Topics: AI | Data Science | Data Visualization | Deep Learning | Machine Learning | NLP | Python | R | Statistics

KDnuggets Home » News » 2021 » Jul » Tutorials, Overviews » Pushing No-Code Machine Learning to the Edge ( 21:n27 )

Pushing No-Code Machine Learning to the Edge


Discover the power of no-code machine learning, and what it can accomplish when pushed to edge devices.



Figure
Photo by Markus Winkler on Unsplash
 

Pushing No-Code ML to the Edge

No-code or low-code platforms have gained a lot of traction in recent years as those with limited programming and development experience become interested in related technologies. These platforms allow the inexperienced to do things like build mobile apps, deploy web applications on websites and within work environments, or even leverage AI and machine learning.

Where traditionally one would need a strong background in various development and programming disciplines to really take advantage of ML, that’s no longer the case. Of course, deployments can be much more sophisticated for those who do have the proper background, but TinyML and no-code ML solutions are making these technologies more accessible to everyone.

Realistically, it’s another form of no-code AI, where the goal isn’t just to “democratize AI,” but also to greatly expand its capabilities in a business setting. No-code ML or autoML are not as popular as their AI brethren, and it’s definitely an emerging field. TinyML, although slightly different, offers the same benefits because the logic is embedded within the device itself.

 

Pushing Computations to the Edge

These technologies, even in low-code form, require constant and massive streams of data, and they also ingest and process that data in real-time. AI and ML solutions can process information exponentially faster than the human mind. As such, they already improve efficiency and output. But they’re also limited by today’s networks, which can experience bandwidth and service interruptions, bottlenecks, and any number of delays or complications.

It stands to reason that unlocking these technologies by providing the active connections they require can enhance the performance benefits even more. That is precisely where edge computing comes into the picture.

Edge computing involves taking the computation and processing solutions and moving them closer to the source, or incredibly close to where that processing needs to happen. It bridges the gap between complex computations that are carried out on the actual device versus those carried out remotely on an external server or system.

Handling events locally means equipping the device(s) with enough power to do the work, which is not always possible, convenient, or cost-effective. Comparatively, handling it remotely affords much more power but often at the cost of speed and performance. Edge computing solutions offer the best of both worlds, with improvements to scalability, reliability, and customization.

It’s also possible to optimize where that “edge” truly is, to reduce latency, lower operating and service costs, increase visibility, accelerate deployments, and more. One IoT startup was able to utilize the AWS Compute Optimizer, with the assistance of a consultant, to implement a compute savings plan — reducing its overall bill by 37.5% per month. It highlights the potential flexibility and dynamic nature of cloud and edge computing solutions.

 

What’s the Relation to No-Code ML?

Again, no-code ML aims to democratize and improve accessibility for the related technologies. That means more and more people will have access to machine learning, AI, and neural network solutions, and both dependability and reliability will grow that much more imperative.

The problem is that most people opting to develop with no-code or low-code solutions generally do not have the power or infrastructure in place to support machine learning platforms on a macro level. Imagine a small startup launches a mobile app, and that app uses machine learning to analyze a user’s movie and TV preferences to provide recommendations. Neat idea, right? Solutions like this are already being explored and are already in use.

Initially, processing power won’t be a problem. But once the install base grows to hundreds of thousands or even millions of different users and accounts, it’s going to take a ton of computational power to support the platform.

At some point, the developers will need to scale up their infrastructure to handle larger quantities of work, without affecting performance and reliability. Edge computing solutions are a more manageable way to do that, from a cost and performance perspective. They’re also much more accessible, which is one of the major benefits of cloud technologies as a whole.

Edge Computing for No-Code Applications

Another element to the equation is how no-code ML applications are supported. Pushing the no-code platforms themselves to the edge of the network comes with the same benefits in efficiency, performance, and cost.

When you’re working with a development platform, such as an IDE, you want near-instant feedback. If the web applications are taking full advantage of edge processing, that improves the entire experience and also makes quick feedback possible.

TinyML, on the other hand, doesn’t need edge computing or internet-enabled hardware. Everything happens locally, on the device or a microcontroller. It has become the go-to practice for the industrial internet of things (IIoT), where embedded computing offers unparalleled performance and speed.

 

Edge Computing vs. Cloud Computing

It’s important to understand that edge computing and cloud computing are not interchangeable and are not meant to replace one another. They are for distinct purposes. Edge computing is often the solution of choice for time-sensitive data processing and delivery, whereas cloud computing is for data that has no time constraints.

Edge computing is best for localized situations, where open connections may be spotty and the bulk of processing power must be done fast and close to the device(s) or system. It’s also an excellent solution when the technology hinges on speed and reliability, like with medical and health care devices. For all other scenarios, cloud computing likely applies.

So, pushing no-code ML platforms and computational efforts to the edge does not technically mean pushing them into the cloud. There are similarities, but they are not the same.

 

On the Edge of the Future

Edge computing is a viable and beneficial form of data processing that is extremely relevant to more traditional machine learning applications as well as no-code or low-code ML deployments. It moves the power closer to the end-user, or the output of the system, to improve performance and reliability and offer a much more scalable solution that can handle the requirements of an ML-driven application.

Bio: Devin Partida is a big data and technology writer, as well as the Editor-in-Chief of ReHack.com.

Related:


Sign Up

By subscribing you accept KDnuggets Privacy Policy