Types of Models in Machine Learning
Related to course: Machine Learning
Answer
To perform a specific task, each model is trained on data. This is necessary for recognizing patterns, for classification, and for producing results or predictions. Data is the foundation. Only with data can patterns or dependencies be discovered.
There’s a classic approach to programming where a programmer writes algorithms for how a program should operate. These are based on specific conditions, for example: if a particular event occurs, certain actions need to be performed. And there can be a vast number of such branches. This is how the operating algorithm of a robotic vacuum cleaner is built: upon encountering a wall, it needs to turn, say, to the right, then move forward, turn around, and so on.
Machine learning differs in that the data analysis algorithm is not explicitly programmed. In Machine Learning, a large volume of labeled data is given, based on which the model is then trained (humans often play a role in this training process as well). In other words, a classic programmer writes rigid conditions, while in machine learning, the algorithmic connections are fine-tuned, essentially paving the way for the model to act.
Important Note: The output of neural networks is not always a precise value but rather a probability. An algorithm written by a programmer operates on the principle of 3+2=5. However, a calculator based on a neural network, given the same question, would output an answer like: “with a 99% probability, the answer is 5” or “the answer is close to 4.99.”
The task of the person training the model is to bring the result closer to the most accurate value possible.
Models are created using algorithms:
— Linear Regression: This displays the relationship of one set of data to another. The values are plotted on a graph, and the relationship is shown as accurately as possible.
— Decision Trees: This type of model resembles a test. By answering a series of questions, the machine goes through a specific path and makes a decision. Each new question-answer forms a branching path, which is why the model is called a decision tree.
— Neural Networks: This algorithm can be compared to a huge panel with many lights. For example, you give the system the task of determining what is depicted in an image, say, a cat. It processes the data you provided and recognizes whiskers, paws, and a tail. In a simplified sense, several “lights” illuminate, and the model concludes that it’s a cat. Of course, our example is quite basic. In reality, the system will deal with thousands of data points and include thousands of “lights” in its processor.
To ensure machines provide accurate results and are effective, they need to be trained correctly: avoid overfitting or underfitting, and configure the parameters correctly. It’s essential to remember that any mechanism—from a diagnostic machine in an operating room to a home coffee maker—needs to be monitored and improved from time to time. As we can see, machine learning engineers have a lot on their plate.
There’s a classic approach to programming where a programmer writes algorithms for how a program should operate. These are based on specific conditions, for example: if a particular event occurs, certain actions need to be performed. And there can be a vast number of such branches. This is how the operating algorithm of a robotic vacuum cleaner is built: upon encountering a wall, it needs to turn, say, to the right, then move forward, turn around, and so on.
Machine learning differs in that the data analysis algorithm is not explicitly programmed. In Machine Learning, a large volume of labeled data is given, based on which the model is then trained (humans often play a role in this training process as well). In other words, a classic programmer writes rigid conditions, while in machine learning, the algorithmic connections are fine-tuned, essentially paving the way for the model to act.
Important Note: The output of neural networks is not always a precise value but rather a probability. An algorithm written by a programmer operates on the principle of 3+2=5. However, a calculator based on a neural network, given the same question, would output an answer like: “with a 99% probability, the answer is 5” or “the answer is close to 4.99.”
The task of the person training the model is to bring the result closer to the most accurate value possible.
Models are created using algorithms:
— Linear Regression: This displays the relationship of one set of data to another. The values are plotted on a graph, and the relationship is shown as accurately as possible.
— Decision Trees: This type of model resembles a test. By answering a series of questions, the machine goes through a specific path and makes a decision. Each new question-answer forms a branching path, which is why the model is called a decision tree.
— Neural Networks: This algorithm can be compared to a huge panel with many lights. For example, you give the system the task of determining what is depicted in an image, say, a cat. It processes the data you provided and recognizes whiskers, paws, and a tail. In a simplified sense, several “lights” illuminate, and the model concludes that it’s a cat. Of course, our example is quite basic. In reality, the system will deal with thousands of data points and include thousands of “lights” in its processor.
To ensure machines provide accurate results and are effective, they need to be trained correctly: avoid overfitting or underfitting, and configure the parameters correctly. It’s essential to remember that any mechanism—from a diagnostic machine in an operating room to a home coffee maker—needs to be monitored and improved from time to time. As we can see, machine learning engineers have a lot on their plate.