Womansplaining Supervised Learning
😊 Let’s begin…
Deep learning algorithms can be trained in several ways. These types of learning include: supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.
🔍 So what is supervised learning?
Supervised learning uses labeled data 🗂️ as input into the deep learning model. If successfully trained, the model, when given new input data, accurately predicts the label of the data. Labeled data 🗂️ simply refers to data with a pair of values: the item and the label.
👩🏫 Let’s look at some classification examples…
Let’s say we have a dataset consisting of dog 🐕 and cat 🐈 photos. We want the deep learning algorithm to classify the images as “cat” or “dog”. In this classification task, each image 🖼️ in the dataset would contain a label: cat or dog. If the algorithm is trained successfully, it will, when shown an image, output the correct image label.
As another example, let’s say we’re tasked with predicting whether patients have a brain 🧠 tumor based on MRI scans. To train the model with supervised learning, radiologists will generate a dataset of MRI images labeled either as “tumor” or “no tumor”. If trained well, the model will be able to predict the correct label for each of the MRI scans.
Datasets for supervised learning do not have to be images though. The data only needs to be labeled. A common example of supervised learning at work can be seen in spam detection. When you get an email 📧 in your Gmail, you can flag 🚩 it as spam, by clicking the octagonal 🛑 shape, or not spam, by leaving the email alone. Over time, Gmail learns to detect spam messages by looking at which emails you flag 🚩 as spam or not. Consequently, it automatically places spam messages into your junk 🗑️ folder.
📉 How is accuracy measured?
In supervised learning, a model should be able to correctly predict the label. A loss function is used to measure the accuracy of the deep learning model. The model continues to adjust and learn until a sufficiently small error 📉, or loss function, is achieved.
📈 These algorithms also perform regression.
Supervised learning algorithms can do more than simple classification tasks. Regression problems can also be handled with supervised learning algorithms. Here, the model tries to predict an output variable based on the input data.
As an example, let’s say we want to predict the temperature on a certain day. We’ll feed the network a variety of input data that might contain information about humidity, air temperature 🌡️, wind speed 🌬️, probability of rain 🌧️, etc. These variables will be associated with an output label: the temperature on that day 🌡️. If successfully trained, the model should be able to predict the temperature based on these input variables.
🔍 What supervised learning algorithms exist?
A range of supervised learning algorithms exist. The most common algorithms include: support-vector machines, linear regression, logistic regression, naive bayes, neural networks, and decision trees. More detail on the specifics of these algorithms will be presented in another post.
🗒️ So what did you just read?
Machines learn in a variety of ways. In supervised learning, labeled datasets 🗂️ are used to train algorithms, which should correctly predict some output, typically measured by a loss function 📉. Many different supervised learning algorithms exist and are used to solve classification or regression problems. In subsequent posts, I’ll delve deeper into other methods of learning, including unsupervised, semi-supervised, and reinforcement learning methods.