Overview of Deep Learning Algorithms: Supervised, Unsupervised, and Semi‑Supervised Methods

In recent years, deep learning has become widely popular due to its high accuracy, effectiveness, efficiency, and ability to handle massive data.

Algorithms are the core of deep learning; they enable machines to work and process data similarly to the human brain, relying heavily on artificial neural networks modeled after brain structure and function.

Deep learning algorithms are mainly divided into three types: supervised, unsupervised, and semi‑supervised.

1. Supervised Deep Learning

Supervised deep learning algorithms predict target variables from a set of input features using labeled datasets, learning by minimizing the difference between predicted and actual labels. Typical applications include Convolutional Neural Networks (CNN) for image classification, which split input images into small features for analysis, and Recurrent Neural Networks (RNN) for sequential data such as time series or natural language, remembering previous inputs to inform later processing.

2. Unsupervised Deep Learning

When data are unlabeled, unsupervised algorithms discover patterns and structures, grouping similar observations. Examples are Autoencoders, which encode input data into a hidden representation and decode it back to reconstruct the original, and Generative Adversarial Networks (GAN), which consist of a generator that creates new data and a discriminator that tries to distinguish generated data from real data, enabling powerful data synthesis.

GANs have been used to generate realistic faces (e.g., “this person does not exist”) and to create new images from simple sketches, as demonstrated by NVIDIA Canvas.

3. Semi‑Supervised Deep Learning

Semi‑supervised methods leverage both labeled and unlabeled data to learn patterns, with examples such as Self‑Supervised Learning, which trains networks using only input data and internal supervisory signals (e.g., predicting the next video frame), and Transfer Learning, which fine‑tunes a pre‑trained network on a smaller target dataset, transferring knowledge from the larger source task.

For more in‑depth knowledge on deep learning algorithms and their practical applications, you can download the e‑book “Deep Learning Algorithm Practice”.