In recent years, machine learning (ML) is gaining more and more popularity, but what exactly is it? In this section, I will deep-dive so you will have a better understanding of what machine learning is, types and how it is used.
What is Machine Learning
The name “Machine Learning” initially originated from famous gaming researcher Arthur Lee Samuel. Samuel is the first person to bring self-learning programs into society. This remarkable discovery shortly laid the foundation for ML algorithms. In later years, raising popularity in Artificial Intelligence (AI) give birth to many innovations in the field of Computers and Automation. Similar definitions and usage of ML & AI created ambiguity in distinguishing these two fields. In fact, few beginners in this field often use AI and Machine Learning interchangeably, but the fact is that they are the same.
Artificial Intelligence is the integration of ML algorithms. Artificial Intelligence models are used to perform multiple tasks such as self-driving cars, humanoid robots, etc.
On the other hand, Machine Learning is used to accomplish only specific tasks like spam detection, Movie recommendation, and Image classification.
Actually, it is a sub-field of AI, the picture below clearly explains what I mean.
3 Segments of Machine Learning
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
Supervised Machine Learning
Supervised learning is the most commonly used technique. Many industries use supervised learning techniques to train machine algorithms.
In supervised learning, we supervise or teach the machine using labeled data. In other words, we show the sample data and tell the machine what the label is, likewise we do it for every sample in the data set.
Figure 1 explains the working of supervised Machine Learning.
In figure 1 Dataset consists of ‘n’ Labelled cat and dog images, each image is labeled with a tag. For instance, in figure 1 Image 1 is labeled as a cat. Likewise, there will be ‘n’ labeled images from 1 to n.
In supervised learning, the teacher holds the actual values for every corresponding image in the dataset. Similarly, the learning system will give predicted values for every corresponding image in the dataset. Once we got the image output values from the teacher and learning system error function will calculate the error between actual and predicted values.
Using the feedback error, the learning system will keep on updating its parameters (weights) to minimize the error value. Eventually, this process of learning parameters (weights) will help the Learning system to understand the model.
Unsupervised Machine Learning
In contrast to supervised, unsupervised learning doesn’t have any monitoring or teaching system. We let the machine to figure out its own model from unlabeled data.
In figure 2 the data set consists of ‘n’ images of cats and dogs but now this image is unlabeled (means we don’t mention their tags to learning system), for instance, in figure 2 now Image 1 is unlabeled it may be cat or dog. Likewise, there will be ‘n’ unlabeled images from 1 to n.
So, to tackle this unsupervised learning system uses the concept of clustering. Clustering is a grouping technique in which similar featured objects are grouped together. In the above example figure 2, we have two different categories (cats vs dogs). We design our learning system capable of clustering into two groups of images separately. We achieve this by the concept of feature selection in other words similarly in images are grouped together and dissimilar images group together.
Reinforcement Learning
Reinforcement learning is a special type of machine learning especially used for game theory, signals and system and control systems.
In reinforcement learning, whenever the environment is sending the information to the learning system, its output values and critic will comment on those values. For instance, if the output is appropriate for the machine then it will reward the learning system to perform the next task through actions. However, if the output is erroneous it punishes learning system and sends a single through actions to repeat the same task with the different data point.
Likewise, it checks every point in the environment to get the reward from the critic. This process of validating every single point will consume huge memory, time and require high computation speed. This is a major drawback of reinforcement learning. However, ML researches are still working on it to compensate for these issues and to create much more cost-efficient systems. Hopefully in the near future we able to see one.
What’s Next
Most Machine Learning algorithms fall under these categories, but current advancement created many more different algorithms, such as Semi-Supervised Learning.
Why and Where We use Machine learning
According to this Forbes article, Machine Learning is going to a common buzzword in Medical, Finance, Transportation, E-commerce and many more sectors. But why are popular companies focusing on ML nowadays? The answer is because of its capabilities.
