Machine learning is a branch of Artificial Intelligence, which deals with helping machines to learn best possible response of an unseen problem. Machine Learning term was coined in 1959 by computer scientist Arthur Samuel.Since then machine learning has come a long way. Machine learning models have demonstrated their productivity in last couple of decades.
Watch this complete course video on Machine Learning
Many complex problems were solved using machine learning algorithms. One such problem is predicting behaviour of stock exchange. For most of us stock exchange graphs look more random, forget about predicting its behaviour. But a duo of mathematician and computer scientist decided to use ML algorithms to solve this problem. David Siegel is a computer scientist and John Overdeck is mathematician who won silver medal at international math Olympiad. They founded Two-Sigma, a hedge fund company. Two-Sigma uses machine learning, artificial intelligence and big data algorithms to create trading strategies. Two sigma proved to be one of the most successful on the planet. Drawing lessons from the success story of Two-Sigma, other hedge funds also started to hire coders, computer scientist and mathematicians. Both David and John are now billionaires. Machine learning has applications in Gaming, Economics, artificial intelligence, mathematics and many other industries.
Read on why companies are using Machine Learning and how you can reap its benefits in our blog on the growing use of Machine Learning.
So, let’s understand what is machine learning and why it is so powerful?
Machine learning is an area of study of intelligent algorithms which try to infer a model from a set of labelled or unlabelled observations and uses this model to make predictions. Most important distinction between machine learning algorithms and other algorithms of artificial intelligence is its capability to predict. Machine learning algorithms just do not process the information or compute but they also test the output in natural setting. Effectiveness of machine learning models depend upon its successful predictions.
Let’s understand machine learning model by taking an example. Imagine a system which predicts asthma attacks. This system can use satellite data to know pollution levels, environmental condition, places etc. This models can further study the relationship between asthma attacks and its factors. This relationship knowledge and real time data can be used to predict time and place where a user may have asthma attack.
For the best of career growth, check out Intellipaat’s Machine Learning Course and get certified.
There are 3 kinds of machine learning models broadly:
- Supervised Learning:
Supervised learning based models are exposed to previously collect labelled datasets. Models then tries to infer knowledge from these datasets. These datasets are called training dataset. The knowledge model acquired, is then applied to predict a value against unseen datasets. Let’s take an example of a machine learning model based on supervised learning which can detect pictures of cat among many images. This model first has to be fed with many labelled images of cat. Model then draws knowledge about cat and how to identify the images. Various markers like colour, tail, size etc. can be used to identify the cat. This model then can be used to identify images of cat among many unseen images of cat.
Become Master of Machine Learning by going through this online Machine Learning course in Singapore.
- Unsupervised Learning:
Problem with supervised learning is that its scope is very restricted. Mostly its very daunting work to manually label thousands of images of cat. Now imagine if we want to create a model which should be able to differentiate between animals and birds. It would be out of scope of supervised learning. For these kind of problems unsupervised learning is used. Unlike supervised learning where model tries to predict a value, unsupervised learning model try to cluster or group the information. It tries to find commonalities among various data points. Feature vectors are used to guide the process of clustering. This cluster of information is then used to predict the possible group the value may exist. Main difference to understand between supervised and unsupervised learning is that in supervised learning neural network try to predict a value and in unsupervised learning neural network tries to predict a group. Let’s understand this situation by an example. Take an example of a system which helps to decide the leisure activity for office employees.This system have the data of hobbies and age group of all employees. This information can be used to answer many questions like what is average age of all employees, what is most popular hobby etc.
Go through this Artificial Intelligence Interview Questions And Answers to excel in your Artificial Intelligence Interview.
Watch this complete course video on Machine Learning Algorithms
- Reinforced Learning:
Biggest disadvantages with both supervised and unsupervised learning is that they both need dataset to work. Now it’s not possible to have dataset of every problem. In those set of problems where datasets are not available at all, model has top create its own dataset in order to acquire knowledge. Gaming bots are most popular use cases of reinforced learning. For example it’s not possible to feed dataset of every possible gameplay of the game of chess. Reinforced learning solves this problem by playing game of chess with itself and come up with best moves of chess. Reinforced learning uses something called policy network. By which neural network tries to find significant moves. Likelihood of an action to be a valid output move is identified by policy gradient. Reinforced learning is mostly used in game playing agents. Most famous game between AlphaGo Vs Lee is a good example of reinforced learning model. Now it was not possible in that scenario to come up with a dataset of moves that would defeat lee. Lee was never defeated before. In that scenario reinforced learning is best possible solution. AlphaGo basically created its own move by playing with itself. And over time it will only get better. Reinforced learning is a compute intensive process. It requires powerful machines to carry out its move simulations.
Machine learning is used everywhere. From a movie suggestion in Netflix to self-driving cars. There is hardly any industry which does not have any application of machine learning. Technological advancement cutting across industryverticals means cloud computing is facilitating the application of machine learning. In last couple of decades, compute capacity has also increased. Now small devices like smartphones have large memories and GPUs which are able to carry out compute intensive task.
If you have any doubts or queries related to Data Science, do post on Machine Learning Community.
This phenomenon has impacted every domain of computer science. Machine learning algorithms have also have taken advantage of this. Better ML algorithms are now possible to be created. More compute intensive tasks can be carried out by using ML algorithms. Within the next 5 years almost every industry would be actively deploying machine learning algorithms and models in order to understand the customer better, gain bigger market share and grow the revenues and profitability.
Interested in learning Machine Learning? Click here to learn more in this Machine Learning Training in Bangalore!