The many layers of intelligence
If you have not seen one of the many Venn diagrams, those images of circles enveloping other circles, describing the relationship between the different types of artificial intelligence, I will attempt to paint an image for you here without those spherical objects. Machine learning (ML) is a subset of AI. It enables a system to learn from data rather than through explicit programming. Deep learning (DL) is a subset of ML. It is more complex in its functionalities and its architecture.
It is important to remember that these definitions are not interchangeable but are their own forms of computer demonstrated intelligence. So, DL is a form of ML, which is a form of AI. Simple, right? Well, within each of these categories of AI there are also many subtypes within the subtypes. However, for the sake of simplicity, here we will only focus on machine learning and deep learning as wider entities and discuss the differences between the two without delving deeper into the mechanics of their subtypes.
Architectures of learning
ML was created with the objective of computers or their processes to learn and improve this learning in an autonomous fashion. To fulfil this aim algorithms are built to interpret data through statistical techniques and apply this information in order to make knowledgeable decisions or predictions. Rather than coding rules with a specific set of instructions, ML learns from the data it is given. Perhaps a recognizable example of ML in action is Uber’s price surging. The engineering lead at the company recently explained that machine learning is behind this; predicting rider demand and therefore tacking on that added cost at popular times, such as that 1am trip home from the bar(s).
Machine learning can be formed of many different types of models including neural networks, the arrangement of three or four layers of connected computational units. Deep learning is the next generation of ML. It is uniform in its architecture always relying on a form of neural network. What brings the deep to deep learning is the addition of more layers: from tens to hundreds. These are what have allowed for huge improvements in AI and for us to solve problems we could not a decade ago. What also distinguishes DL from its predecessor is the way it learns. ML requires some hand holding, and by this I mean it needs hard coded features to learn from. DL is more autonomous, constructing an idea of the features automatically.
Image analysis powered by deep learning
One of the most powerful applications of DL is image recognition and analysis. In fact, it is already surpassing human capability in these tasks. When we humans see something our brain recognizes distinct patterns to infer what it is that we are seeing. Neural networks see images as a grid of numbers represented by the pixels in an image. In order to use DL for image recognition the networks must first be trained, with for example images of a specific type of tumor. They then develop an idea of what an image of that tumor contains and learn to recognize it.
By providing more accuracy deep learning has allowed for the field of artificial intelligence to advance even further. It is enabling more complex problems to be solved and for more intricate and larger data sets to be analyzed. Deep learning is setting new records and high standards in fields such as image recognition; a major technological advancement in healthcare. The recognition and discovery of patterns lies at the heart of scientific progress and deep learning AI is by far the best tool for this. So why not upgrade your work?
With Aiforia, our AI just needs to know what you want to find.