Several excellent practices for building machine learning platforms on the cloud

​Translator | Bugatti

Reviewer | Sun Shujuan

Most people are familiar with major technology platforms such as iOS, Windows and AWS. A platform is essentially a set of technologies that serve as a foundation for building, contributing to, experimenting with, and extending other applications. They bring many of today's advanced technological capabilities and cutting-edge customer experiences to the table.

To keep up with the scale and complexity of the technological capabilities enabled by big data, artificial intelligence, and machine learning, many companies are developing complex in-house platforms. In fact, Gartner predicts that cloud-native platforms will form the basis of more than 95% of new digital initiatives by 2025, up from less than 40% in 2021.

In my experience, enterprise technology platforms are transformative: they enable cross-functional teams to test, launch, and learn quickly, reduce duplication, standardize capabilities, and provide consistently integrated experience. In short, they help turn technology into a competitive advantage.

1. The evolution of enterprise platforms

#Many organizations leverage cloud-native platforms such as Kubernetes that can run the heavy lifting of artificial intelligence and machine learning. Laiyu is good at providing first-class customer experience. Capital One has become the first financial institution in the United States to fully invest in the cloud. The ability to redesign the architecture of the data environment is indispensable for consolidating its cloud-based platform capabilities. With this solid foundation, Capital One is better positioned to leverage big data to build new machine learning capabilities across its enterprise platforms to accelerate, enhance and deliver new, more meaningful customer experiences.

Much of Capital One’s work in this area has produced significant results for the company and its clients. Its fraud decisioning platform, for example, was built from the ground up to make complex, real-time decisions. By leveraging large amounts of data and enabling models to be updated in days instead of months, the platform helps protect millions of customers from credit card fraud and can be used by the company’s various stakeholders.

Based on my experience leading teams to deliver enterprise technology platforms, here are the key lessons and best practices learned along the way:

• Everything Starts with Teams: Build cross-functional teams made up of the best people, even if it slows down your work at first. Bigger teams aren't always better! The team must at least include product managers, engineers and designers. Staff these functions with people who truly understand the users of the platform. For example, if you are building a platform that will be used primarily by data scientists, hire a product manager who is a former data scientist, or add a data scientist to the leadership team. If your team is made up of people from multiple departments, make sure you have a common goal.
• Clearly define the end state first: Before you start building, take the time to clearly define the architecture and plan for the end state, and achieve your goals in an iterative manner. Make sure the architecture is designed for self-service and contribution from the start. Better yet, design your platform with the assumption that you will extend it to users outside of your organization or business unit. Also assume that over time, as technology changes, you want to be able to replace components.
• Estimate how long you think it will take, then double it: The important thing is to take the time to brainstorm all the capabilities you need to build from the beginning, and then devote the appropriate amount of effort to each part. Once the technical team combines this with speed and estimates how long it will take to build each feature, increase the buffer amount by 50%. In my experience, this estimate ends up being pretty accurate.
• Focus on business outcomes: Building a great platform can take a long time. It's important to sequence your work so that business value is continuously realized. This motivates the team, builds credibility, and creates a virtuous cycle.
• Strive for transparency and communication: freely communicate decisions, progress and roadmap with stakeholders. In addition to clarifying the work at hand, also clarify what is not currently a priority. Write good documentation to encourage others to contribute and easily join the platform.
• Start Small: Even the best testing and quality assurance (QA) environments may miss issues that only become apparent after deployment to production. For major changes that will have a clear impact on customers, always start with a small group and then expand the scope of application after seeing it is effective in a small-scale production environment.
• Be radically transparent and over-communicate: Share decisions, progress and roadmap freely with stakeholders. In addition to clarifying what you are doing, also clarify what you are not currently prioritizing. Invest in documentation that makes it easy to contribute and join the platform.
• Start Small: Even the best testing and QA environments may miss some issues that aren’t discovered until production. For big changes that will have meaningful customer impact, always start with a small group of people and then gradually increase as you see things happening in small-scale production. If possible, use employees only for the initial population when changes affect external customers.
• Pay attention to proper management: Platform owners should pay attention to platform performance. All problems should be revealed through control mechanisms and automated alerts. Exceptions should be handled quickly. Priority should be given to root cause analysis and changes to prevent problems from recurring. If there are no issues, celebrate appropriately so the team knows it is appreciated.
• If it seems too good to be true...Exception monitoring is a great way to ensure execution is consistent with intent. The goal is often zero exceptions. For example, the delay should not exceed 200 milliseconds. If the exception report never shows any exceptions, there's probably something wrong with your monitoring. Always force an exception to ensure it fires correctly. I understand this very well.
• A happy team is a productive team. Celebrate achievements, praise team members when they perform well, and create an environment of inner fulfillment. Regularly measure your team’s happiness and give your team a chance to discuss what would make them happier and try it out on their own to address areas of dissatisfaction.

#When a team has a strong culture strongly supported by the right platform technology, the opportunities are endless. By combining cloud-native platforms with large-scale data, companies can better advance and experiment with newer, more innovative products and experiences. When these experiences enable end users and customers to get the product or service they need, when they need it, it makes a huge difference.

Original link: https://venturebeat.com/ai/best-practices-for-building-machine-learning-platforms-on-the-cloud/

The above is the detailed content of Several excellent practices for building machine learning platforms on the cloud. For more information, please follow other related articles on the PHP Chinese website!