AI vs. ML vs. DL
With the advent of innovative and path breaking changes in computing technologies especially computing power, storage and networking, there is a sustained interest in new technologies that drive over these changes. Among them, artificial intelligence (AI), machine learning (ML), and deep learning (DL) are most talked about. All these acronyms are often used in the same breadth in many discussions. They are similar but not the same. The main purpose of this blog is to clearly articulate the difference between these three terms.
First things first, all these acronyms come under the Artificial Intelligence (AI) umbrella. AI is not a standalone technology but an umbrella or agglomeration of technologies that covers everything related to making machines smarter that can learn and improve with time and experience. Machine Learning (ML) is not equivalent to AI but it is only a subset of AI. ML refers to an AI system that can self-learn based on an algorithm that can be of several kinds. An ML system can learn with time and experience and thus we can say that an ML system can get smarter and smarter over time without human intervention. Deep Learning (DL) is a superset of machine learning (ML) applied to large data sets.
Let us briefly delve into each of them to understand how they differ.
AI enables high computing machines to think like humans but without any human involvement. It is a broad and interdisciplinary area of computer science. AI systems can be categorized into three types
- Artificial Narrow Intelligence or ANI systems. It is a goal-oriented system that is programmed to perform a single task only.
- Artificial General Intelligence or AGI systems. It allows machines to learn, understand, and act in a way that is akin to the way a human does in a given situation. In other words, it enables a machine to mimic a human.
- Artificial Super Intelligence or ASI. It is still in the research phase and not yet realized. These AI systems empower a machine to show such intelligence that surpasses brightest humans and thus can control any human.
In brief, Artificial Intelligence deals with Reasoning and Problem Solving, Knowledge representation, Decision making, auto learning, Natural Language Processing (NLP), Perception, Motion and Manipulation, Social network analysis and problems involving general intelligence.
To understand better about types of artificial intelligence, here is a pictorial representation:
Machine Learning or ML is a subset of AI that uses statistical learning algorithms to build intelligent and smart systems that can automatically learn and improve without any human intervention. The review and recommendation system on Amazon is an example of ML. The machine learning algorithms are classified into four categories as per the purpose
- Supervised learning
- Unsupervised Learning
- Semi-supervised Learning
- Reinforcement Learning
In supervised learning, we feed the computer with labeled training data containing the input/predictors and we show it the correct answers (output) and from the data the computer should be able to learn the patterns so that it can predict on new data items. Supervised learning basically maps relationships and dependencies between the target prediction output and the given input features so that we can predict the output values for any unseen data based on those learned relationships. The main types of supervised learning problems include regression and classification problems. Popular algorithms under this category are Nearest Neighbor, Naive Bayes, Decision Trees, Linear Regression, Support Vector Machines (SVM) and Neural Networks.
In unsupervised learning, the machine is trained with unlabeled data with no output categories or labels and hence the algorithm cannot model relationships based on labels. Instead, these algorithms process the input data to mine for rules, detect patterns, and summarize and group the data points based on some parameter. All this is done in order to derive meaningful insights from the data and present it in a better way to the end users. The most popular unsupervised learning algorithms are Clustering algorithms and Association rule learning algorithms. k-means clustering and Association Rules are main unsupervised learning algorithms used.
Semi-supervised learning is a hybrid of the above two methods. In many real world scenarios, the cost to label generation is quite high as it requires skilled human experts to do labeling. So, we normally label only select observations and use semi-supervised algorithms to build the model. These methods cash in on the fact that even though the group memberships of the unlabeled data are unknown, the group parameters of this data carry important information.
Reinforcement Learning is an advanced category of machine learning. This system uses observations gathered from the interaction with the environment to take actions that would maximize the reward or minimize the penalty of the agent. Here the underlying algorithm is referred to as an agent. The agent relentlessly learns from its experiences with the environment so as to get the entire range of possible states. Agents automatically determine the ideal behavior within a specific context so that their performance is maximized. Reinforcement Learning is defined by a specific type of problem, and all its solutions are classed as Reinforcement Learning algorithms. Some popular reinforcement learning methods are Q-Learning, Temporal Difference (TD) and Deep Adversarial Networks. Reinforcement learning algorithms are heavily used in computer played board games such as Chess and Go, robotic surgeries, and self-driving cars.
Deep learning or fondly called DL is a subset of AI as well as a subset of ML. DL works in the same way as the human brain does. Just like the Human brain’s Neuron, an Artificial Neuron called perceptron starts working when data is fed to it. It then processes it by finding patterns, deciphers the information, tries to minimize the loss and gives a satisfactory result.
DL systems focus on learning data representations rather than focusing on task-specific algorithms. It makes use of Deep Neural Networks that are inspired by the structure and function of the human brain. DL is associated with learning from examples. DL systems help a computer model to filter the input data through layers to predict and classify information. Deep Learning processes information in the same manner as the human brain. It is used in technologies such as driverless cars. DL network architectures are classified into Convolution Neural Networks, Recurrent Neural Networks, and Recursive Neural Networks.
Deep Learning is used in Natural Language Processing (NLP), drug discovery and toxicology, bioinformatics, gene culture etc. Deep networks are made of multiple layers that data must pass through before producing the output. Deep Learning improves AI by enabling many of its practical applications.
To conclude, AI is a superset of ML and ML is a super set of DL. First AI came into this world. It was followed by ML and finally DL completed the triplet and made possible all that was still only to be dreamed of at the moment. Remember Data science produces insights, Machine learning produces predictions and Artificial intelligence produces actions.