If you are new to Data Science, it is possible that you have seen the three buzzwords being used time and again (often, interchangeably but mistakenly).
Artificial Intelligence, Machine Learning & Deep Learning. So, what is the difference? In short, each of them is a subset of the earlier. We will let the diagram do the explaining
Artificial Intelligence | Source: BuzzRobot
Let’s begin by understanding each of them separately.Artificial Intelligence
AI refers to the intelligence exhibited by machines (like, problem solving and learning) using techniques or algorithms like search, logic, if-then rules, decision trees and machine learning (including deep learning).
Artificial Intelligence Illustration | Source: IBM
AI can be categorized into Applied AI and Generalized AI. While the former is defined as an application to enable a system that replicates human intelligence for a dedicated purpose; general AI is aiming to build ‘thinking machines’; that is, general-purpose systems with human intelligence.
Which gets us to the three levels of AI- Narrow AI refers to a machine which is good at performing a single task. For example, playing chess or forecasting weather. This is the only form humans have been able to achieve.
- Artificial General Intelligence (AGI) will be machines which are able to comprehend their environment just like we do.
- Superintelligent AI. As Nick Bostrom describes it, this is “an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom, and social skills.”
Understanding of human brain and advancements in technology (like, GPUs and Big Data movement) has evolved AI overtime. The focus has shifted to emulating human brain so the machines are able to make decisions and carry out tasks in more human ways.
Clearly, artificial intelligence and machine learning are not entirely separate. Artificial Intelligence is the umbrella term which encompasses Machine Learning and Deep learning.
Machine Learning
Machine learning uses algorithms to parse data, learn from the results and make decisions/predictions. Instead of coding or providing instruction to the machine in any form, it is trained using data and advanced algorithms that enables it to learn how to perform a task themselves. The techniques include linear and logistic regression, k-means clustering, decision trees, random forests, and more.
Machine Learning Illustration | Source: IBM
Machine learning technology falls into two major types: supervised machine learning and unsupervised machine learning. Google’s improved algorithm named Rankbrain or facebooks automatically suggesting friends to tag are a few use cases for machine learning
Supervised learning, as the name suggests, requires that the output is known and data which is used to train the machine is labeled with correct answers. According to Jason, it is called supervised learning because the process of an algorithm learning from the training dataset can be thought of as a teacher supervising the learning process.
The data used in unsupervised machine learning is not completely labeled. It depends on recognizing patterns within the data or comparing it with other data. As the data set grows, the accuracy of algorithms increases as more patterns emerge. Clustering, concept search, and near-dedupe (compares similar data and excludes data based on degree of similarity) are a few models which are used.
This image from Datconomy sums it up
Image courtesy- Kontagent
While spam detection and voice recognition (based on training the device with your voice, especially on your phone) are a few examples of supervised learning, this video shows unsupervised learning model created by Facebook AI Research (FAIR) team. According to the team, it has developed a system that can “watch” a series of visual tests — in this case, sets of precariously stacked blocks that may or may not fall — and predict the outcome.
Similarly, we can take an example of Deep blue and DeepMind. While Deep Blue and DeepMind are both types of AI, Deep Blue was rule-based and dependent on programming and hence not ML. DeepMind, on the other hand, beat the world champion in Go by training itself on a large data set of expert moves.
When we talk about artificial intelligence and machine learning, we can safely assume that all machine learning counts as AI, but not all AI counts as machine learning.
Deep learning
With improved processing capabilities being available now, a key ML approach that remained dormant for a few decades has now gained attention. So, let’s see what artificial neural networks are.
Artificial Neural Network is an attempt to replicate the human brain. It is a technique which teaches a machine to do what is natural to a human mind – learn by example. It does not need to use any task-specific algorithms like data classification.
According to Jacquilline, A neural network passes data through interconnected layers of nodes, classifying information and characteristics of a layer before passing the results on to other nodes in subsequent layers. The difference between a neural network and a deep learning network is contingent on the number of layers: A basic neural network may have two to three layers, while a deep learning network may have dozens or hundred
While this technique requires large amounts of labeled data, it requires minimal feature engineering (as compared to a linear regression model). The machine will learn classification based on the images, text or sound you have put in.
