AI platform and lower the threshold of AI platform application

The PaaS-based AI platform we built has also gone through several stages of change. 1.With version 0, we focus on solving the problem of data-based online modeling and move from the tool idea in the pre-technology accumulation phase to platform building.

Once the platform is completed, customers can stop writing programs when using the platform and only need to configure hyperparameters. 1.Another feature of version 0 is the use of parameter servers.

2.In version 0, we started to focus on doing evaluations for the algorithm. To achieve the best business enhancement, choose the algorithm that achieves the best results. Also, it supports the application of different models to them based on different data sources, and provides model lifecycle management. In addition, we pioneered the introduction of an open interactive environment where users can put cutting-edge algorithm packages on the platform for automated execution.

As for version 3.0, what we mainly do is to allow more cutting-edge metrics for tuning parameters and algorithm selection, which in turn enables automated optimization of algorithms, also known as Auto Machine Learning. We want to make it simpler for customers to choose algorithms to build machine learning models based on SkyCloud's PaaS-based AI platform, just like Android-based open applications. It enables containerized deployment in notebook environments, automatic determination of algorithms based on task types, feature engineering automation, automatic derivation or synthesis of features, and intelligent optimization of model superparameters.

Overall, after these platform iterations, the SkyCloud platform has been able to achieve intelligent production of models and truly realize Auto Machine Learning. Through a series of automated methods, such as automatic algorithm selection, automatic parameter tuning, and automatic feature engineering, the reliance on data scientists has been successfully reduced, lowering the threshold for AI adoption in the enterprise market.

Later, we will share a few successful cases, such as the credit card application anti-fraud done for a large joint-stock bank, by analyzing the "credit card" information and "fraudulent credit card" information of the bank credit card, finding the relationship contained in the registration information, while statistical analysis of the relationship information, calculating relevant indicators, and then building a social network through the results of statistical analysis, and finally supporting fraudulent users.

Thomson Reuters uses a semantic classifier built by SkyCloud to machine read hundreds of thousands of public company announcements each year, which can read over 200,000 announcements per year, replacing 10 senior financial analysts.

By extracting the features of the schematic image, SkyCloud builds a model for automatic fault identification and early warning of oil wells. Taking early action before a well fails prevents further damage or even scrapping, while reducing the need for manpower, increasing the accuracy of judgments, and greatly reducing the cost of oil company operations.

Making access to machine intelligence as easy as reading a book. The future direction of SkyCloud will still be along the lines of enabling companies that make up the majority of the industry with weak or even zero AI capabilities to better embrace the AI era.

Love Analysis Hot Articles

Recommended>>
1、What do people who know programming think about the WeChat Jump that broke out in the friend circle
2、Ubiquity Education Advanced Building Block Robot Owl
3、Why is your version of python a mess Because this operation is missing
4、The Group acts as a matchmaker for UWM and City Womens Hospital
5、Ether is being threatened claiming to have a weapon that can destroy it who is so arrogant

已发送

发送中