This article was submitted by our partner AWS.
Low-Code No-Code is a relatively recent trend in programming paradigms, and it’s all about developing application and flows without having to write a single line of code (NoCode) or by reducing the amount of code written to a minimum (LowCode).
Amazon Web Services (AWS) is the world’s most comprehensive and broadly adopted cloud platform, offering over 200 fully featured services from data centers globally. Millions of customers—including the fastest-growing startups, largest enterprises, and leading government agencies—are using AWS to lower costs, become more agile, and innovate faster. Today, more than 100,000 customers do Machine learning on AWS. And more and more of these customers want to enable more employees to do Machine Learning at scale, looking with interest at LCNC technologies in AWS portfolio.
Who is Davide Gallitelli?
Davide is a Specialist Solutions Architect AI/ML at Amazon Web Services. Born and raised in Bari, Italy, he has moved between a few European countries, including Spain, France and Belgium, where he currently lives. Automating IT tasks has always been his passion, since the early days of development through the years working in Southern Italy. Started learning about Machine Learning in university, where it struck him as the best tool for replacing most of the heuristics that come from programming thanks to statistical patterns in the dataset. Since those days, he still hasn’t let go of that passion, and he’s using it to assist customers throughout Benelux to solve challenges with Machine Learning and Cloud Computing.
Why Low-Code No-Code For Machine Learning
Low-Code No-Code paradigm has two main purposes: first of all and most importantly, it’s to empower any user regardless of their development experience to create their own applications, maintain them, and customize them as they see fit, without requiring development skills or an IT team supporting them; then, it is to accelerate the lifecycle of these applications, by removing the burden on limited constraints such as engineers or developers. This concept can be applied on anything, from analytics to application development, from BI (Business Intelligence) to ML (Machine Learning).
All of the above mentioned problems and challenges apply in the world of Machine Learning as well. ML experimentation is time-consuming and requires both domain expertise and coding skills. Data science teams are a scarce resource and often oversubscribed. According to a Gartner study, ML talent is in high demand, growing twice as fast with respect to other emerging job roles, with an annual compound growth of 74% in the past 4 years. In other words, this means that we have an increasing demand for ML skills, an increasing number of workloads that could be disrupted by ML, while the data science teams cannot grow at the same speed. Low-Code and No-Code Machine Learning enable true democratization of Machine Learning to virtually any user: they speed up data scientists thanks to pre-built solutions, drag-and-drop UIs, and simplified consumption of pre-trained models; they empower business analysts with tools that can help them past the steep learning curve of ML, and having them focus on inputs and outputs rather than the process.
Low-code ML helps data scientists be productive and effective
Enterprises are adopting Low-Code ML as tools that enable their data scientists to be more effective and more productive. The name of the game here is “faster time-to-market” and “strong operationalization”. Data Scientists want to increase the number of prototypes that they can work on, and reduce the time it takes them to get to a working prototype phase, by visually selecting some of the most common transformation they use in the data preparation phase, and visually building the flow that it would take them to train the model.
Additionally, very often Data Scientists test multiple well-known algorithms and combinations of them which have been proven very effective on a huge variety of problems. For reference, XGBoost has been one of the default go-to algorithms for many Data Scientists participating in Kaggle competitions, a well-known platform for learning data science through examples and competitions. So why not automating some of those ML models in a package, and use it as a perfectly functioning baseline for further developments?
Finally, it has been said over and over that “we shouldn’t reinvent the wheel”. This means leveraging existing solutions, or existing pre-trained models that are available. For instance, the Hugging Face model zoo is a well-known hub for natural language processing (NLP) models. Hubs can help with Low-Code ML, by reducing the time it takes Data Scientists to get to a fully functioning and performant model, without having to train one from scratch, and occasionally fine-tuning it to the specific task to be solved.
Empowering analysts and non-coders to do ML with no-code ML
Think about the tools that business analysts have today to introduce Machine Learning in their workflows. Most of them work on spreadsheets and/or BI tools. Some of them use macros and advanced capabilities of the platform of their choice. Few of them have the experience to open a Jupyter Notebook and start writing some code to build a linear regression or use a pre-trained PyTorch algorithm. However, all of them have to eventually try to understand why customers are churning, or predict the sales of the upcoming holiday season.
If we can mask away the complexity of ML with automation and visual UIs, then we can enable them to take the spreadsheet they’re working with everyday, and generate some smart business decisions – powered by ML. This way, they can reduce the time it takes them to create a ML-powered prototype, while focusing on the business metrics.
However, transparency for business analysts does not imply that everything has to be a black-box. A good No-Code ML tool must expose a certain level of explainability to analysts, so that they can always look into why the predictions where generated that way, track model evolution over time, and report on its usage. Furthermore, none of the automatically and visually generated artifacts should be lost. Business analysts should collaborate with data scientists, which could eventually look into the model and insights generated, in order to include them in existing applications or automate them as needed.
AWS Customers are adopting Low-Code No-Code ML at an increasingly fast pace
One example of an industry where we see a lot of usage is manufacturing; and supply chain. With what has happened in the world most recently, there are many challenges in the supply chain area – so being able to forecast demand is very important. Customers ask us: “How can you help us to anticipate changes, to anticipate demand, to save cost to make our customers happy and deliver on time?”. Customers are increasingly asking us to empower even more people to generate insights from data, and speed up those who already can. That’s why we have introduced Low-Code and No-Code ML capabilities in Amazon SageMaker. The first offering is Amazon SageMaker Canvas, which expands access to ML by providing business analysts with a visual point-and-click interface that allows them to generate accurate ML predictions on their own — without requiring any ML experience or having to write a single line of code. It also has built-in capabilities for sharing models and datasets with data scientists so they can validate and further refine ML models. In addition, data scientists can also find low-code ML capabilities, assisting them throughout the ML lifecycle: Amazon SageMaker Data Wrangler for data preparation, Amazon SageMaker Autopilot for AutoML, and Amazon SageMaker JumpStart for pre-built solutions and built-in algorithms with pre-trained models.
How can you learn more about LCNC ML on AWS?
In order to get started with Amazon SageMaker low-code/no-code tools, please explore the links mentioned above, and play around with the different workshops that we have available. We have also recently launched a free Coursera course, “Practical decision making on AWS with no-code ML”. In addition, you can review the AWS Machine Learning blog, where our solutions architects and some of our customers publicly speak about new features, architectures, and success stories.
Do you prefer in-person hands-on workshops? Please, Join us on December 13 at the AI Innovation Center in Eindhoven to participate in an Amazon SageMaker Canvas hands-on lab during our Improving Supply Chain with Machine Learning event. We’re looking forward to help you accelerate and democratize Machine Learning with Low-Code No-Code ML!