“If you’re looking at classifying images for a cat, you’ll have to feed your data set with a bunch of images of a cat. But you don’t necessarily have to say that a cat is something with cute ears or whiskers,” explained Verizon Wireless Data Scientist Aurora LePort during the opening panel of DataScience: Elevate.
One of the most popular types of deep neural networks is known as convolutional neural networks (CNN or ConvNet). According to Mathworks, A CNN convolves learned features with input data, and uses 2D convolutional layers, making this architecture well suited to processing 2D data, such as images.
You don’t have to manually identify features with CNN. It does it automatically by extracting features from the images. This is what makes deep learning models highly accurate. As mentioned earlier, CNNs have hundreds of hidden layers which help recognizing any shape or object.
what is the difference between Deep learning and machine learning?
- Data: While deep learning algorithms don’t perform well if your data pool is small, machine learning algorithms use handcrafted rules which will lead to comparatively more accurate results.
Data science techniques scale | Source: Analyticsvidhya
- Hardware required for deep learning and machine learning: Deep learning require high end machines to work accurately which is not the case with traditional machine learning. GPUs are an integral part of deep learning algorithms working because it involves a large amount of data processing.
- Feature engineering: In machine learning, features need to be identified and hand-coded. For example, features can include colors, shapes, textures etc. The success of a machine learning algorithm depends on how accurately these features were extracted. Deep learning on the other hand learns these features from the data.
- Interpretability in deep learning and machine learning: Machine learning algorithms like decision trees or linear regression can explain why it gave a particular output. It is easier to interpret the reasoning behind it. On the other hand, this isn’t possible with deep learning. According to Faizan, Deep learning does not reveal why it has given that score. Indeed, mathematically you can find out which nodes of a deep neural network were activated, but we don’t know what their neurons were supposed to model and what these layers of neurons were doing collectively. So we fail to interpret the results.
Is deep learning a subset of machine learning?
Yes. It’s also AI, which is where the confusion usually starts. The three terms get used as if they compete with each other, when in fact one sits inside the next. Take a simple example from a bank. The fraud team might use a machine learning model that looks at transaction amount, location and time of day to score each payment. Someone decided those features mattered and fed them in. That’s machine learning. Now suppose the same bank wants to read scanned cheques. Nobody can write down every feature of every handwritten digit, so a deep learning model, built from many layers of neural networks, works out those features by itself from thousands of images. Both are machine learning, and both are AI. Only the second is deep learning. One way to keep it straight is to think of nesting dolls. AI is the big outer doll. Open it and you find machine learning; open that and you find deep learning. Whatever is in the smallest doll is also inside the bigger ones. The reverse isn’t true, and a lot of AI in use today involves no deep learning whatsoever. Beyond that, the two part ways on data, hardware and explainability.How machine learning and deep learning compare
| Machine Learning | Deep Learning | |
|---|---|---|
| What it is | Algorithms that learn patterns from data | A type of machine learning built on many-layered neural networks |
| Data needed | Can work well with smaller datasets | Usually needs a lot of data |
| Feature engineering | People pick the features | The model finds features on its own |
| Hardware | Standard CPUs are usually enough | Generally runs on GPUs |
| Training time | Minutes to a few hours | Hours, sometimes weeks |
| Interpretability | Fairly easy to explain | Often hard to explain, a black box |
| Examples | Spam filters, fraud scoring, product recommendations | Image recognition, voice assistants, generative AI |
Akshay Badkar, on iamwire.com, has tabled the differences between deep learning and machine learning for ease of understanding.
Difference between deep learning and machine learning illustration | Source: iamwire.com
By now, it is clear that deep learning is a subset of machine learning and both of them are subsets of artificial intelligence.
Modernize with confidence. Build for what's next.
Talk to our team about embedding AI into the core of your business.