Deep learning (DL) is a subset of machine learning. It allows artificial intelligence (AI) to use neural networks to predict outcomes based on data analysis. It essentially allows the AI to mimic how the human brain works to “think” intuitively based on what it already knows.
Think of how a child learns to speak. Every day they are hearing new words and sounds. Their brain begins to file those pieces of information based on what the child understands from the context of the language. That is how AI deep learns. The AI scans massive quantities of data rapidly to predict an outcome based on previous occurrences.
Deep Learning vs Machine Learning
Machine learning is when a computer can learn from data using algorithms to complete a task. They do not need to be explicitly programmed, but they do need time to train with existing data sets in order to produce the desired action or response when they are given a certain intent.
For example, a chatbot on a banking website will be trained using language snippets based on intents that seek to elicit banking-related responses. If the user types in “see my account,” the bot will be trained to respond with something like, “do you want to see your checking or savings account?”
Deep learning takes this concept a step further. Deep learning means the computer uses a structure vaguely inspired by the synapses in the human brain. This allows the system to process unstructured data and learn how to make decisions after training on a large dataset. For example, if elections are coming up in town, a deep learning algorithm could scan news headlines and, based on the sentiment it finds in the text, predict who is going to be leading in the polls. And that’s just one example of the many ways deep learning is mixed in with our everyday lives.
How Does Deep Learning Work?
Deep learning works using layers of processing units that allow for the extraction and transformation of numerous variables. Each network in the hierarchy transforms pieces of data and passes it on to the next layer to build a static model with iterations of output. The number of layers varies based on the level of precision and learning the output sought requires.
It sounds complicated, so let’s add a visual:
(Image from Vitalflux)
Deep learning depends on networks that mimic the neural networks in our human brains. Deep learning depends on networks that mimic the neural networks in our human brains. It allows the network to transform the data into what data scientists call different representations, which highlights salient information like the edges of objects in images or particularly relevant words in a text. This progressively abstracts away the raw data, reducing an image with millions of pixels to what might be a list of only a few dozen numbers, each representing the “cat-ness”, “dog-ness” or other properties we care about.
This process is called dimensionality reduction or feature extraction. Once found, classical analytical methods like linear regressions or k-means clustering, which would struggle with the original input because of its size, can be then used to make a final decision.
Applications of Deep Learning (AKA How To Use It In Real Life)
So let’s get out of the technical and into the practical. Where do we see deep learning in our lives? Pretty much everywhere. Here are just a few of the many examples. You’ll definitely be able to think of more!
One of the first and most commonly referenced uses of DL are smart virtual assistants, like Alexa or Siri. Online service providers use DL to respond to users' needs through speaking. Many companies use these virtual assistant tools to get to “know” their employees’ needs and provide them with useful information as needed throughout the day.
Virtual assistants can be either voice activated, or chatbots, or both. They can serve as excellent training tools, teaching new employees the ins and outs of the job on the fly as they have questions about policies or procedures that are documented within the company.
Virtual assistants are also great just to have around the house! Playing trivia or watching Jeopardy? Ask Alexa! In an argument about a factoid? Your VA can provide the correct answer. Deep learning powered VAs can manage your schedule, take notes, keep inventory, and generally help you manage your life with minimal thought. There are so many ways to take advantage of their help.
Deep learning is revolutionizing healthcare and helping medical professionals make diagnoses and treat illnesses based on data that may not be apparent to the human eye. While most deep learning in healthcare is still at the experimental stage, its potential use is certainly growing, which means you are in time to become an expert.
Deep learning is used in healthcare for imaging analytics and diagnostics. The neural network model works really well for analyzing MRI results or x-rays since it is specifically well-suited to analyze and extract data from images. Deep learning models can detect anomalies on a medical scan much more quickly and accurately than the average nurse practitioner.
Of course, deep learning affects all of us when we are consuming entertainment. If you watch series or movies on an online platform, you are experiencing deep learning models in action.
Have you ever stopped to consider how your streaming service knows what shows or movies to suggest to you? They are able to match your tastes, and even provide suggestions for different moods or genres that you tend to shift between. This is a perfect example of how deep learning algorithms observe massive quantities of data and filter it to understand the complex layers of your preferences. The algorithm then learns from your previous choices and uses that information to sort through new additions to the library to suggest things you will like.
The same goes for music. If you’ve ever signed in to Youtube and seen music video suggestions that you know you’ll like, it’s because their deep learning algorithm keeps track of music you’ve listened to before. Not to alarm you, but these suggestions can also be provoked from music you heard outside of Youtube if you have a microphone enabled on your device. Those virtual assistants are always listening, and deep learning allows them to process sound bytes even if you aren’t actively engaged with a device or platform.
The future of deep learning and video games is absolutely fascinating. Remember those choose your own adventure books? Well now, role playing games are using deep learning to allow you to make choices within the game, and the game will change to predict how you are going to act based on your previous choices. The deep learning agents within the game can actually compete, even on the professional level, in complex strategy games. The game is always learning, and consequently transforming.
This phenomenon began with machine learning and chess, when, in 1997 we witnessed the historic match between IBM supercomputer Deep Blue against champion Garry Kasparov. Spoiler alert: Deep Blue won. Since then, AI and gaming have gone hand in hand to create increasingly challenging problems for humans to solve in the name of fun!
According to Google, shifting Google Translate to deep learning led to a 60% increase in translation accuracy. Before the change to deep learning, Google relied on phrase-based machine learning. But with deep learning, their tool is constantly improving its language knowledge and translation accuracy as more and more people use it.
Also known as neural machine translation (NMT), AI models that use deep learning for language understanding and translation deliver more accurate and natural sounding translations. In most of these models, each word in the input sentence is encoded as a number to be translated by the neural network into a new sequence of numbers that coincide with the target language for output. The model then uses complex mathematical formulas that it refines to meet parameters that are defined through training with data sets of each of the languages.
NMT works with any language pair, and can learn new languages very quickly and easily. It can also be fine tuned to understand different accents or dialects (like British English vs. American English) and translate appropriately. The only downside is that since the algorithm learns based on new inputs, consistency can be diminished. New vocabulary may affect future translations, so additional neural models may be needed to overcome such faults.
Those are just a few of the many examples of how deep learning is working in our every day lives. Imagine the possibilities if we apply it to even more aspects! Of course there need to be some rules and oversight, but we are on the brink of some very amazing technological breakthroughs with deep learning models.
How to Learn More About Deep Learning
Do you want to learn more about deep learning and tech? You should! This is just the tip of the iceberg.
If you're ready to take the next steps in your tech career, check out an Ironhack Bootcamp! Or, join one of our events to learn a little bit more and connect with other like-minded technology enthusiasts!