Machine learning is something new – a technology that builds itself. It is an engine using data, not programmers, to more quickly write programs, and it has more predictive power than ever. Machine learning is turning data into one of the most precious commodities on the planet, expanding the range of what is possible with a variety of information. Companies now face an exciting challenge: what questions can, and what questions should, be asked of their data?
The IoT (which includes the IIoT) is a network of physical devices that uses sensors, software, and connectivity to exchange data between devices (sometimes called “the edge” in IIoT) and the Internet. This ecosystem can be used to solve a range of problems, from something as simple as finding your lost keys using the device Tile, to something as complex as transforming supply chains like Amazon did by using connected robots to fulfill orders and manage stock. The IoT can even help us rewire our own brains: by measuring brain waves, the device Muse is meant to improve the effectiveness of meditation.
Machine learning doesn’t care about reasons why things might be related. Instead, machine learning is about connecting inputs to an output. In the past, if you wanted to predict the price of a new house, you would take inputs – information about the house – and a model written by a programmer and combine them to get your output: the value of that house. With machine learning, you give the computer the inputs and outputs, in the form of data, and it returns the model.
Many people confuse ‘machine learning’ with ‘artificial intelligence’ (AI). AI is a quest, one that has existed for over 70 years, and over the years interest in the AI mission has come and gone. But today, something is different. Machine learning now provides a set of tools that are accelerating the development of intelligent systems. By rapidly utilizing data, machine learning can also combine connected devices, sensors and robotics to build systems that respond to a user or client.
Machine learning is no silver bullet; it’s important to know when machine learning can and cannot be used. Currently, machine learning can only be applied to very specific problems such as predicting good search results, recommending specific services or detecting particular types of fraud. Furthermore, machine learning only achieves good results with lots of manual data cleaning and input (feature) engineering. But this sensitivity to clean data may not last forever. Current research is applying machine learning to the task of automatic data cleaning in order to make applications that are more tolerant of unrefined raw data. This will have the effect of further broadening the types of problems machine learning can solve.
So how much data do you need to do machine learning? Well, it depends. Depending on what question(s) you’re asking, depending on what algorithm(s) you decide to use, and depending on what outcome(s) you’re looking for, you might need different amounts, types, and formats of data. Usually, more data are better for machine learning to best train the algorithm, but we recommend seeking out a data scientist to guide you in answering these questions since there’s no explicitly right or wrong way to use machine learning.
But if you have (a lot of) the right data and ask good questions, machine learning can enable businesses to turn data into predictions. And the ability to predict in actionable ways is powerful. That’s why the technology giants – companies like Google, Amazon, Baidu, Microsoft, IBM – are so interested in machine learning; it has the potential to drive business decision-making based on actionable data.
But you don’t have to work at one of those tech giants to take advantage of their advances in machine learning. The open-source movement is accelerating the pace of innovation in the field of machine learning through cutting-edge resources such as Google’s Tensorflow and Microsoft’s CNTK toolkits, as well as the open sharing of research and dataset by groups like OpenAI and DeepMind.
Although we know that machine learning will continue to impact how companies make decisions and predictions based on data, we don’t yet know whether machine learning is just for predicting numbers or if machine learning might finally give rise to general artificial intelligence.
In the end, modern machine learning is all about taking lots of data and making a prediction about future data. So what sorts of predictions does your business need to make?