13.5. Machine Learning and Deep Learning
Now that you understand how AI systems differ from humans and machines and their evolutionary arc, the next question is how they work? Intelligent systems are created through machine learning and deep Learning techniques. These techniques are used to train, test, and validate AI models that can be implemented into different systems to automate simple tasks, process massive amounts of data and even develop new products. Let’s look at the relationship between artificial Intelligence, machine Learning, and Deep Learning, as illustrated in the image below. It is important to understand the relationship between these terms as they are similar and can be confusing.
| Artificial Intelligence | Refers to a machine mimicking human behaviour. |
| Machine Learning | A technique in which a computer is able to solve problems using sets of data. |
| Deep Learning: | A subset of Machine Learning – Deep Learning refers to when computers are able to solve more complex problems that cannot be solved using the traditional Machine Learning method (IBM Cloud Education, 202,0 June 3). |
Machine Learning
Machine learning is a technique that is used by an AI system to analyze data, find patterns and make decisions automatically or with minimal human support. Machine Learning enables a system to sort, organize, and analyze data in order to draw important conclusions and make predictions. Machine learning techniques can be subdivided into three categories: supervised learning, unsupervised learning, and reinforcement learning.
| Supervised Learning | The form of Machine Learning in which data is labelled and categorized into groups by humans such as fruits and vegetables or engineers and doctors. Algorithms are used to classify data based on labels. Techniques can include decision trees, neural networks, and regression. |
| Unsupervised Learning | The form of Machine Learning in which a machine is given unlabelled data and must find its own connections through analysis, clustering, and identifying patterns. |
| Reinforcement learning | The trial and error-based machine learning method is used in which a machine learns from mistakes and improves upon them. Uses unlabeled data and learns through a series of trial-and-error actions. |
Supervised learning is typically used to create algorithms where the model must learn how the inputs affect the outputs. Therefore, these models are used in instances like predicting the prices of houses, image classification, and even weather prediction.
AI systems created through unsupervised learning algorithms are given unlabelled data and like the human brain, the model is expected to connect information together, sort of like detecting patterns within data sets. The data given to the machine is unorganized, and the machine does the work of drawing inferences and conclusions from existing data points to sort it out into groups or clusters.
Example: Supervised and Unsupervised Learning
If a company wants to create a new marketing campaign for a particular product line, they may look at data from past marketing campaigns to see which of their consumers responded most favorably. Once the analysis is done, a machine learning model is created that can be used to identify these new customers. It is called “supervised” learning because we are directing (supervising) the analysis toward a result (in our example: consumers who respond favourably).
If a retailer wants to understand the purchasing patterns of its customers, an unsupervised learning model can be developed to find out which products are most often purchased together or how to group their customers by purchase history. This is called “unsupervised” learning because no specific outcome is expected.
Machine learning models can form the core of logistics and supply chain solutions in terms of optimizing the product packet size, delivery vehicle selection, delivery route selection, and delivery time computation. For instance DHL uses Amazon’s Kiva robotics (improve speed, accuracy) for network management.
Reinforcement learning is similar to how humans learn: they learn from mistakes. Systems created using reinforcement learning become smarter as more data is collected, and the machine learns from any mistakes it makes. In this type of algorithm, no supervision is involved as it is based on a reward or goal-driven system where the machine is rewarded (assigned positive values) for desired behaviours and punished (assigned negative values) for undesired ones. Therefore, this algorithm is optimal for use in instances where a very specific behaviour needs to be automatically determined.
Example: Reinforcement Learning
Fanuc, a Japanese industry-based robotics company, has been leading with its innovation in this field – it is working actively to develop reinforcement learning in its own robots. They use reinforcement training so that the robots can train themselves on how to pick an object from one box and place it into another box. They use this process for different tasks, and as a result, they can build robots that can complete complex tasks more quickly than humans (Sharma, 2020).
Deep Learning
Deep Learning is a subfield of Machine Learning and is another technique in which a machine is trained with massive datasets. However, there is significantly less human work involved. In Deep Learning, the model can extract features and classify the data itself instead of having a human identify the features. We can view Deep Learning as an intelligent technique that uses multiple layers of a computing system known as Artificial Neural Networks to categorize information within a data set. Much like the neural pathways in a brain, scientists and engineers created a web of connected paths for electrical messages, which are called Artificial Neural Networks (A.N.N). These neural networks function similarly to the neurons that fire electrical impulses within the brain, which then send messages to and from the brain in living organisms to accomplish tasks. The data in A.N.Ns are activated and sent to other layers through a series of mathematical calculations to simulate decision-making in computers.
| Neural Networks | Algorithms that process information in a similar way to the human brain. Contain a collection of interconnected nodes. |
Deep learning and neural networks have been deployed in several fields, such as computer vision, natural language processing, and speech recognition. It has been used in many healthcare applications for the diagnosis and treatment of many chronic diseases. These algorithms have the power to avoid outbreaks of illness, recognize and diagnose illnesses, and minimize running expenses for hospital management and patients. While machine learning, deep learning, and neural networks are all subsets of artificial intelligence, deep learning is actually a subset of machine learning, and neural networks are a subset of deep learning.
Deep learning can be thought of as the automation of predictive analytics. Deep learning is essentially a neural network with three or more layers which allows it to learn from a large amount of data. Deep learning is behind many artificial intelligence applications improving on automation and analytics. It is used in such applications as voice-activated electronics, self-driving cars, and credit card fraud detection. The evolution of deep learning started after the invention of neural networks, by adding more neurons and additional hidden layers to neural networks makes deep learning more cultivated. Like machine learning, deep learning is also categorized into subcategories, i.e., supervised and unsupervised.
Machine Learning vs Deep Learning
Machine learning focuses on developing algorithms that can alter themselves without human involvement to take defined data and generate a required output. Deep learning uses neural networks to learn unsupervised from unstructured or unlabeled data.
Machine learning uses algorithms to analyze data, learn from it, and make smart decisions based on the knowledge learned, while deep learning organizes the algorithms into layers to form artificial neural networks that can learn and make intelligent decisions independently.
| Deep Learning | Machine Learning | |
|---|---|---|
| Data Requirements | Requires a large dataset | Works well with a small to a medium dataset |
| Hardware Requirements | May require machines with a GPU | Works with low-end machines |
| Training Time | Longer training times (ranging from a few hours to weeks) | Shorter training times (from a few seconds to hours) |
| Processing Time | Faster processing times | Slower processing times |
| Number of Algorithms | Fewer Algorithms required | More algorithms required |
| Data Interpolation | Difficult | Difficulty Varies |
“Module 1: Introduction to Artificial Intelligence” & “Module 2: Machine Learning and Deep Learning” from Thriving in the Age of Artificial Intelligence by the Faculty of Engineering and Applied Science at Ontario Tech University under a Creative Commons Attribution license, unless otherwise noted.
