Have you ever found yourself under the spell of YouTube? You can’t help watching one video after another because YouTube just gets you and is recommending all the right things. What’s a person to do? Although it may seem like it is reading your mind, YouTube recommendations work according to a concept called machine learning (ML).
What is machine learning?
Machine learning is the process of teaching machines to learn and improve from experiences like humans do. While most of the machines that you use have been programmed to perform certain repetitive tasks, it is possible for machines to learn automatically to improve the outcome.
This may have conjured up images like these –
But hold your horses. Today machine learning looks much more like this –
All of this looks impressive but how does machine learning work?
‘Practice makes perfect’ works not only for humans but also for machines.
A machine gets better at performing a task through experience. In machine learning terminology, experience refers to large amounts of data.
Today, the world is moving online. This allows businesses to have a large amount of data about their customers which is called Big Data. It can be time-consuming for humans to derive insights from this data. So we can train the machine learning algorithm to do this for us which can be more efficient in the long run.
We can use different programming techniques to train a machine-learning algorithm to process this massive data and give us helpful insights. The algorithm evolves each time it goes through this process.
This training can majorly take place through 3 types of machine learning.
Types of machine learning
To train a machine learning algorithm, you can use two types of data – labelled and unlabelled. Labelled data, as the name suggests, is a training dataset that someone took pains to label. The labels include input and output. Unlabelled data, on the other hand, can be partially or completely unlabelled. Although it reduces the hassle of labelling, one needs to put more effort into the process. This data needs to be in a machine-readable format like text, images, numbers, sounds, and tables.
Depending on the type of data or no data, you can use supervised learning, unsupervised learning, or reinforcement learning.
1. Supervised learning (predictive modeling)
Remember when your books used to look like this? Turns out that machines can learn in a similar way. Supervised learning uses sample data which is labelled to train a machine learning algorithm. For example, if you want to teach an algorithm to recognise a dog (output), you can use different images of dogs as input. This way the algorithm gets an idea of the problem and solution. It is then able to figure out the relationship between input and output, and makes predictions. Next, you need to provide feedback on this prediction so that the algorithm can become more accurate. Once the algorithm has been trained, it can be put into practice with a larger dataset during which it improves with each iteration.
Supervised learning is used when you have correct answers in your past data and you want the machine to make future predictions based on that. So the machine learning expert needs to ensure that the data being used is varied and accurate.
One of the most common applications of supervised learning is email spam filters. The Gmail algorithm uses keywords such as free, winner, double your income, etc. to detect unwanted emails. It can identify these keywords as spam because it tests a given email against several other emails that have been marked as spam.
2. Unsupervised learning
Unsupervised learning uses unlabelled data. In the absence of labels, the algorithm tries to structure the data into clusters. It does so by discovering relationships between different data points. This makes it an important technique because much of the data that is produced today is unlabelled.
Unsupervised is commonly used in the e-commerce industry. For example, Amazon’s algorithm clubs together people with similar buying behaviour and gives ‘You may also like’ recommendations to each customer.
To summarise, supervised and unsupervised machine learning take the following route:
3. Reinforcement learning
In 2016, a machine called AlphaGo defeated an 18-time world champion of Go game. This was partly possible because the machine had been trained using reinforcement learning.
Reinforcement learning works without a dataset and is based on rewards. Here, the machine learning algorithm is an agent that chooses an action that will lead to rewards. Since there is no preexisting data to tell it which is the best course of action, the algorithm discovers it through trial and error.
In the example cited above, the machine played more games than any human could possibly play and discovered the best pattern to keep winning.
Advantages and disadvantages of machine learning
1. It can be used to perform complex tasks more efficiently. For example, it can be used in healthcare to diagnose diseases more accurately and in a lesser amount of time. In fact, Google Health has developed a model that was able to detect breast cancer more accurately than doctors.
2. It can predict trends and patterns from large amounts of data. This is currently being used in various ways like Amazon recommendations, Google search, targeted advertising on social media, etc.
1. If the data is not varied and accurate, it can create a biased model. For example, if a machine learning algorithm was trained to recognise scientists and it was only shown pictures of male scientists, then the algorithm would not recognise female scientists.
2. It can require a lot of resources. You might need a lot of data and computing power to process that data.
How to learn machine learning?
If machine learning has caught your attention and you want to know more about it, then here’s how you can go about it –
1. Brush up on your knowledge of statistics
Chances are that the first time you learnt statistics, it was not a joyride. However, this time, it’s going to be different because you are going to see it being put into actual use! Some concepts that you should be familiar with are probability, data structures, and sampling.
2. Make use of online resources
There are various free and paid online courses that you can take advantage of to learn about different concepts in machine learning. If self-learning is your strong suit, then you can take Google Machine Learning Crash Course. If you want to venture into machine learning with more guidance, you can check out Internshala’s Machine Learning training.
3. Choose a programming language
There’s no escaping this so you might as well get started. Python is one of the most popular languages for machine learning. Other languages include C/C++, Java, and R. You can start with one language and then unleash the geek in you to explore other languages.
4. Build your own project
Once you have understood the theoretical aspect of machine learning, it’s time to take it to the next level. You can begin with supervised models and move onto unsupervised models. You can also create a data analysis project to get a firm understanding of how data works.
5. Participate in competitions
You can compete in various challenges on Kaggle. The Titanic survivor prediction is one of the most popular machine learning challenges for beginners.
6. Do an internship
Even machines agree that there’s no better way to learn about anything than getting hands-on experience in it. You can apply to machine learning internships on Internshala, and learn the ropes of the trade.
If you have made it this far, give yourself a pat on the back! Now you know about one of the most revolutionary technologies in the world today. You can register for Internshala’s Machine Learning training now to start your journey!
Image credits: The New York Times (https://www.nytimes.com/2017/02/23/automobiles/wheels/self-driving-cars-standards.html)