The first question that I was asked a month ago during a job interview for a position as a Data Scientist was: “What is Machine Learning, and what is Deep Learning?”.
I have been working as an analyst in credit risk management for over five years now. In this time, I have implemented many models from scratch. My answer to these two questions immediately proved to the interviewers that I would not be suitable for this position, so they never asked me another question…
The point here is not to complain about being asked simple questions during job interviews. But instead, it’s because of how little knowledge employees tend to have when approaching topics such as these two. These points out our lack of understanding, from which we can learn a lot about Machine Learning itself.
Machine Learning gives computer systems the ability to know without programming. In other words, it enables computers or machines to have the ability to learn by themselves when exposed to new data and use this learning to make predictions without being programmed rationally for these tasks. This emulates how humans work: we gather knowledge from examples through exposure – hence why it is called ‘learning’ – and apply it whenever possible – hence why it is called ‘learning’.
In short, Machine Learning provides computers with the capability of adapting to an environment based on experience instead of explicit programming. At its core, this means that, through various learning models, computers can come up with their own rules to classify data without any user guidance or pre-programming.
When does one use Machine Learning?
Machine Learning is simply an algorithm that emulates human learning. Therefore, the answer to when you should use it is always: “when you would do so in real life.” But there are some elementary examples where you could immediately see the applicability of using Machine Learning algorithms to solve problems. These include but are not limited to:
- Detecting spam/ham in emails
- Classifying objects in an image (e.g., cats vs. dogs)
- Identifying handwritten digits (0-9)
- Determining someone’s age based on their facial features
- Predicting the default of a loan applicant based on their credit report
These are all examples where Machine Learning can be used to solve problems. For example, one would use the last example (i.e., predicting someone’s age) via using Deep Learning, an advanced type of Machine Learning that gives computers the capability to learn by themselves without any human guidance or pre-programming. They do this by providing what is known as an artificial neural network (ANN), which loosely mimics how our brains work. In turn, ANNs process information through interconnected nodes – neurons – each with different linear algebraic equations/weights and biases. This allows them to create their own rules from examples without being programmed rationally. The ANN is then trained against a training dataset to identify patterns before extrapolating this learning to new data that it has not yet seen during the learning phase. This learning process is iterative between the model and the training data until the error is minimized. Hence, it is called ‘deep’ learning as there are many layers of mathematical relationships between an input object and its output classification.
Machine Learning has grown out of the intersection between the fields of Statistics and Computer Science. It is a highly active area in computer science that takes on more significance as data, especially massive amounts of data, become available to everyone.
Machine learning aims to enable systems that automatically learn from their experiences to improve performance based on examples rather than being explicitly programmed for this purpose by a human programmer. This may seem very similar to artificial intelligence, which tries to create computers that can solve problems with intelligent behavior without telling them what we want them to do – but actually, there are important differences that you will understand if you read further.
The automatic way machine learning works might be compared with the work done by a teacher who grades student assessments and uses the grades as feedback to improve the way she teaches. Unlike a human teacher, however, who might (for example) be tired on one particular night and give lower marks than usual; machine learning systems don’t get bored or frustrated, and they will always try to learn from their experiences, whether that is what we want them to do or not! Fortunately, they are benign in their actions – unless they are programmed otherwise.
Machine learning and artificial intelligence
Machine Learning (ML) is also sometimes called ‘Artificial Intelligence. Artificial Intelligence has existed for many decades now, and it produces valuable tools such as speech recognition software, search engines, and autonomous vehicle control systems. All these applications have something in common: They are all about automation. A programmer writes some rules or information about what is needed. The computer program then finds out how to do it automatically. The critical point here is that the programmer needs to tell (write) the computer what to do; he doesn’t get any feedback on whether this was successful until he tries it! In machine learning, as we shall see, this situation is reversed – we let the program run and collect the data from real-life experiences, which are used as a source of feedback. We now know that machine learning can be beneficial for sorting through all kinds of data available efficiently and making predictions about what might happen next. To know more check RemoteDBA.com
Machine Learning got its start during World War II with research into military applications such as recognizing enemy planes from radar signals. But the broader study of this area really took off in the 1980s with a vital breakthrough based on a concept called ‘Bayes Theorem’.
This theory’ gives us a reliable way to calculate how likely it will happen, given that some other event has happened. Visit Answer Diary. For example, suppose I want to know how likely it is that someone who purchases cigarettes is also a smoker. From my experience so far, I have found out the following: Among people who buy cigarettes, 95% are smokers, and 5% are non-smokers. Now for my new customer, if they purchase cigarettes, what would you say is the probability that they are smokers?