Deep learning vs. machine learning
Valérie Bécaert Valérie Bécaert
August 23 5 min

Deep learning vs. machine learning

There are many different technologies that fall under the broad category of artificial intelligence. There are types of AI, and within these there are also subtypes – different variations with sometimes very dramatic distinctions. Deep learning vs. machine learning is one such example.

Let’s explore how these two technologies differ.

Technology can be trained

Machine learning originated from the study of "pattern recognition” – developing the capabilities of a machine to recognize patterns in data – but the technology has expanded far beyond those boundaries since then. Today, it’s concerned with how artificial intelligence can learn to perform a task by analyzing task-related data. It constantly learns how to better interpret this data, so it can become better at performing the task.

This sense of “a machine with a job to do” is an important idea to grasp. Accomplishing a task, and striving to do so more effectively, is the foundation of all human technology – since we first made fire, started using stone tools and invented the wheel. And artificial intelligence is just the latest expression of this.

When thinking about machine learning, and AI as a whole, it’s useful to consider the Slavic root of the word “robot”: robotnik. It’s a term for “worker”. An artificial intelligence functions chiefly as a worker, with a specific task or set of tasks to do and the imperative to do them as well as possible. Much of AI development today is concerned with how a machine can learn to do its job and keep learning how to do it better.

Machine learning in action

A simple example of machine learning in action would be tasking a machine to sort images into two categories: for instance, photographs of the sea and others of mountains. The machine learning algorithm needs to start with structured data.

In this case, pictures that are already labelled “sea” or “mountains” so it can begin to tell them apart. Once it trains itself using this structured data, it can then continue to sort the rest of the images without needing them to be pre-labelled. It has “learned” to differentiate them.

A higher level of reasoning

Deep learning is essentially a highly sophisticated application of machine learning. It automatically learns to extract a useful representation of inputs (e.g., image -> a few characteristics of it), in order to draw its own conclusions without human guidance/training. Whereas, in regular ML, one needs to manually specify how this representation is extracted.

That’s because it doesn’t just use one algorithm to make a decision like simpler machine learning does. It uses an artificial neural network – a kind of structure inspired by the human brain, although currently not nearly as complex – which essentially consists of many components that deal with the data differently and send it between each other, before coming to a decision.

In other words, deep learning models are composed of several layers of small models (called neurons) that progressively transform the data into a representation that allows for an accurate decision.

How deep learning tackles tasks

In the example of differentiating between pictures of the sea vs. mountains, the neural net would examine various specific aspects of the images. Different parts of the network would analyze each photograph, apply their own criteria for judging its content, reach their own conclusions, compare these conclusions (each factoring the others’ “thoughts” into its reasoning) and finally make a judgement call; sea or mountain.

And, of course, because this is a form of machine learning, the neural network remembers the insights it gained from each decision-making process and applies them when making future evaluations. It gets smarter all the time.

Thinking like we do

This is similar to how human reasoning works, and how our brains analyze sensory data. We experience something (see it, hear it, etc.) and this is translated into electrical impulses passed around the network of neurons in our brain, each performing a stage of our analysis.

It’s not an exact replica of a human brain, but the idea of the artificial neural network is a useful way to introduce the concept of deep, hierarchical machine learning to those unfamiliar with AI. And it’s the closest we’ve come to approximating the processes of complex biological thought.

Smarter learning machines

There is a popular saying these days: “every company is a tech company”. This reflects how important technology like deep learning has become to all kinds of enterprises. In the insurance sector for instance, machine learning can perform a range of valuable functions, from providing insights into customer needs, to finding the right pricing levels and detecting fraud.

Meanwhile, in the financial world, future market trends can be predicted with greater accuracy than ever. In personal banking, AI-driven chatbots can now service customers’ requests with more sophistication, reducing the burden on the helpdesk and increasing customer satisfaction.

The future of machine learning

Deep learning brings a new level of advancement to the field of artificial intelligence. Whereas a simple learning machine’s methods ultimately come down to “monkey see, monkey do”, the more complex reasoning of deep learning is a step closer to the ultimate aim of AI: machines with a thinking capacity equal or greater to that of a human. With the leaps ahead that deep learning constitutes, we're getting closer. But it's not right around the corner: we still have a lot to understand and discover before silicon-based machines really surpass the carbon-based ones.