Benefits in the industry: Explore your data immediately

Machine learning has become the basis of the majority of intelligent systems and a pillar of modern Artificial Intelligence. It allows a mathematical model to learn to do one or several tasks. This learning process can be done in many different ways, but they all have in common the need for data!

The new popular motto “Data is the new gold” shows how important data is today. Optimising the processing of these data and using them to create applications should be the primary objective of any organisation that has access to them.

The three main learning methods are :

supervised learning
unsupervised learning
reinforcement learning

These impose different ways of utilizing the data. In this blog, we will focus on the first two because they are the most widely used nowadays, due to their efficiency/resource requirement ratio.

Supervised learning

Supervised learning consists of training a system or model to perform a task based on already annotated, labelled or categorised data. To do this, the features or descriptors extracted from the data are provided to the input of the model and the output of the model is compared to the labels that are also provided. This comparison allows the parameters of the model to be adjusted so that they are better suited for the task. A certain amount of data is required to make these parameters converge to optimal values.

In short, for a model to learn with this method, the process needs data that have already been annotated or categorised.

The supervised learning process is then relatively simple: you require a model, annotated data (input and output data), and the parameters of the model will be adjusted according to these data.

The annotation/categorisation of the data requires a lot of time, effort and resources. It also sometimes requires expertise and is therefore dependent on access to qualified people. For example, the detection of cancerous or benign cells on an X-ray requires the expertise of doctors.

Unsupervised learning

Unsupervised learning, on the other hand, means that the model learns with input data without having or even knowing the corresponding outputs.

The process is therefore more complicated to set up than the supervised method because it requires more thought and effort.

For example, the K-Nearest Neighbors (KNN) method is an unsupervised method based on clustering. Clustering organises data into groups according to their proximity. Within each group the data should then have common characteristics. Therefore, all that is needed for this technique is the input data. The model does not need the data to be pre-categorised.

Another more graphic example is depth estimation using an unsupervised method for autonomous navigation. One of the methods used for this task, for example, is to take an image as input and to estimate the output by automatically comparing it with the sequence of images (also from the dataset). It adjusts the weights of the model based on this estimation. This model therefore does not need manual annotation or prior categorisation of the training data.

Contrary to supervised learning, unsupervised learning does not need to spend time, effort or resources on data categorisation. But it does require spending more time thinking about and setting up a process that allows training.

The unsupervised learning also offers the possibility of discovering new categories in the data. Indeed, the fact of not having a prior categorisation and of "letting the model do its thing" allows the discovery of new categories, even as they appear through time. This is less likely to happen when using supervised methods, as in this case, the categories are very often predefined and considered fixed.

If you want to go further: how we use unsupervised learning at Phedone

At Phedone, we exclusively use "modern" Artificial Intelligence to take full advantage of the constant and hyper-fast innovations of AI.

We offer our clients a significant increase in their capabilities:

By bypassing the long and expensive process of manual annotation/categorisation, therefore eliminating the need for staff training thus saving time and resources,
With ease of implementation,
Without bias, compared to manual creation, allowing to discover new trends or categories.

At Phedone, we make sure that you benefit from all the advantages of unsupervised services by preparing the implementation, job by job. We start with the Voice of the Customer, the patient or citizen. We integrate with support and ticketing tools.

The business experts do not have to intervene. The result is immediate, it is effective from the moment it is implemented and it will get even better with time.

Natural Language ProcessingModern artificial intelligence : What is unsupervised learning?

Découvrez l'IA avancée quiautomatise le traitement de vos retours clients

Restez connecté à Phedone