What is so interesting about machine learning?Why is machine learning considered the future?
Do you think a cognitive system will ever be able to beat the human experience?
These are the main questions that pop into the heads of the computer scientists working in the Machine Learning field.
What we are trying to do is that we are trying to train agents to perform some tasks; some of these are classification tasks, some are prediction tasks, and many more. What would enhance our agent? What would make our agent more accurate? The answer is simple, data. When we train our agent, what happens is that the agent is exposed to what is called a training set, the agent runs certain algorithms to detect features and classify a point into one of the defined classes.
Machine learning works the same way as the human brainWhen a child is born he initially knows nothing and then he is exposed to experiences in life when he learns. That is exactly what happens, the human experience here is the training set for the agent.
If you are sick and you are going to visit a doctor which would you rather visit?An experienced doctor or a less experienced one? I guess the answer will be the experienced one.
In any professional life, the more experience you get, the higher you are paid and the higher your position will be in the working chain. Experience is essential for working as you are exposed to many situations and when these situations are faced again you are less likely to make a wrong decision as you faced this problem before.
As we stated before, the machine works exactly as the human brain, and the more experience it gets, the better it will perform its task. For a machine to be more experienced or for an agent to be more accurate that means that the training set should be bigger. If we assumed that two agents run the same algorithm and one of them is trained with a bigger set than the other or we assumed that the training set of one agent is a subset of the other, which agent will have the higher accuracy in classifying or predicting? Of course, the one with the bigger training set will be more accurate.
The questions that must pop into your mind right now is,
Is there a point where adding data to the training set will not result in any enhancement in the accuracy of the agent?For example, if the training set consists of a huge number, for example 300 million data points and we tested the accuracy of the agent and we found out that the agent scored a score of 87% and then we increased the amount of data points in the set to double and we trained the model on the new 600 million data points. Would that mean that the accuracy of the agent will be doubled? Of course not. Would the agent have the same accuracy? The answer is also no.
Many researchers attempted to research this topic and to find out whether increasing the training data set have a settling point or not. After many attempts, they found out that there is no settling point reached, in other words, the more you add to your training set, the more accurate your agent will become and your agent will never have enough of learning. The relation between the accuracy of the agent and the size of the data set can be seen in the below graph.
What can be easily concluded from the above graph is the following:
- At the beginning by adding to the training set the accuracy increases significantly.
- At a certain point the increase in the accuracy is not as significant as in the period of the beginning of the training.
- Coming to a certain point, even doubling the size of the training set will result in a slight increase in accuracy.
In conclusion, as you increase the size of the training set, the accuracy will always increase however the increase is not as significant as the size gets bigger.
So, now we can address the main question
How IoT could unleash the real power of the Machine Learning?The IoT network will be considered the biggest source of data in the new era. Studies have shown that by 2020, 50 billion objects will be connected to the IoT network worldwide. Could you imagine the number of data points that could be used in training our agents only from the internet of things data?
Data science and analytics will be used to store the data collected from the devices of the Internet of Things in a way the data can be used easily by machine learning agents. The problem now is how to arrange and sort this data to be sorted on relevancy, this problem is now addressed and scientists are working on algorithms and agents to sort the data to be used later in an efficient, beneficial way.
So here come the answers to the questions raised in the beginning of the article:
What is so interesting about machine learning? Why is machine learning considered the future? Do you think a cognitive system will ever be able to beat the human experience?
The interesting thing about Machine learning is that soon, machines will do our jobs better than us. Why? Because they can focus on the job they are doing without any external factors and that is exactly why machine learning is considered the future. Imagine that you will have a doctor who knows every single case with doctors and patients from the whole of history. Imagine that he knows every single mistake that any doctor has made before. I guess this doctor will never make a mistake. Well that is your machine learning agent, disguised as a doctor and you can relate to this example in any other profession and I think now I answered the question. What do you think? Will a machine learning agent beat the human experience?