Deep Learning

Neural networks, and in particular their training through backpropagation, present several practical issues:

• they require labeled training data, while in contexts based on large amounts of data (big data) most of the data may not be categorized;
• the training time scales poorly both with the increase in network size and with the increase in the amount of data;
• the optimization can get stuck in local minima, making the network sub-optimal;
• the supervised training of deep models (networks with many hidden layers) is a particularly complex optimization problem.

To address these challenges, a branch of machine learning has been developed, known as deep learning, which leverages deep architectures and advanced techniques to improve learning effectiveness.

Humans tackle complex problems by breaking them down into sub-problems and multiple levels of abstract representation. Similarly, deep learning allows a system to learn hierarchical representations of data, directly mapping complex functions between input and output without relying on manually engineered features.

This approach makes it possible to generate high-level abstractions, often not expressible by humans, but more manageable by the computer.

With the growing availability of data and applications of machine learning, automatic learning techniques are evolving rapidly. The goal of deep learning is to build high-level representations of data through the use of multiple layers of nonlinear operations, as in Deep Neural Networks (DNN). ands, techniques that enable automatic learning are continually growing. The goal of deep learning is to create high-level representations of data through the use of multiple layers of nonlinear operations (DNN Deep Neural Network).