Press Clipping
06/21/2018
Article
What automated machine learning brings to the IoT | Q&A Pedro Alves, CEO Ople

The trouble with big data is that there’s just so much of it. Because of that, mining big data for nuggets of gold is the real prize.

Internet of Business spoke to Pedro Alves, founder and CEO of Ople, about why he believes that automated machine learning (AutoML) should lie at the core of this process of extracting real business value.

Internet of Business: What, in a nutshell, is automated machine learning?

Pedro Alves: “Machine Learning is one of many tools that a data scientist can use to have machines – computers – learn from data. What this means is that instead of having a human codify a set of instructions for a machine to act on, machine learning algorithms can create their own rules on how to model data, and the appropriate responses it should exhibit.

“The tasks that a data scientist faces when building machine learning models include choosing algorithms and tuning them, among other things. Automated machine learning is simply a script, or a set of scripts, that runs through a predetermined set of tasks that the data scientist would otherwise have to perform manually.

“The real benefit of AutoML is that it reduces the time needed for a data scientist to run though experiments in the initial assessment of methods that one might use to build a model.

“The downside is that because it is just a script calling a defined collection of machine learning algorithms – and for the more technically minded, it’s worth saying that I’m including hyper-parameter optimisation here – it has no intelligence. AutoML does not evolve as a data scientist would.”

So what does AutoML in its current form bring to our ability to better understand big data?

“The goal of AutoML is to enable data scientists to be more efficient with their use of time and run through projects more quickly. And given the fact that they might be working with big data, this will indirectly increase our understanding of that data.

“The current status of AutoML is a great step forward – more people can build quality models – but it is far from achieving the goal of greater intelligence. AutoML lacks the intuition and learning that delivers real breakthroughs in understanding.”

Can AutoML help with working out which data points it might be useful to have, as well as with analysis?

“The short answer is: No.

“Machine learning, in general, is fantastic at determining relationships between data points that humans overlook. AutoML makes it easier to develop these models and get better predictions, but it does not have the intelligence to help figure out which data points to look at and analyse.”

Where is the crossover between AutoML and the Internet of Things?

“The IoT is generating more data than any technology before it. Extracting the signal from the noise and embedding intelligence directly into the device is critical for the IoT to truly become mainstream.

“AutoML is playing an enabling role on the labs and experimentation side for IoT companies, as they try to find ways they can work with the massive amounts of data they are generating.

“However, the unique needs of the IoT – extremely lightweight models that can be built into embedded devices, the ability to natively work with sparse and unclean data, and just the pure volume of data available – are all beyond the current state of capabilities for AutoML.

“IoT companies should look for solutions that incorporate intelligence into the model development process and can deliver lightweight models as their output.”

In a future scenario where AutoML opens up access to learning that was once the province of only the wealthiest companies, what might that opening up mean for SMEs, and for the whole commercial landscape in general?

“We applaud AutoML’s role in expanding access to AI and machine learning. Artificial intelligence and machine learning can dramatically impact all businesses large and small – once they can get access to it. There is a shortage of data scientists, and SMEs and startups do not have access to those specialties today. Intelligent AutoML solutions can help bridge that gap.

“However, solutions for the SME need to go beyond AutoML’s brute-force approach and become smarter.”

Internet of Business says
Alves’ comments about some aspects of AI and machine learning lacking intelligence and the ability to evolve are useful in an environment in which so much store is being set by organisations about the current ability of these technologies to help their businesses.

AI and ML are on a long, evolutionary journey towards greater intelligence and utility, and it is worth adopting that perspective as a user.