As recently as the 1990s, few could have comprehended the capabilities that would eventually be common to 21st century mobile phones. They perform an extraordinary number of general and specific tasks, many having nothing to do with speaking or text messaging. Thanks to state-of-the-art technology, paired with the outside-the-box thinking of the human brain, smartphones have redefined modern-day convenience for billions of people all around the globe.
Much of their genius derives from machine learning. While the origins of machine learning predate the personal computer, mentioning the term during the era of pagers, Walkmans and VCRs would likely have led to confused looks. But given the extent to which machine learning is leveraged in a host of household items, appliances and gadgets, the raised eyebrows these days are in response to those who don’t know about it or haven’t at least heard of the once-obscure phrase.
As the digital technology and vast amounts of data have expanded, so has the jargon. Machine learning and deep learning are spoken about as if they’re synonymous, but they’re not. What exactly is the difference between machine learning and deep learning? Will deep learning replace machine learning? Where does artificial intelligence come into play?
With an online Master of Science in Engineering (MSE) from the University of California, Riverside, you can discover the remarkable advancements in data science and machine learning, and leverage your knowledge of it into a successful career.
What is the difference between deep learning vs. machine learning?
To understand the distinctions between machine learning and deep learning, you first have to define artificial intelligence, because each one of these methods is a subset of artificial intelligence. As its title implies, artificial intelligence is a technology where computers perform the types of activities and actions that typically require human intervention. Instead, they’re done by mechanical or computerized means. Such activities may include speech recognition, visual perception, language translation or memorization. Some AI consumer products may leverage all of these capabilities, such as virtual assistant devices made by Amazon or Google.
In short, artificial intelligence is the ability of a machine to replicate human intelligence or behavior.
Machine learning
Machine learning is a branch of artificial intelligence that deals directly with data. Although there are slight differences in how machine learning is defined, it generally refers to a series of complex processes that make certain conclusions in data patterns without requiring programming. In other words, it can act on its own. Whereas artificial intelligence requires input from a sentient being — i.e., a human — machine learning is typically independent and self-directed.
A classic example of machine learning is the push notifications you might receive on your smartphone when you’re about to embark on a weekly trip to the grocery store. If you typically go around the same time and day each week, you may receive a message on your device, telling you how long it will take to get to your destination based on travel conditions.
Another is the television or film recommendations you may get after you’re through watching a program on one of the streaming entertainment services. Based on what you’ve viewed in the past, Netflix, Hulu or Disney+ will make suggestions for new shows that align with your interests. And those recommendations are all based on the data patterns that the system assesses independently.
Social media websites like Facebook, Twitter and Instagram and search engines like Google, Bing and Duck Duck Go all use machine learning algorithms in one way or another to improve their services and make them more personalized for the end user.
Deep learning
Whereas machine learning is a subset of artificial intelligence, deep learning is a subset of machine learning, only it is more precise in terms of performance. For example, when Netflix makes a recommendation about an action/adventure film, documentary or biopic — and the suggestion proves to be on point — it does so by leveraging deep neural networks: multilayered algorithms that are composed of high volumes of data. These vast amounts of data that are parsed and assessed make machine learning processes — such as television recommendations — that are much more accurate. In essence, deep learning is machine learning only better, more targeted and more advanced. You might think of it as machine learning 2.0.
What are some other examples of deep learning?
While deep learning remains in its infancy relative to artificial intelligence, the speed with which machine learning models are advancing has allowed it to be leveraged more broadly. Translation services are a classic example. As anyone who is fluent in more than one language knows, the closed captioning on television or in movies may not always be precise when it comes to translating French, Spanish or German to English or vice versa. However, as Forbes points out, deep learning is allowing for translation services to be pinpoint accurate, translating a verbatim statement into its equivalent in another language. It does this by leveraging the bundle of neural networks that were unavailable with machine learning models alone.
An additional way deep learning is used in practice is news gathering. You may receive emails from a mobile app, website or organization that aggregates the daily headlines from well-known publications such as The Wall Street Journal, The New York Times or The Washington Post. Depending on what service you have — and what kinds of current event updates you’d like to receive — news aggregators rely on something called sentiment analysis to gauge how readers respond to certain types of news headlines. Related interest in a particular article may be guided by clicking on the link or time spent on the resulting page. This technology is what allows news aggregators to be more precise when it comes to compiling stories that are of interest to the intended audience.
