Shaping our future - discover how to use AI to build Predictive Models
_Have you ever heard about the types of machine learning algorithms and how they differ from each other in their approach?_
By Luís Mendes, Role — Feb 04, 2022 — 13 min. — AI, Machine Learning
Before we talk about Deep Learning, it is important to understand the types of machine learning algorithms and how they differ from each other in their approach. We will go from Supervised to Unsupervised Learning, passing by Reinforcement Learning and understanding how some of these types of algorithms already impact our day-to-day lives.
When are we in the presence of deep learning? Essentially, we call it deep learning when we use an artificial neural network with one or more hidden layers. The logic behind is essential to understand a different paradigm. We will start with the basics to show the principles on how a Machine Learning algorithm actually learns.
Feedforward architectures are the most common ones. Others like RNN – Recurrent Neural Networks take advantage of back propagation and aim to make predictions using time series sequences.
Inside RNN, there are some more advanced architectures solving the problem of the short term memory present on the RNN. This is the case of LSTM (Long Short Term Memory), which takes advantage of 3 gates that allow to mitigate the short term memory concern.
The potential use cases of Deep Learning are endless, and a lot of companies are now making huge progress using this type of technology. Frameworks like TensorFlow, created by Google Brain and used by Google itself, now play a big role on most of the new products launched by the main tech companies in the world.
The Machine Learning cycle and the understanding process of building a predictive model is crucial to be able to build a project for the long term. From gathering data to the deployment of the model, every stage takes a different time to complete. As a matter of fact, one is expected to spend little time building the model, 10% of the time, when in comparison to 70% one would need to dedicate to the process of gathering and preprocessing data.
In conclusion, building machine learning algorithms is very appealing and opens a wide range of new approaches to solve problems that were otherwise hard to even understand. Machine Learning will solve problems of very complex logic, but is it better to keep it away from simple problems we would be able to solve otherwise, at least while the field of Explainable AI doesn’t evolve to provide more satisfying results. The future is dependent on us, humans, being able to understand the rationale behind the decisions of machine learning algorithms. This way, not only will we be able to build more intelligent solutions, but also trust these models for making critical decisions for us to improve our future.