Deep Learning Fundamentals - Premium Edition

AI vs. Machine Learning vs. Deep Learning - Relationship Overview

We'll first start our deep learning journey by understanding where the field of deep learning falls in relation to the fields of machine learning and artificial intelligence (AI).

Deep learning is a subfield of machine learning, which itself is a subfield of artificial intelligence.

While our focus in this course will be on deep learning, we'll take a moment to get introduced to the broader fields of artificial intelligence and machine learning.

Artificial intelligence introduction

Generally speaking, artificial intelligence is intelligence that is inhabited by a machine, rather than from a natural biological life form.

If we wanted to get philosophical, then we could dwell on this definition for quite a while, contemplating the definition of intelligence and debating about what is considered "natural" life forms. For now, though, let's agree to think about this more broadly and generally for the time being.

We may consider a machine, or a specific program on a machine, to be artificially intelligent if it can perceive its environment and act in that environment in such a way to reach a given goal.

Applications of AI

Artificial intelligence may seem a bit sci-fi or futuristic sounding on the surface, but actually, it's widely implemented and already in use today!

Ever wonder how YouTube chooses which videos to recommend to you next? That's AI. How about when you use a face filter available on social media apps? How is it able to fit so seamlessly to your own and everyone elses face? AI! What about when Facebook automatically tags you in a photo? You guessed it, AI.

Oh, and I'm sure you're familiar with all the craze around self-driving cars, right? If you weren't sure how that was happening before, I bet you can guess the answer now.

Narrow AI

These are just a few of the many applications of AI already in use today, and each of these AI applications are what we call narrow AI. Narrow AI is AI that is focused only on a narrow subject or task.

For example, the narrow AI involved with auto-tagging photos on Facebook may just have the task of accurately tagging photos of people. It has the goal to maximize the probability that the tag it applies to a photo is correct.

For self-driving cars, generally speaking, we can think about the narrow AI as having the task of safely driving the vehicle from point A to point B. This is an oversimplification, but still the task is narrow. The self-driving car is not trying to safely drive at the same time as trying to come up with the recipe for the best donut.

Artificial General Intelligence (AGI)

Artificial general intelligence (AGI), on the other hand, is an AI system that is able to understand and learn to do any intellectual task that a human can.

So, an AGI self-driving car would be able to safely drive you around, develop the recipe for the best donut, accurately predict the weather, be your physical trainer, have deep and meaningful conversations, diagnose and treat cancer, etc. Throw robotics into the mix as well, and we're in for a totally new world!

AGI is a very active area of research and has not yet been solved. Once it is, our world as we currently know it will be forever changed.

Our focus in this course will be on narrow AI via deep learning, which we'll be introducing soon.

Machine learning introduction

Now that we have a basic understanding of the field of AI, we can now be introduced to the subfield of Machine Learning (ML). Machine learning is one approach that we can take to implement AI.

Machine learning is the practice of using algorithms to analyze data, learn from that data, improve, and then make a determination or prediction about new data.

The key distinction of machine learning algorithms as opposed to traditional non-ML algorithms is that ML algorithms learn from the data.

Within the field of machine learning, there are many types of learning algorithms that work with all types of data to achieve all sorts of tasks. Some common ML algorithms include:

• Regression
• Decision Trees
• Support Vector Machines (SVM)
• Naive Bayes Classifiers
• K-Nearest Neighbor
• Artificial Neural Networks (belongs to subfield of deep learning)

With machine learning, rather than a developer manually writing code with a specific set of instructions to accomplish a specific task (as with traditional programming), the machine is instead trained using data and algorithms that give it the ability to learn from the data and perform the task without being explicitly told how to do so.

This "learning" may sound like magic, but we'll see in later lessons exactly how this is achieved.

Let's look a basic example of how a task may be done using machine learning verses traditional programming. Let's suppose the task is to analyze articles from a fitness publication and classify the articles as either being about diet or exercise.

Traditional programming approach

The algorithm may be programmed to first look for particular words associated with diet or exercise.

With conditional statements, the algorithm would classify articles as either diet or exercise based on the words that it knows falls into these categories.

# pseudocode
diet = [
    "diet", 
    "food", 
    "eat"
]

exercise = [
    "exercise", 
    "movement", 
    "weightlifting"
]

These words are arbitrarily chosen by the programmer. Once we have the list of diet and exercise examples, one simple algorithm is to simply count up the occurrences of each type of word in a given article. Then, the article can be classified as diet or exercise based on which word count is higher, the diet words or the exercise words.

Machine learning approach

The algorithm analyzes given media data and learns the features that classify what a diet article looks like versus a exercise article.

With what it has learned, the algorithm can then classify new articles as diet or exercise. A machine learning developer in this case won't explicitly specify the words for the algorithm to recognize. Instead, the algorithm will have “learned” that certain words are associated with diet vs exercise based on labels given to each article it examines during the learning phase.

# pseudocode
articles = [
    {
        label: "diet",
        data: "Wholegrains are a great source of fiber..."
    },
    {
        label: "diet",
        data: "Processed meats, like hotdogs..."
    },
    {
        label: "exercise",
        data: "Once your heartrate reaches..."
    },
    {
        label: "exercise",
        data: "Squats activate the quads..."
    }
]

Deep Learning Introduction

Now that we have a basic understanding of machine learning is, the next topic to introduce is the topic of this course itself, deep learning! As previously mentioned, deep learning is a subfield of machine learning.

Artificial Neural Networks (ANNs)

Just as with machine learning, deep learning uses algorithms learn from data. It is the specific type of learning algorithms that deep learning uses that creates the boundary between it and machine learning in general.

Deep learning makes use of algorithms called artificial neural networks (ANNs) to learn data. These algorithms are loosely inspired by the biological structure and function of the human brain's neural networks, hence the name.

We'll get a more thorough introduction to ANNs in an upcoming lesson.

We should now have a general understanding of the relationship between artificial intelligence, machine learning, and deep learning.

We have much more to discuss regarding deep learning and artificial neural networks – an entire course worth, in fact! We'll pick up next time with a more thorough introduction into deep learning itself.