Apple, Amazon, Google, Facebook and Twitter all utilize deep learning in the products they sell or services they offer.
Will deep learning eventually replace machine learning?
Because deep learning is inherently more accurate than machine learning — making it presumably better for customer satisfaction, translation, language recognition and other services — some question whether it will eventually render machine learning obsolete. But several tech experts believe otherwise, mainly because certain actions or activities don’t always require advanced customization. For example, as noted by Sambit Mahapatra, a tech contributor for the website Towards Data Science, deep learning may be preferable to machine learning in cases where data sets are large. This may include services like voice, speech or image recognition or natural language processing.
But in cases where data sets are smaller — such as logistic regression or decision trees — machine learning may be sufficient because the same result can be reached but in a less complex fashion.
Deep learning vs. machine learning: What specialized hardware and computer power are needed?
Another reason why machine learning will endure is due to infrastructure. As Mahapatra pointed out, deep learning techniques require high-end infrastructure. This includes hardware accelerators, such as graphic processing units (GPUs), tensor processing units (TPUs) and field programmable gate arrays (FPGAs). In addition to the cost of such infrastructure, the calculations take longer to perform.
As InfoWorld points out, classical machine learning algorithms have their place and may be a more efficient form of artificial intelligence. It all depends on the issue or service that’s necessary and how much data is involved.
Are there some companies that use machine learning more than others?
While some organizations that now regularly use machine learning predate the AI-based technology, an increasing number of companies likely wouldn’t exist in their present form without it. A classic example is Uber. With tens of millions of active users around the world — from 63 countries and over 700 cities, according to its website — Uber is among the largest global ridesharing providers, with drivers helping people get to their intended destinations by connecting them electronically via the mobile app.
Uber is able to do this through a platform called Michelangelo. As elaborated on at its website, Michelangelo is an internal software-as-a-service program that “democratizes machine learning” and helps its internal teams manage data, make and monitor predictions and provide time series forecasting at scale.
Logan Jeya, lead product manager at Uber, noted that Michelangelo is a multipurpose solution that the company uses for a wide range of needs, from training incoming employees to tracking business metrics. And it’s all made possible by machine learning.
“Machine learning is one of the coolest areas to work in, period,” Jeya explained. “It is this cyclical process on how you can use data to effectively create these smaller-scale decision systems that ultimately allow you to operate Uber at the scale we do.”
As Uber has expanded its reach, it’s relied more heavily on machine learning to grow its base and become the leader in ridesharing that it is today. From optimizing their maps so that they’re more accessible and accurate, to helping drivers narrow down how quickly they’ll arrive at their intended destination, Uber leverages machine learning in several different ways so the user experience is more enjoyable for riders as well as those who work for the company.
UberEats, the delivery arm of Uber, also takes advantage of machine learning so that those who download the mobile app can receive recommendations on which restaurants to try that are in the area and feature menus that are in line with certain tastes or cuisine choices.
Other delivery apps, like Instacart, Postmates, Slice, Grubhub and ChowNow, also use machine learning to improve the customer experience and compete or partner with other entities in the restaurant, grocery or ridesharing space.
While the fruits of machine learning and deep learning are increasingly apparent and growing in use, you may still be wondering how it all works from a hands-on process perspective. You’ll discover the answers to this and much more through the online Master of Science in Engineering program at UC Riverside. The data science specialization will not only give you a better grasp of the core concepts that revolve around the branches of artificial intelligence, but it will also give you the connections, expertise and training to turn your intellectual curiosity into a career filled with possibilities.
Visit the program page to learn more about the curriculum, faculty, eligibility requirements and what graduates have to say about their experience.
Recommended Reading
Career Spotlight: Data Scientist
How to Become a Machine Learning Engineer
Sources
Forbes, 7 Types of Artificial Intelligence
Forbes, 10 Amazing Examples of How Deep Learning AI Is Used in Practice
Towards Data Science, Why Deep Learning Over Machine Learning?
InfoWorld, Deep Learning vs. Machine Learning: Understand the Differences
Uber, Science at Uber: Powering Machine Learning at Uber
Uber, Meet Michelangelo: Uber’s Machine Learning Platform