Machine Learning systems that are trained with or without humans intervention

Machine Learning systems that are trained with or without humans intervention

First of all, we have to understand what machine learning really is. There are many definitions of it with some really good examples. As I have understood, it is something like you've programmed your computer (not explicitly) that open browser with some particular tabs that you mostly visit at a particular time. Every day it records your most visiting sites and learns through it. It learns with you from you. You can experience machine learning on the Facebook wall every time. The account you're interacting with the most will come first on your wall most of the time. Posts will come with timestamps but not in ascending order. Posts there always will be algorithm-based, you can choose the newest first but still, you can experience why your wall always becomes addictive.

Now as in the heading I have stated what actually we are going to train?

In Machine Learning we have types of systems in which some require human intervention or batch learning or instance-based learning or model-based learning.

Here we will be talking about a starting point and most of us will start machine learning with that only.

These are:

  1. Supervised: In typical supervised learning you will have value and labels. Like you want to predict the price of real estate where values with labels like price, number of rooms, area, locality will be provided. From the data, you will need to implement the desired model which can generate the most accurate price prediction. Here the output will be expected by the client.

Generally in a supervised model linear regression learning algorithm will be:

a. logistic regression b. decision tree c. random forest d. support vector machines e. k-nearest neighbors algorithms will be used.

  1. Unsupervised: here, the data will be the unlabeled training set in which learning algorithms will be:

a. Clustering: Suppose you're standing at the crossing, you will see kinds of automobiles, bikes, different aged people, animals, etc. Your eyes will cluster it in visualization and you'll have full understandings of the surroundings. This algorithm works that way, with clustering it will visualize highlighted semantics clusters.

b. Anomaly detection and Novelty detection: Major difference between these two is novelty detection algorithms expect to see only normal data in training while anomaly detection can tolerate data and learns if the data is normal or a deviation.

c. Visualization: Here you can feed a lot of complex and unlabeled data and it can output representation of your data in 2d & 3d which you can easily plot.

d. Association rule learning: The purpose of this algorithm is to dig into large amounts of data and find out relations between the attributes. Let me give you one example to understand. You are in the multiplex with your family and in the interval, you buy items from the stall. Salesperson after the day of experience will understand with what items people generally buy cold drinks the most. Like people buy popcorn and coke together mostly. Here this algorithm practices the same exercise.

  1. Semisupervised: It is a combination of supervised and unsupervised data, you can understand it that way. You have already experienced this one on Facebook. Remember, you upload a photo in the post, and on the faces, the block appears to ask you to tag a person, that way it labels the data for future use.
  1. Reinforcement Learning: This learning is my most favorite and excited about. If you remember, in 2017 a company called Deepmind used this model in the game of Go in which the model was able to defeat the world champion, it won 4 rounds out of 5.

    Here is the link:

It observes the environment first then selects action using policy, gets reward or penalty, updates policy with every reward and penalty, and iterates until it reaches the optimal policy.

So here we can say that robots, tesla cars most probably use this model. I will write a dedicated article on this one so stay tuned.

EndNote: I am not a genius and like you, I am also learning machine learning. I know python and have an interest in machine learning. If you find this article good enough, please share. You can find me on Twitter: [NitishJ2704][twitter.com/NitishJ2704]. I will continue writing articles and some interesting tweets on machine learning. Remember Machine Learning is the practice process, there is no endpoint