Deep Learning: Machine Learning at its Finest

What you need to know about the evolution of artificial intelligence

woman with computer code projected onto face
 The Image Bank / Getty Images

Deep learning is a powerful form of machine learning (ML) that builds complex mathematical structures called neural networks using vast quantities of data (information).

Deep Learning Definition

Deep learning is a way of implementing ML using multiple layers of neural networks to process more complex types of data. Sometimes called hierarchical learning, deep learning uses different types of neural networks to learn features (also called representations) and find them in large sets of raw, unlabeled data (unstructured data).

One of the first breakthrough demonstrations of deep learning was a program that successfully picked images of cats out of sets of YouTube videos.

Deep Learning Examples in Daily Life

Deep learning is not only used in image recognition, but also language translation, fraud detection, and to analyze data collected by companies about their customers. For example, Netflix uses deep learning to analyze your viewing habits and predict which shows and films you prefer to watch. That’s how Netflix knows to put action films and nature documentaries in your suggestion queue. Amazon uses deep learning to analyze your recent purchases and items you’ve recently searched for to create suggestions for the new country music albums you’re likely to be interested in and that you’re in the market for a pair of gray and yellow tennis shoes. As deep learning provides more and more insight from unstructured and raw data, corporations can better anticipate the needs of their customers while you, the individual customer get more personalized customer service.

Artificial Neural Networks and Deep Learning

To make deep learning easier to understand, let’s revisit our comparison of an artificial neural network (ANN). For deep learning, imagine our 15-story office building occupies a city block with five other office buildings. There are three buildings on each side of the street.

Our building is building A and shares the same side of the street as buildings B and C. Across the street from building A is building 1, and across from building B is building 2, and so on. Each building has a different number of floors, is made from different materials and has a different architectural style from the others. However, each building is still arranged in separate floors (layers) of offices (nodes)–so each building is a unique ANN.

Imagine that a digital package arrives at building A, containing lots of different kinds of information from multiple sources such as text-based data, video streams, audio streams, telephone calls, radio waves and photographs–however, it arrives in one big jumble and is not labeled or sorted in any logical way (unstructured data). The information is sent through each floor in order from 1st through 15th for processing. After the information jumble reaches the 15th floor (output), it is sent to the 1st floor (input) of building 3 along with the final processing result from building A. Building 3 learns from and incorporates the result sent by building A and then processes the information jumble through each floor in the same manner. When the information reaches the top floor of building 3, it is sent from there with that building’s results to building 1.

Building 1 learns from and incorporates the results from building 3 before processing it floor-by-floor. Building 1 passes the information and results in the same way to building C, which processes and sends to building 2, which processes and sends to building B.

Each ANN (building) in our example searches for a different feature in the unstructured data (jumble of information) and passes the results to the next building. The next building incorporates (learns) the output (results) from the previous one. As the data is processed by each ANN (building), it gets organized and labeled (classified) by a particular feature so that when the data reaches the final output (top floor) of the last ANN (building), it is classified and labeled (more structured).

Artificial Intelligence, Machine Learning, and Deep Learning

How does deep learning fit into the overall picture of artificial intelligence (AI) and ML? Deep learning boosts the power of ML and increases the range of tasks AI is capable of performing. Because deep learning relies on the use of neural nets and recognizing features within data sets instead of simpler task-specific algorithms, it can find and use details from unstructured (raw) data without the need for a programmer to manually label it first–a time-consuming task that can introduce errors. Deep learning is helping computers get better and better at using data to help both corporations and individuals.

Was this page helpful?