Crystal clear view of AI, Machine learning and deep learning

What does the term 'Artificial Intelligence' signify?

Artificial Intelligence is the ability to achieve goals, especially in a complex, uncertain environment. It's a broad field that covers many things, from comparatively simple algorithms to computer vision related algorithms used in self-driving cars that can recognize images,sound and any pattern.

Machine Learning (ML) is the most successful approach to Artificial Intelligence (AI). It is a subset of AI and it is based on learning from data itself. There can be other approaches like rule based approaches that can be said to be AI, but not ML.

===================================

What does the term 'Machine Learning' denote?

ML algorithms are evolved forms of normal algorithms. They make the computer programs "smarter", by letting them automatically learn from the data that is provided to it.

Machine learning algorithms help us to make decisions with great confidence and these algorithms can be improved over time through reinforcement learning. The accuracy will improve with the feeding of more test data. The same algorithms can be used to train different models

Machine Learning is a set of technique which programs the computers helps it in the automated decision making processes. the computer makes decisions by understanding patterns in the historical data and making generalizations for the future data, on the basis of this knowledge or learning from past data.

The decisions, being discussed here can range from making predictions of a customer's behaviour, to forecasting the weather conditions, to even identifying voices and faces from an audio and a video clip respectively!

Supervised ML Algorithms are further divided into three kinds:

A. Supervised Learning

In supervised learning, we have both input and output variables, and the ML algorith is required to generate a function which can make accurate predictions of the output on the basis of the given input variables.

i. Regression: A supervised ML algorithm is said to be regression algorithm in case the output variables are continuous in nature like weight, height etc. Sales prediction is an example where revenue is predicted based on a set of attributes(better known as predictors)

 ii. Classification: A Supervised ML algorithm is termed as 'classification' when the output variables are categorical in nature like yes, no or male, female etc. Spam prediction/classification is an example where email is classified/predicted to be spam or not-spam based on text of the email.

Ever wondered what helped Amazon, Google, Facebook, Netflix, or Youtube, suggest recommendations to you ? Sure, supervised ML application was at play here! It is also extensively being used for fraud detection.

B. Unsupervised Learning

In unsupervised machine learning algorithms, we have only input variables(predictors/set of X's), not the corresponding response(Y). The main objective of this type of algorithms is to perform any action based on the pattern of predictors. For example, similarity, closeness of instances/rows based on predictors(set of X's) is to be measured. The unsupervised ML algorithm is further divided into two kinds:

i. Cluster analysis: Here, the built-in groupings in the data is discovered through ML algorithms. Customer segmentation is an example where you try to understand similar customers by discovering groups based on similarity/closeness in attributes.

ii. Association: Here, the algorithm discovers the relationship between variables in the dataset that might be useful. Market basket analysis is an example where you try to find out which items are sold together frequently.

Unsupervised machine learning algorithms can be used to a great extent in Marketing domain, for segmentation and target marketing purposes.

C. Reinforcement Learning(RL)

In case of reinforcement ML learning algorithms, the machine is trained to to understand the activities going on in the world around just like human beings by applying a certain level of intelligence. The machine learns from the repercussions of its own actions without being taught a thing! In short, the purpose of reinforcement learning is to have a good policy, not the good decision. To understand it, let us take an example of your pet dog. Consider teaching the dog a new trick: you cannot tell it what to do, but you can reward if it does right things and penalize if it does wrong things. RL agent(Dog) has to figure out what it did that made it get the reward/punishment. An RL agent(Dog) learns by interacting with its environment by observing the results of these interactions

=========================================================

 What does the term 'Deep Learning' mean?

Deep Learning is a subset of Machine Learning that tries to learn high level features from the given data. Thus, the problem it solves is reducing task of making new feature extractor for each and every type of data specially speech, image etc. Deep Learning algorithms try to learn features such as length of forehead,distance between ears etc. These features are used for classification, prediction and other such tasks. Thus, this is a major step away from typical ML Learning Algorithms. In short, Deep learning is a special type of machine learning that mimic approaches of the neural system.It needs huge computation and hence gpu.

Hope there were some takeaways for you. In case, you have some suggestions or thoughts on this, I would like to hear from you!