Machine Learning - The Subset of AI

You often heard the term Machine Learning. Today, we are trying to answer a few questions that arise in your mind. What is Machine Learning? What discipline of science does it belong to? What connection does Machine Learning (ML) have with Artificial Intelligence (AI)? Are ML and AI the same concepts and can be used interchangeably? What strong capabilities does ML have that make it the most exciting and dominant discipline? What are the types and applications of ML?

 Let’s try to understand the underlying concepts by starting from AI. 

Artificial Intelligence (AI)

Artificial Intelligence (AI) is the branch of computer science that deals with building machines and robotic systems to perform complex tasks requiring human intelligence. Put it simply, AI is the phenomenon of creating machines that are intelligent.

Human Intelligence is the mental capacity to learn from the past, understand and react to complex or evolving situations, and adapt to new environments or conditions. So, AI is trying to induce human intelligence in machines to mimic human behavior or thinking process. 

Based on capabilities, AI is categorized as:

• Weak AI

Also known as Narrow AI, capable of performing a single dedicated task. For example, google translate can translate from one language to another, but it cannot reply to your queries as Siri  does, and vice versa. 

• Strong AI

Also known as Artificial General Intelligence (AGI), capable of understanding and performing the tasks as human beings do. The strong AI machine should be as smart as the human mind to concentrate and solve various intelligent tasks. We did not get much success in creating strong AI solutions yet, except for a few tests like Turing Test. 

• Super AI

Intelligence beyond Human Intelligence, Super AI is the theoretical concept where machines can perform much better than humans.

Machine Learning

Machine Learning (ML) is the branch of AI, where machines learn from the experience without being explicitly programmed. It is the form of AI where the historical or past data is fed into the algorithms as an input, and the predictions of the outcomes or future values are received as an output. We can refer to it as data-driven AI. 

 

ML algorithms have strong capabilities to get insights into the past. Businesses can use ML to know the hidden patterns of customer behavior and operational processes, devise future strategies, and better decision-making after analyzing the previous trends and patterns. ML is successfully being deployed in many application domains including recommendation systems, object recognition, business intelligence, customer relationship management, self-driving cars, smart security systems, precision agriculture, smart farming, medical diagnostics, genome analyses, financial analysis, and so on.

Based on the learning principle, ML models are characterized as:

• Supervised Learning

Supervised learning models are trained on the labeled datasets to predict the future values accurately. The input data along with the labels is fed into the model, the model is trained on the input values and keeps on optimizing using the feedback from the associated labels until the required outcome is achieved. Supervised learning has achieved huge success in solving various real-world problems including spam filtering, image classification, and face recognition.

• Unsupervised Learning

Unsupervised learning models use unlabeled datasets to analyze and find the patterns or clusters hidden inside. Unsupervised learning algorithms learn on their own without taking supervision through labels. They are suitable for user segmentation, exploratory data analysis, pattern recognition, and cluster analysis. 

• Semi-supervised learning

Semi-supervised learning lies in between supervised and unsupervised learning. It serves as a bridge where labeled data is not available in enough volumes for supervised learning problems. Dataset labeling is one of the challenging problems for big data sets, so semi-supervised learning could be the best option for solving future real-world problems.

• Reinforcement Learning

Reinforcement learning is about adopting a sequence of actions in a particular situation with an objective to maximize the reward. Unlike supervised learning, reinforcement learning considers an agent who learns from its environment using trial and error. Every action the agent performs depends on the estimated reward will it gain.