Written by Michelle Dufflocq Williams and Jordyn Kronenberg
In the world of technology, concepts seemingly overlap and it’s easy for them to become jumbled. For example, the terms “deep learning,” “machine learning,” “artificial intelligence” and “cognitive computing” seem to be used interchangeably at times, making grasping these concepts confusing and overwhelming.
The purpose of this blog is to communicate simply and clearly: what exactly is the difference between deep learning, artificial intelligence (“AI”), machine learning, and cognitive computing?
It’s helpful to think of these concepts as Russian dolls nestled within the other. Machine learning is a type of AI, and deep learning is a type of machine learning. Therefore, both machine learning and deep learning are subsets of AI, but AI is not a subset of machine learning or deep learning.
Let’s start with AI. What is it exactly? Artificial intelligence is defined as “the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.”
Machine learning is a type of AI, and a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that machines should be able to learn and adapt as they mature. A basic understanding of machine learning is important because it’s currently the field of AI which is showing the most promising future, as it provides many of the tools that will be useful to the tech industry and society as a whole.
So if machine learning is cutting-edge, you can think of deep learning as the cutting-edge of the cutting-edge. Machine learning focuses on solving real-world problems with neural networks that are designed to mimic human decision-making, while taking some of the core ideas of AI. Deep learning, however, has a more narrow focus. It is a subset of machine learning tools and techniques, applied to solving pretty much any problem that requires human or artificial thought.
As a type of machine learning, deep learning has the capabilities to train a computer to perform human-like tasks, for example learning how to recognize speech, identify images or make predictions. Deep learning moves from communicating how to solve a problem to the computer, training the computer to solve the problem on its own. Deep learning sets up basic guidelines about the data, instructing the computer to learn on its own by recognizing patterns through using many processing layers.
The algorithms in deep learning are inspired by the structure and function of the brain and are called artificial neural networks. A neural network is a series of algorithms that attempts to identify underlying relationships in a set of data by using a process that mimics the way the human brain operates. Neural networks have the ability to adapt to changing input so the network produces the best possible result without the need to redesign the output criteria.
Much like a brain, a deep neural network has layers of neurons, which are artificial and figments of computer memory. Natalie Wolchover from Wired states that “when a neuron fires, it sends signals to connected neurons in the layer above. During deep learning, connections in the network are strengthened or weakened as needed to make the system better at sending signals from input data.”
This takes us to cognitive computing. So what is the difference between cognitive computing and deep learning? Cognitive computing is, in fact, another subfield of AI. Essentially, cognitive computing is focused on reasoning and understanding at an even higher level than basic AI, imitating the manner in which humans think. Instead of dealing purely with data or sensor streams, it processes more symbolic and conceptual information, with the goal of possessing the ability to make high-level decisions in complex situations. Although cognitive computing makes use of machine learning techniques, it is not exactly considered a machine learning method. “It is often a complete architecture of multiple A.I. subsystems that work together,” said Lynne Parker, director of the division of Information and Intelligent Systems for the National Science Foundation.
Our next blog post focuses on how deep learning and image recognition represent the start of unleashing the untapped potential within artificial intelligence and its diverse set of applications. For example, deep learning allows for users to give voice commands such as with Siri, or even to recommend products similar to the techniques Amazon employs on its consumers.
Deep Learning: Opening Eyes and Clearing Vision
Upon hearing the phrase “deep learning,” one may think of something along the lines of intense focus or working feverishly; in reality, deep learning is a subset of machine learning and artificial intelligence. Although machine learning and artificial intelligence are relatively well known in comparison to deep learning, one may not realize that deep learning makes itself useful in our current, everyday lives, very frequently and essentially undetectably. The purpose of this blog post is to explore applications of deep learning in our lives. Companies such as Facebook, YouTube, Twitter, Apple, and many others all employ a version deep learning to assist in their business endeavors in a variety of ways.
One of the most common uses of deep learning revolves around the idea of “seeing.” Deep learning helps “see” things that otherwise would fly under the radar, or require a significant quantity of manpower to achieve the same results. For example, deep learning serves as a form of facial recognition, which explains how Facebook recommends who to tag in your photo before you even start typing, or how YouTube can flag or remove inappropriate images. Deep Learning ends up being the a younger, sharper and more accurate set of eyes than any human employed, especially when assigned to a specific task. Deep learning can be usefully applied to many areas, but its use in the realm of social media and technology as a second set of eyes may qualify as the most recognizable.
An example of deep learning at work in a more specialized market is the company WEpods. WEpods is the world’s first driverless public transportation vehicle, located in the Dutch province of Gelderland. WEpods’ use of deep learning allows for the vehicle to evaluate its surroundings in order to safely navigate traffic and utilize the public roadways. It utilizes deep learning and neural networks to build an image of the surrounding environments and situational assessments. Eventually, these technologies begin to adapt automatically to changes and to teach itself to adjust accordingly.
