Unsupervised Machine Learning Explained For Beginners

welcome back to another machine learning explained video by assembly AI in this video we talk about unsupervised learning and one note we also have a video about supervised learning on our Channel and I highly recommend that you check this out as well and now without further Ado let's get started in the

last video we learned about supervised learning where the machine learning model learns by making use of labeled data now in unsupervised learning we have unlabeled data meaning we have data but we don't know the corresponding class or Target for the features this process is often times more complex but it has the huge advantage that it does not require upfront human intervention to label the data but if the data isn't

labeled then how is the model learning essentially the model is trying to figure out some kind of structure in the data and will extract the useful information from it a popular example is object recognition a common task in computer vision here the model finds similar looking objects and marks them without knowing its specific label each group has its own characteristics that the model learns to identify and afterwards we can very easily Define the associated label for this group now let's look at different applications of unsupervised learning one of the most

used applications of unsupervised learning is clustering which is the process of grouping a set of objects in such a way that similar objects fall into the same group imagine you are the founder of an e-commerce website for each user you could somehow collect the age and income but other than that the user stays Anonymous now you want to determine if the user is likely to buy your product or not this can be done by analyzing the data and identifying two possible clusters one cluster for the group that is interested to buy and another not interested of course this example is a little bit oversimplified and real world data often times does not look that simple that's why the model needs to use a more sophisticated way to find the Clusters there are different approaches for example connectivity models here we build models based on distance connectivity this method is also known as hierarchical clustering centroid models represent each cluster by a single mean Vector a common example is the K means clustering algorithm that represents each cluster with its mean Vector also called centroid in an iterative approach it finds the best centroids and assigns all the surrounding data points to this cluster and we have distribution models here clusters are modeled using statistical distributions another possible

application is outlier detection often times data sets contain a few samples that fall out of the typical range these samples are called outliers if we were to train a supervised model with this data set the outliers might confuse the model and lead to incorrect results that's why often times it makes sense to identify and remove these outliers with an unsupervised learning technique before moving on to a supervised learning task another application is latent variable modeling in statistics latent variables are variables that are not directly observed but are rather inferred from other variables that are directly measured mathematical models that aim to explain obser variables in terms of latent variables are called latent variable models a popular method in this category is principle component analysis or short PCA is commonly used for dimensionality reduction it computes the so-called principal components and then transforms the data by projecting all data points onto these new axes in the simple example we end up from a 2d space in a one-dimensional space on the line but of course this can be used for any number of Dimensions neural networks can

also be used for many different unsupervised learning tasks one particular neural network type is the auto encoder Auto encoders have an encoder part that encodes the image to get a low dimensional embedding of the image and after that it gets decoded Again by the decoder with a goal to reconstruct the original image as good as possible during this process there is no need to know the actual label of the image but what can this be used for two possible use cases are Den noising the image by encoding the image and learning meaningful features of the image to reconstruct it we can remove noise or other unwanted things from the image or we can do image and video compression where only the encoded representation is used to be sent over a network from one end and on the other end it gets decoded again to see the normal Image Auto encoders are indeed a very important Concept in deep learning so if you want to see a dedicated video for these let us know in the comments all right I hope

you now understand the concept of unsupervised learning if you enjoyed the video then please leave us a thumbs up and consider subscribing to our Channel also if you want to try ably a i speech to text API for free then go and grab your free API token using the link in the description below and then I hope to see you in the next video bye