Understanding Core Data Science Concepts: Big Data, Machine Learning, Data Mining, Deep Learning, and AI

Testing the challenges of data science through the relationships among several key concepts in the field. As we will see, differing opinions on specific concepts are inevitable; this is another viewpoint to consider.

There are virtually no articles online that compare and contrast data‑science terminology. Many people have written various pieces expressing their opinions, and the volume is overwhelming.

So, let me state plainly for anyone wondering whether this is one of those posts: yes, it is.

Why another? While many opinions partially define and compare these related terms, the reality is that most terminology is fluid and not universally agreed upon. Exposing oneself to other viewpoints is one of the best ways to test and improve one’s own understanding.

Therefore, even if I do not fully (or even minimally) agree with most of the prevailing definitions, there is still something to be gained by examining the core concepts of data science, or at least what I consider to be core. I will try to outline their relationships and how they combine as individual pieces of a larger puzzle.

As an example of differing opinions, Gregory Piatetsky‑Shapiro of KDnuggets compiled a Venn diagram that outlines the relationships among the same data‑science terms we will discuss. Readers are encouraged to compare this diagram with Drew Conway’s well‑known data‑science Venn diagram, as well as with my own discussion and the modified relationship diagram near the end of the post. Although differences exist, the concepts share a degree of similarity.

We will now examine the same six core concepts described in the Venn diagram and provide insight into how they fit together within the data‑science puzzle. First, we briefly set aside one of the most popular topics of the past decade.

Big Data

There are many articles that define big data, so I will not dwell on it here. In short, big data can be defined as a dataset that exceeds the capture, management, and processing capabilities of conventional software tools. The definition is both vague and accurate enough to capture its core characteristic.

To understand the remaining concepts, it helps to look at their search‑term popularity and N‑gram frequency, which distinguishes fact from hype. For the older concepts (1980‑2008), the N‑gram frequencies are shown above.

Recent Google Trends show two emerging terms rising, two others maintaining upward trends, and one term gradually declining. Note that big data is not included in the above graphs because it has already been quantified. Continue reading for further observations.

Machine Learning

According to Tom Mitchell’s seminal work, machine learning “concerns the question of how to build computer programs that automatically improve.” It is inherently interdisciplinary, drawing techniques from computer science, statistics, and artificial intelligence. The primary artefacts of machine learning research are algorithms that can automatically improve from experience and be applied across many domains.

There is no doubt that machine learning is a core aspect of data science. If the goal of data science is to extract insight from data, machine learning is the engine that automates that process. Machine learning shares much with classical statistics because it uses samples to infer and generalize. While statistics often focuses on description, machine learning rarely does; it uses descriptive steps only as an intermediate stage to enable prediction.

Machine learning is often considered synonymous with pattern recognition; although I do not dispute this, I prefer to avoid the term because pattern recognition can imply a broader, more complex set of processes than machine learning actually entails.

Machine learning has a complex relationship with data mining.

Data Mining

Fayyad, Piatetsky‑Shapiro, and Smyth define data mining as “the application of specific algorithms to extract patterns from data.” This emphasizes the use of algorithms rather than the algorithms themselves. We can define the relationship between machine learning and data mining as follows: data mining is a process in which machine‑learning algorithms are employed as tools to extract valuable patterns stored in a dataset.

Data mining, as a sister term to machine learning, is also crucial to data science. Before the explosion of the term “data science,” data mining enjoyed greater success as a Google search term. The trend extended five years beyond the graph shown above, indicating that data mining was once more popular. Today, data mining appears to be split between machine learning and data science. If we accept this explanation, it makes sense to view data mining as a process and consider data science as a superset of data mining.

Deep Learning

Deep learning is a relatively new term, although it existed before its recent surge in online searches. Because of its incredible success across many fields, research and industry are booming. Deep learning is the application of deep neural‑network techniques (i.e., neural‑network architectures with multiple hidden layers) to solve problems. Like data mining, deep learning is a process that employs a specific type of machine‑learning algorithm.

Deep learning has achieved impressive results. In light of this, it is important to remember a few points:

Deep learning is not a universal “silver‑bullet” solution for every problem.

It is not a legendary “master algorithm” that will replace all other machine‑learning or data‑science techniques—at least not yet.

Expectations must be tempered; despite recent breakthroughs in computer vision, natural‑language processing, reinforcement learning, and other areas, contemporary deep learning cannot yet handle extremely complex problems such as “solving world peace.”

Deep learning and artificial intelligence are not synonyms.

Deep learning can provide substantial assistance to data science as an auxiliary process and set of tools, making it a valuable complement to the field.

Artificial Intelligence

Most people find a precise definition of artificial intelligence—often even a broad one—difficult to grasp. I am not an AI researcher, so my perspective may differ from those in other fields. Over the years I have philosophically reflected on AI and concluded that AI, at least as we commonly think of it, does not truly exist.

To me, AI is a moving target, a shifting benchmark that never becomes fully attainable. Each time we claim an AI achievement, the accomplishment is soon re‑characterized as something else.

I once read that if you asked AI researchers in the 1960s what they meant by AI, they might have agreed on a pocket‑sized device that could predict our next actions and wishes and provide access to all human knowledge—a device that would be considered true AI. Today we all carry smartphones, yet few of us call them AI.

Where does AI fit into data science? I do not believe AI is a tangible entity, making it hard to place. However, many areas related to data science and machine learning are powered by AI and can be as valuable as any tangible tool. Deep learning research, for example, benefits from the AI spirit, and computer‑vision applications naturally arise from it.

AI is likely the deepest‑pocketed R&D engine, having never produced a concrete product in the same way other technologies have. While the direct line from AI to data science may not be the best way to view their relationship, many intermediate steps between the two have been developed and refined by AI in some form.

Data Science

After discussing the related concepts and their positions within data science, what exactly is data science? For me, it is the hardest concept to define precisely. Data science is a multidisciplinary field that includes machine learning and other analytical processes, statistics and related mathematics, and increasingly draws on high‑performance scientific computing—all aimed at extracting insight from data and using those insights to tell stories. These stories often involve visualizations and serve industries, research, or even personal curiosity, with the ultimate goal of generating new ideas from data.

Data science employs a variety of tools from many related domains (see all the content above). It is both a synonym for data mining and a superset that contains data mining.

Data science produces many different outcomes, but they all share the common thread of insight. Data science is all of this and more; for you, it may be something entirely different… We have not even covered data acquisition, cleaning, debating, and preprocessing! By the way, what exactly is data? Is it always large?

My view of the data‑science puzzle aligns well with the Piatetsky‑Shapiro Venn diagram shown at the top of this article, and I also believe it largely matches Drew Conway’s diagram, although I would add that Conway’s graphic refers more to data scientists than to data science itself.

Of course, this is not a complete picture of a constantly evolving landscape. I recall reading not long ago that data mining was a sub‑field of business intelligence; even though opinions differ, I cannot imagine that this is still a valid view today.

And there you have it: your favorite terms have morphed in new ways, and you may not forgive me for it. If you feel angry and want to tell me how wrong I am, remember that the purpose of this article was to present one person’s opinion. In that spirit, feel free to voice contrasting viewpoints in the comments (perhaps sharply). Otherwise, I hope this either introduces new readers to the data‑science puzzle or forces them to consider their own version of the puzzle in their minds.