How low-code platforms enable machine learning

摘要： Multiple options make machine learning available to professional data scientists, citizen data analysts, and software developers.

▲圖片來源：metamorworks / Getty Images

Low-code platforms improve the speed and quality of developing applications, integrations, and data visualizations. Instead of building forms and workflows in code, low-code platforms provide drag-and-drop interfaces to design screens, workflows, and data visualizations used in web and mobile applications. Low-code integration tools support data integrations, data prep, API orchestrations, and connections to common SaaS platforms. If you’re designing dashboards and reports, there are many low-code options to connect to data sources and create data visualizations.

If you can do it in code, there’s probably a low-code or no-code technology that can help accelerate the development process and simplify ongoing maintenance. Of course, you’ll have to evaluate whether platforms meet functional requirements, cost, compliance, and other factors, but low-code platforms offer options that live in the gray area between building yourself or buying a software-as-a-service (SaaS) solution.

But are low-code options just about developing applications, integrations, and visualizations better and faster? What about low-code platforms that accelerate and simplify using more advanced or emerging capabilities?

I searched and prototyped for low-code and no-code platforms that would enable technology teams to spike and experiment with machine learning capabilities. I focused mainly on low-code application development platforms and sought machine learning capabilities that enhanced the end-user experience.

Here are a few things I learned on this journey.

Platforms target different development personas

Are you a data scientist looking for low-code capabilities to try out new machine learning algorithms and support modelops faster and easier than coding in Python? Maybe you are a data engineer focusing on dataops and wanting to connect data to machine learning models while discovering and validating new data sources.

Data science and modelops platforms such as Alteryx, Dataiku, DataRobot, H20.ai, KNIME, RapidMiner, SageMaker, SAS, and many others aim to simplify and accelerate the work performed by data scientists and other data professionals. They have comprehensive machine learning capabilities, but they are more accessible to professionals with data science and data engineering skill sets.

Here’s what Rosaria Silipo, PhD, principal data scientist and head of evangelism at KNIME told me about low-code machine learning and AI platforms. “AI low-code platforms represent a valid alternative to classic AI script-based platforms. By removing the coding barrier, low-code solutions reduce the learning time required for the tool and leave more time available for experimenting with new ideas, paradigms, strategies, optimization, and data.”

There are multiple platform options, especially for software developers who want to leverage machine learning capabilities in applications and integrations:

• Public cloud tools such as GCP AutoML and Azure Machine Learning Designer help developers access machine learning capabilities.
• Low-code development platforms such as Google’s AppSheet, Microsoft’s Power Automate’s AI Builder, and OutSystems ML Builder expose machine learning capabilities.
• Low-code learning libraries like PyCaret target data scientists, citizen data scientists, and developers to help accelerate learning and implementing machine learning on open source toolkits.

These low-code examples target developers and data scientists with coding skills and help them accelerate experimenting with different machine learning algorithms. MLops platforms target developers, data scientists, and operations engineers. Effectively the devops for machine learning, MLops platforms aim to simplify managing machine learning model infrastructure, deployment, and ops management.

No-code machine learning for citizen analysts

An emerging group of no-code machine learning platforms is geared for business analysts. These platforms make it easy to upload or connect to cloud data sources and experiment with machine learning algorithms.

I spoke with Assaf Egozi, cofounder and CEO at Noogata, about why no-code machine learning platforms for business analysts can be game changers even for large enterprises with experienced data science teams. He told me, “Most data consumers within an organization simply do not have the required skills to develop algorithms from scratch or even to apply autoML tools effectively—and we shouldn’t expect them to. Rather, we should supply these data consumers—the citizen data analysts—with a simple way to integrate advanced analytics into their business processes.”

Andrew Clark, CTO and cofounder at Monitaur, agreed. “Making machine learning more approachable to businesses is exciting. There are not enough trained data scientists or engineers with expertise in the productization of models to meet business demand. Low-code platforms offer a bridge.”

Although low code democratizes and accelerates machine learning experimentation, it still requires disciplined practices, alignment to data governance policies, and reviews for bias. Clark added, “Companies must view low code as tools in their path to benefiting from AI/ML. They should not take shortcuts, considering the business visibility, control, and management of models required to make trusted decisions for the business.”

Low-code capabilities for software developers

Now let’s focus on the low-code platforms that provide machine learning capabilities to software developers. These platforms select the machine learning algorithms based on their programming models and the types of low-code capabilities they expose.

• Appian provides integrations with several Google APIs, including GCP Native Language, GCP Translation, GCP Vision, and Azure Language Understanding (LUIS).
• Creatio, a low-code platform for process management and customer relationship management (CRM), has several machine learning capabilities, including email text mining and a universal scoring model for leads, opportunities, and customers.
• Google AppSheet enables several text processing capabilities, including smart search, content classification, and sentiment analysis, while also providing trend predictions. Once you integrate a data source, such as Google Sheets, you can begin experimenting with the different models.
• The Mendix Marketplace has machine learning connectors to Azure Face API and Amazon Rekognition.
• Microsoft Power Automate AI Builder has capabilities tied to processing unstructured data, such as reading business cards and processing invoices and receipts. They utilize several algorithms, including key phase extraction, category classification, and entity extraction.
• OutSystems ML Builder has several capabilities likely to surface when developing end-user applications such as text classification, attribute prediction, anomaly detection, and image classification.
• Thinkwise AutoML is designed for classification and regression machine learning problems and can be used in scheduled process flows.
• Vantiq is a low-code, event-driven architecture platform that can drive real-time machine learning applications such as AI monitoring of factory workers and real-time translation for human-machine interfaces.

This is not a comprehensive list. One list of low-code and no-code machine learning platforms also names Create ML, MakeML, MonkeyLearn Studio, Obviously AI, Teachable Machine, and other options. Also, take a look at no-code machine learning platforms in 2021 and no-code machine learning platforms. The possibilities grow as more low-code platforms develop or partner for machine learning capabilities.

When to use machine learning capabilities in low-code platforms

Low-code platforms will continue to differentiate their feature sets, so I expect more will add machine learning capabilities needed for the user experiences they enable. That means more text and image processing to support workflows, trend analysis for portfolio management platforms, and clustering for CRM and marketing workflows.

But when it comes to large-scale supervised and unsupervised learning, deep learning, and modelops, using and integrating with a specialized data science and modelops platform is more likely needed. More low-code technology suppliers may partner to support integrations or provide on-ramps to enable machine learning capabilities on AWS, Azure, GCP, and other public clouds.

What will continue to be important is for low-code technologies to make it easier for developers to create and support applications, integrations, and visualizations. Now, raise the bar and expect more intelligent automation and machine learning capabilities, whether low-code platforms invest in their own AI capabilities or provide integrations with third-party data science platforms.

轉貼自Source： infoworld.com

若喜歡本文，請關注我們的臉書 Please Like our Facebook Page： 　　　Big Data In Finance