Artificial intelligence (AI), machine learning and deep learning have all long been major areas of interest for enterprise and consumer technology vendors, as well as for computer science researchers. All three involve the concept of intelligent machines or programs that can think and reason like humans. This idea predates the invention of the computer itself, but it has only become somewhat realistic in recent memory, thanks to advances in processors, networks and data storage.
AI and machine learning are often used as interchangeable terms. However, there are important differences between them, and between these two and the concept of deep learning. As a data scientist or other IT professional, you may work with projects that touch upon one or more of these topics, so it’s important to know the distinctions. Let’s focus on machine learning versus deep learning.
Machine learning vs. deep learning: What they have in common and how they differ
In most contexts, machine learning refers to systems that can make inferences and identify patterns in data by themselves, without requiring explicit programming each time. This self-directed behavior sets machine learning apart from other forms of AI such as rules engines, which can only perform tasks they’ve specifically been set up to do.
In contrast, machine learning-driven applications and services are supposed to be similar to how humans learn, i.e. by being able to dynamically draw conclusions as information changes and/or more of it becomes available. For example, a facial recognition program could use machine learning to become increasingly good at identifying individuals based on certain features, in the same way that an actual person might improve their skills at spotting specific types of clouds in the sky or species of birds in a park.
The key to machine learning’s viability is how it eliminates the need for extensive human intervention. Accordingly, it can help process huge amounts of data with relatively little overhead. Companies including on-demand transportation service Uber and online physician scheduling app ZocDoc have put machine learning to work in tasks involving large collections of information that would be impractical for a person to comb through on their own.
In addition to trip ratings and feedback via the main app, Uber riders also contact its support team on channels including email and social media. Ticket volumes are unsurprisingly high, as Uber connects its customer base with billions of rides each year.
To handle this flow of tickets, the Uber machine learning team assembled an architecture that could draw inferences from each message’s textual contents, the associated trip fare in question and other features to efficiently categorize and route each one. The new setup helped reduce customer service handling times and improve response accuracy.
ZocDoc is designed to streamline the process of scheduling medical appointments. One of the biggest potential hurdles in this process is finding in-network providers for each patient’s insurance coverage, as an in-network provider is usually much more affordable than one that’s out of network.
To improve its suggestions, ZocDoc implemented a machine learning engine that analyzed user-submitted photos of their insurance cards. This task was inherently challenging, due to the low resolution of the photos and the repetition of data on many insurance cards. ZocDoc created a machine learning structure that featured a core classifier system that analyzed multiple types of card information, used image recognition to align these findings and applied optical character recognition to ensure ultimately accuracy.
As a term, deep learning is less widely used than machine learning. It generally refers to a more intense form of machine learning, with sophisticated mathematical models and greater overall adaptability that together allow for more accurate results. The “deep” in its name describes its relatively high number of layers, i.e. mathematical operations that get applied to its data.
Deep learning is based on advanced artificial neural networks, which are modeled on current understanding of how the human brain works. Key applications for deep learning include image and sound recognition software, recommender systems (e.g., for YouTube or Spotify), natural language processing applications and video game AI.
One widely covered example of deep learning is the appropriately named DeepMind, a Google creation designed to play both traditional board games and some video games. DeepMind became initially famous for its mastery of Go, a game with so many possible operations that simply brute-forcing it ― i.e., going through all possibilities to find the best one ― is less efficient than playing it like a human expert might. DeepMind has applied similar principles to online gaming.
In the years ahead, we should expect machine learning and deep learning to become more capable, thanks to improvements in the underlying technical infrastructure as well as the collection of more training data. Services like Apple Siri, Amazon Alexa and Google Assistant are all testaments to how these technologies continue to progress.
A look at UCR MSE courses on machine and deep learning
As a student in the online Master of Science in Engineering (MSE) track at the University of California, Riverside, you can learn the essentials of machine learning and deep learning as part of the data science specialization. Relevant courses include Machine Learning, Foundations of Applied Machine Learning and Advanced Computer Vision.
The UCR MSE is flexible, with multiple options for starting times and a completely online structure. With expedited completion, you can earn your master’s degree in as few as 13 months and be prepared to pursue careers in the fast-growing category of computer and information research scientists, which the U.S. Bureau of Labor Statistics foresees growing 19% from 2016 to 2026.
To get started, visit the main program page for a quick overview. You can also download the program brochure there.