The applications for deep learning seem to be limitless as technology advances and society becomes more accepting of these applications. In fact, society has developed almost an expectancy for–and a dependency on– deep learning. At this point, the thought of not interacting with our devices seems absurd. Could you imagine not using your cell phone to text a friend or check your most recent notification from Facebook or Instagram? The absence of technology to perform daily tasks, or even more complicated endeavors such as language translation, would leave a void in society’s daily routine. Ultimately, these tasks are achievable due to the advancements in deep learning technology.
Predicting The Future of Deep Learning: Barriers to Overcome
The current interest in deep learning is due in part to the buzz surrounding AI. Although it still has a long ways to go, deep learning enables classification, recognition, detection, and description, fostering a world in which communication and understanding sit at the forefront of society’s framework.
As we previously mentioned, deep learning is present in our daily lives, especially in our phones, computers and social networks. However, the next level of deep learning remains untapped due to certain barriers, including the need for a constant feed of data; enhanced computing power and functionality of smart devices; and a shared understanding about the value of deep learning.
Constant Data Feed. As mentioned in our first blog post, deep learning consists of neural networks, which are essentially a “computer system modeled after the human brain and nervous system.” These neural networks each contain their own data and algorithms, which essentially teach a computer how to complete a function. Within these neural networks layers can be applied to increase the specificity or add to the function itself.
Neural nets feed off of large stores of data and images in order to continue learning and adapting. Each of these uses stems from an increasing pool of data, a growing neural network and an increased understanding of how deep learning develops based on the information people feed it. Without a constant feed of data this natural evolution stalls. As a result, constant data input is required for deep learning’s necessary learning patterns.
Enhanced computing power and functionality of smart devices. Although nearly everyone owns a smart phone nowadays, those phones do not possess the necessary computing power and functionality for AI to reach the next level and perform significant and innate tasks. Devices must feature significant memory storage and historic data for deep learning to teach itself and to evolve with each new piece of information received. In a sense, deep learning needs to possess the aptitude to train itself through the tools and knowledge provided by the users.
Shared Understanding of the Value of Deep Learning. Even if our everyday devices were to attain the necessary memory and operational functionalities to facilitate deep learning applications, people in society lack an overall understanding of the capabilities and value of deep learning. Improved training could help develop an appreciation of deep learning, which would help it advance to the next level for the typical, everyday user.
More advanced deep learning could have a lasting impact through its ability to recognize and outsmart patterns. However, the use of deep learning on a larger, more specialized scale has yet to be identified effectively for mainstream consumers. The advancement of artificial intelligence has been a notoriously slow process up until the last decade or so; advancements made in recent years have created compelling applications for deep learning. These advancements have benefited society and offers the potential of bigger and better things to come.
“Dictionary.com.” Dictionary.com, Dictionary.com, www.dictionary.com/.
LeCun, Yann, et al. “Deep learning.” Nature News, Nature Publishing Group, 27 May 2015, www.nature.com/nature/journal/v521/n7553/abs/nature14539.html.
Marr, Bernard. “The Top 10 AI And Machine Learning Use Cases Everyone Should Know About.” Forbes, Forbes Magazine, 30 Sept. 2016, www.forbes.com/sites/bernardmarr/2016/09/30/what-are-the-top-10-use-cases-for-machine-learning-and-ai/#35d7090894c9
Noyes, Katherine. “5 things you need to know about A.I.: Cognitive, neural and deep, oh my!” Computerworld, IDG News Service, 3 Mar. 2016, www.computerworld.com/article/3040563/enterprise-applications/5-things-you-need-to-know-about-ai-cognitive-neural-and-deep-oh-my.html.
Parloff, Roger. “Why Deep Learning Is Suddenly Changing Your Life.” Fortune, 28 Sept. 2016, fortune.com/ai-artificial-intelligence-deep-machine-learning/.
Shapiro, Danny. “NVIDIA Deep Learning & AI Use Cases and Customer Success Stories.”NVIDIA, 1 Feb. 2016, www.nvidia.com/en-us/deep-learning-ai/customer-stories/.
Staff, Investopedia. “Neural Network.” Investopedia, 12 Sept. 2006, www.investopedia.com/terms/n/neuralnetwork.asp?lgl=rira-layout.
“What is deep learning?” What is deep learning? | SAS, www.sas.com/en_us/insights/analytics/deep-learning.html.
Wolchover, Natalie. “New Theory Cracks Open the Black Box of Deep Neural Networks.” Wired, Conde Nast, 9 Oct. 2017, www.wired.com/story/new-theory-deep-learning/