Machine Learning is a method of training a computer to learn from data and allowing it to make predictions and decisions without being explicitly programmed. Think of it like this: instead of giving a computer a set of rules to follow, we give it a bunch of examples and let it figure out therules on its own.
For example, you developed a machine learning algorithm for a spam filter. This algorithm can learn to flag spam and non-spam (ham) emails based on examples flagged by users. The examples used to train the model are called the training set, and each individual example in the training set is called a training instance.
Now, the question is: why are we using a machine learning algorithm to solve this kind of problem instead of a traditional programming language?
First, you would examine the typical characteristics of spam. You might observe that certain words or phrases (such as “4U,” “credit card,” “free,” and “amazing” and “unnecessary images”) frequently appear in the subject lines. Additionally, you might notice other patterns in the sender’s name, the email’s body, and so on.
You would develop a detection algorithm for each of the patterns you observed, and your program would identify emails as spam if it detects a sufficient number of these patterns.
You would test your program and iterate through steps 1 and 2 until it meets your desired performance level.
Since the problem is not trivial, your program will likely become a long list of complex rules, making it difficult to maintain.
In contrast, a spam filter that uses Machine Learning techniques automatically identifies which words and phrases are strong indicators of spam by detecting unusually frequent patterns in spam examples compared to non-spam (ham) examples (given below figure). This approach results in a program that is shorter, easier to maintain, and likely more accurate.
Additionally, Moreover, if spammers notice that emails containing the word ‘4U’ have been blocked, they might start writing ‘For U’ instead of ‘4U’ in their emails. If we take a traditional programming language approach, you would need to add or update ‘For U’ as a flagged word. If spammers keep finding new ways to bypass the filters, you will need to add new rules indefinitely.
In contrast, a spam filter that uses Machine Learning techniques automatically detects that ‘For U’ has become unusually frequent in spam flagged by users, and it begins flagging these emails without requiring your intervention.
Types of Machine Learning Systems
There are many types of machine learning systems, so it’s helpful to group them into broad categories based on their similarities:
- Supervision during training:
- Supervised Learning: The algorithm learns from labeled training data to predict outputs based on input data.
- Unsupervised learning: The algorithm work with the unlabeled data to find pattern or relationship within data.
- Semi-supervised learning: This type involves a combination of labeled and unlabeled data during training.
- Reinforcement learning: The algorithm learns through interaction with an environment and receiving feedback in the form of rewards or penalties.
- Incremental Learning:
- Online learning: The system can learn incrementally on the fly, as new data becomes available.
- Batch learning: The system is trained on all available data at once and does not learn incrementally.
- Data Comparison Methods:
- Instance-based learning: The system compares new data points with known data points to make predictions.
- Model-based learning: The system detects patterns in the training data and builds a predictive model based on those patterns, similar to how scientists work.
These criteria can be combined in a variety of ways, and as a result, a machine learning system may belong to multiple categories according to these classifications. For instance, an advanced spam filter might very well qualify as an online, model-based, supervised learning system. It operates in real-time by effectively utilizing a deep neural network model that has been thoroughly trained on numerous examples of both spam and non-spam messages.