Introduction to Unsupervised Learning.
*All the information and quotes provided below are retrieved from the link added at the end*
Unsupervised learning refers to the knowledge acquisition process carried out by machines. In contrast to supervised learning, we do not provide specific labels to train the model, but allow it to train itself from the environment.
‘It models the world by guessing’, the best example can be dancing. If you were in a zumba class for 1h, would you be able to guess the next moves?
Recognizing characteristics and creating categories is known as Unsupervised Clustering. To program computers to perform clusters, we first need few properties that characterise whatever it is we want to group.
To construct the model, we need to know the observations we will measure and place in a graph. Moreover, how do you want to represent the world? Somewhere, there are X number of groups calles K clusters but to know where they are we need an algorithm called K-Means Clustering.
K-Means Clustering algorithm is used to contrast and compare some observations to guess the number of clusters that exists in our data, as well as its averages.
We need the mean by adding all the datapoints and diving by the number of datapoint. First, our AI will predict; Second, the AI will learn and correct. If we are working with a daatbase made up of flowers, then each datapoint will represent one flower and will be assigned a label — for example ‘Type 1’ — . However, what we will see in our graph — which can be formed by ‘petal length’ in Y axis and ‘petal width’ in X axis — are merely random guesses made by the model.
The average of each cluster should be in the middle, so the model will continue correcting itself by calculating new averages until all datapoints are gathered in the centre.
We can repeat the process: asign new labels, adjust the averages, predict, and repeat. Something that needs to be taken into account is our own beliefs. Do you know that much about flowers that you’re able to tell whether the tulips used to train the models are from the same species? We need to be careful because we, as our model, learn from observation.
Ask yourself:
- What observations can I measure? ‘Does it have fur? How much green does it have? Is there a nose?’ For these predictions, the computer will measure the amount of RGB pixels in our images.
- How do we want to represent our environment? If we get just 2 variables, it may not be accurate. Representation learning helps us to identify patterns to understand and compare images with or without labels.
‘An autoencoder is a sort of neural network that uses the same basic principles of weights and biases to process inputs, pass data onto hidden neuron layers, and finally to a prediction output layer.’
With an autoencoder, our machine would get an image, have some representations in our hidden layers, and construct a full representation of our image. In unsupervised learnign ‘the world is our teacher’.
You can access the course here:
You can follow me on LinkedIn here:
You can read more about technology here:
