The writer asks, how does one successfully implement a machine learning strategy? Photo: Pete Linforth: Pixabay

JOHANNESBURG – One Cloud argue that comparing cloud implementation with machine learning is like comparing an effective hiring strategy with a lead funnel, which are by no means directly related.

The supposition that cloud is better than machine learning assumes a lack of maturity for machine learning, a lack of data or absence of expertise for a complex machine learning strategy.

It’s important at this point to enunciate that priority is important. The fact that any business leader may want to leverage artificial intelligence (AI), or machine learning to avoid becoming an industry laggard, at least qualifies one to be in the late majority.

Luckily for many, machine learning implementation has been poor globally and perhaps one could squeeze into the early majority of successful implementers.

The prevailing question remains, how does one successfully implement a machine learning strategy?

The sure answer is to start with a cloud implementation strategy.

Why is this a better priority and a key first step towards successful machine learning implementation?

This question is a clue and any confusion surrounding the question is attributed to the unordered list of priorities that filter around in our heads while we try to dream and execute. We desire machine learning implementation yesterday because it promises automation and reduced overhead. It also promises the future, rather than being left behind in the past.

To achieve machine learning implementation objectives can be tiring, but it doesn’t have to be. 

The last decade has seen the rise of cloud giants, notably AWS (Amazon), Azure (Microsoft) and Google Cloud. These giants promise reduced cost of ownership and managed infrastructure in return for a little bit of trust and regular monthly payments. 

The true value that these hyper-scalers offer is the toolkit that accompanies their offering. There are two reasons why cloud implementation offers a reduced time to implementation machine learning strategy:

Platforms in the cloud help you collect data that is critical for machine learning success. 

Cloud providers offer machine learning tools that the average citizen can wield.

The application of cloud machine learning to business may not be immediately obvious. The following examples are used by select innovators today to undercut competition:

  • Prediction from images: AES Global uses vision detection to find visual damages to their wind turbine fleet using drone snapshots powered with vision prediction. The American Cancer Society use Google Cloud machine learning to fight breast cancer through clustering techniques.
  • Video intelligence: 20th Century Fox uses the Google Cloud Video Intelligence API and natural language processing to analyse the true DNA of movies to understand the actual target audience. The Greatest Showman movie is a great example of this. Fox adjusted marketing for the target audience from young females to females and those likely to watch Disney musicals, propelling the picture to a $434 million (R6.58 billion) haul at the box office. 

Osaka Prefecture University has developed the concept of a “plant factory” to address dwindling numbers of farmers. They currently use Azure Cloud to activity monitor plant growth through IoT (Internet of Things) sensors and video, streamed into the cloud. Notifications are sent to farmers when undesirable events occur. The system is now machine learning ready to use tools such as video intelligence to predict the onset of plant diseases and ill-nourishment.

  • Natural language: Google Cloud Translation API empowers Bloomberg to automatically provide news to customers in 40 languages in 170 countries. 
  • Ananda Development Thailand saw a problem with snag list processing when handing over a new condominium to the buyer. This process was tedious, with a lot of back-and-forth between buyers and inspectors. 

The Google Cloud Speech-to-text API helps inspectors capture defects on the fly via a mobile app, which automatically generates reports and contacts contractors responsible for repairs. 

The buyer views this process in real-time via the app. Within three months, Ananda development had saved $120 000. 

T Mobile’s customers like to have personal, human connection. Machine learning assists call centre agents by providing contextual information about customer issues. This information was first made available by labelling data that is used to train models. 

Amazon Ground Truth learns how to label that data after seeing some examples, making it easier to train models and provide quality customer support.

  • Auto machine learning: Lenovo partnered with Data Robot on AWS to allow the Lenovo data team to build automated machine learning models that can predict sell-out sales forecasts from a month in advance. BP is using automated machine learning on Azure to predict the recovery factor of potential oil and gas reservoirs. This helps them determine how much raw fuel can be extracted from these reservoirs. 

By choosing the right cloud provider and incorporating their technology toolkit into one’s business, an operational advantage is immediately present. Data collection then becomes easier, machine learning maturity grows and one can utilise democratised machine learning tools to increase the bottom-line.

Mark Mc Naught is a senior cloud architect at DotModus.

BUSINESS REPORT