Unpacking the nuts and bolts of machine learning

With the increasing adoption of machine learning (ML) across industries to solve complex business problems, tasks which previously required a great deal of resources and effort are now undertaken with unprecedented ease and cost efficiency.

However, as with any other technology, some enterprises are still on the fence about ML. One notable cause for hesitancy is that a lot of companies don’t see the big picture yet on how ML will create value for them.

According to Kumar Chellapilla, General Manager of Human-in-loop ML/AI Services at Amazon Web Services (AWS), incorporating ML will actually cut companies’ costs on talent acquisition and training, and will especially be beneficial for those engaged in multiple industries.

“In the past, you had to have subject-matter expertise. And you needed people who could code those domain expertise into a solution and operate. But now, machine learning is actually making it easy for many people to enter new industries, where as long as you’re able to do a principled scientific approach to collecting data, and bringing the best of machine learning tools and technologies, you’re now able to enter new industries,” Chellapilla said during a media conference hosted by AWS.

In order for ML to truly scale, Chellapilla added, the tech industry as a whole needs to make the adoption and incorporation of ML into businesses easier, regardless of people’s knowledge and skill level.

“One key way we do this is by providing training resources to help our customers upscale and learn best practices for machine learning. We offer over 80 free digital training courses that cover a variety of AWS ML services and solutions. The content ranges from foundational level topics such as ‘Machine Learning Essentials for Business and Technical Decision Makers’ all the way to advanced level topics, such as ‘Developing Machine Learning Applications,’” he explained.

Why ML?

To illustrate the benefits of ML, several AWS clients shared industry-specific uses of the technology.

For Kunal Prasad, co-founder and COO of India-based agrotech company CropIn, global problems such as hunger could be solved using the wonders of ML.

“Basically we look at these global challenges in agriculture and try to solve them with ways and means possible, so that it can be effectively used by both our customers, which is the farmer, as well as the industries. Now when we come to the farmers, the needs of the farmers are to have access to advisories, which are much more prescriptive and predictive because of the climate change issues,” he explained.

“We have worked in the last 11 years, impacting the lives of close to 7 million farmers, to digitise entire farm records covering 16 million acres. And in doing so, we have worked with more than 250 different enterprises, government institutions, as well as the development sectors in 56-plus countries covering almost 400 crops in 10,000- plus varieties,” Prasad added.

Meanwhile, James Mylroie-Smith, Vice President of Data Science at Singapore-based retail market intelligence platform Omnilytics, acknowledged the role of ML in helping retailers amid government lockdowns during the past two years.

“I think, for retail, it’s been a very challenging year with the pandemic. And I think with the closures of physical stores, it really meant that retailers have to adapt and be agile,” he said.

“Our goal is really to get data into the hands of retailers both large and small. And we do this by collecting data on over 75,000 brands globally, and have a database now with over 100 million products. To get value from this data, we use ML to classify the products to a high granularity. So this enables our clients to make big business decisions at a granular level,” Smith noted.

ML, Smith added, would greatly help with retailers’ internal processes.

“They (retailers) can reduce waste, from better buying and better selection of their products, as well as optimising the price points of the products they’re offering. So that should mean that they’re able to actually sell the products for the price that they want,” he said.

As for Mark Judd, the Executive Manager of Australian energy firm AusNet, ML is instrumental in avoiding damages to their power lines.

“We believe that machine learning will help us better manage our assets in terms of recognising defects. If you want to get to a little bit of detail, (it’s) faster and more reliable, but also (effective in) managing vegetation across our networks. So we’re able to use modern technologies very efficiently, and accurately monitor vegetation and assess vegetation growth in our networks, and as we know that vegetation can knock down the lines, (and) can also light fires, Judd pointed out.

“Ultimately, if we can operate our networks more cheaply, that makes it cheaper for our customers,” he added.

Creating ML culture

In order to integrate ML into enterprise systems more efficiently, the experts agreed that laying the groundwork requires not only logistical preparations, but a positive mindset as well.

“I think generally, the culture needs to be open first to the benefits of ML. And it’s key to have buy-in across teams, and also in leadership. And you then need the data scientists or ML practitioners to really be focusing on the business value and try to solve some of the business problems,” Smith said.

“Rather than ‘should we be using machine learning,’ it was ‘what’s the benefit we want to get? And what do we need to get there?’ And because of that, (in the) very early versions of our dashboard, we’re already integrating machine learning in the background,” he added.

For Judd, internal trust in the technology has proved invaluable to the performance of their field teams.

“We’re using machine learning now for recognising asset defects in the imagery. For example, a broken cross arm on a pole. The way we’ve deployed that is (that) the machine is assisting our asset inspectors to identify the defect. So the machine processes the data, presents it to our asset inspection, and says, ‘I think there’s a defect here.’ Conversely, the asset inspector is identifying, ‘Yes, that is a defect,’ or ‘No, it’s not.’ So really the track that we’re building is the asset inspectors trusting the process, but the byproduct is that the asset inspectors, by that very nature, are really feeding the machine learning models to make it better,” Judd explained.

“So you can see we’ve got this lovely relationship with our asset inspectors, where it’s building trust, and they’re building the trust themselves. And effectively, that builds a trust across the greater business,” he said.

Industry change

While ML is doing wonders for their respective companies, the experts also predict a sweeping impact across industries, especially as more and more enterprises are scaling up their digital transformation. 

CropIn, for example, has already collected three trillion data sets over the course of several years, which it then uses to predict variables such as pests, diseases, and crop yield. 

“We are building early warning systems for pests and disease-based models, right? So I can inform the farmer seven days in advance that ‘This is the disease that is going to come back to your crop. Can you take this little advisory right now?’ So both at the farmer level, when we are trying to increase the per acre value, we are seeing the impact of data and the models that we are putting on our system, as well as for the stakeholders which are coming and engaging with us,” CropIn’s Kunal Prasad noted.

Aside from providing farmers with insight, Prasad said that ML also simplified loan applications for agri needs. 

“Eventually, what is also happening is that the banks had a problem of reaching out to the farmer because the process was very mundane, like it took almost a month for them to reach to a farmer and process a loan. But now, if it is already coming in a pre-processed manner, we are able to provide the loans to almost 60% of the population with a click of a button,” he said.

“I think that’s the pace at which the industry is transforming. And even the incumbents like now are facing that challenge from startups and new innovators, because they’re transforming it in a very, very different way. And I think that’s how the world is going to transform, an opportunity for all of the industry players as well as the newcomers to use the data and technology to change the way that we have done business over here,” Prasad emphasised.

For AusNet’s Mark Judd, ML is one of the key parts in lowering the cost of energy.

“I think what’s going to happen with machine learning is it’s going to be a key part of operating and modeling the operation of our networks, which will allow us to transmit and distribute significantly more energy in a two-way operation. So we can absorb energy from consumers, and solar farms, and we can manage the network more powerfully. And then from an industry perspective, that allows you to confidently offer the great economy of a very reliable power network and a power system,” he said.

“I think there’s never been a better time to start using ML. And I think the tools and services around mean that that journey is getting easier and cheaper to give you business value much faster now,” Omnilytics’ James Mylroie-Smith concluded.