Deep Learning is one of the most fascinating branches of Artificial Intelligence (AI). Its goal is to enable machines to learn autonomously, drawing inspiration from how the human brain works.
Thanks to this discipline, we now have technologies capable of recognizing images, understanding natural language, or generating content with astonishing accuracy.
If you are just starting out in the world of AI, getting familiar with its vocabulary is the first step.
In this article, we present a comprehensive Deep Learning glossary, featuring clear explanations of the most widely used terms in the industry.
Deep Learning Glossary of Terms
Machine Learning
A subfield of AI that develops algorithms that learn from data. Instead of being explicitly programmed, models adjust themselves through examples and patterns.
Unsupervised Learning
A type of learning where the model works with unlabeled data, searching for natural patterns or groupings without external guidance.
Reinforcement Learning
A method where the model learns through rewards and penalties, exploring actions to maximize a cumulative reward. It is commonly used in robotics and video games.
Supervised Learning
A type of training where the input data is labeled. The model learns to associate examples with their expected outcomes.
Backpropagation
A key algorithm in deep learning. It calculates the gradient of the model's error with respect to its weights, propagating it backward to adjust the network's weights through an iterative process.
Clustering
An unsupervised technique that groups similar data into clusters, identifying hidden patterns.
Computer Vision
A field that enables machines to interpret images and videos. Facial recognition, medical diagnosis, and autonomous driving are some of the applications in this area.
Dataset
A collection of data used to train, validate, and/or test models. Its quality and diversity are crucial for final performance, since a model is successful not when it memorizes the training dataset (overfitting), but when it generalizes well to the validation dataset.
Deep Learning
A type of machine learning based on artificial neural networks with multiple ("deep") layers. It allows machines to learn complex representations of data, such as images, text, or audio.
Gradient Descent
A mathematical method used to minimize the error function. It adjusts the model's parameters in the opposite direction of the gradient, reducing the model's overall error.
Dropout
A regularization technique that randomly breaks connections between certain neurons during training to improve the model's generalization. It can be thought of as combining a set of classifiers within a single model.
Embedding
A numerical representation of categorical data or words in vector form, used to capture relationships within the data. Thus, two similar vectors represent two similar pieces of data. It is essential in Natural Language Processing (NLP) models because it captures semantic relationships in text.
Deep Learning Frameworks
Libraries such as TensorFlow, PyTorch, or Keras facilitate the construction, training, and deployment of neural networks. They allow developers to temporarily set aside the most theoretical concepts and focus on network design and training.
GPU (Graphics Processing Unit)
A processor designed to execute floating-point operations and handle parallel computations. It is essential for training Deep Learning models because, ultimately, these models are always based on matrix operations.
Artificial Intelligence (AI)
A field of computer science that seeks to create systems capable of performing tasks that typically require human intelligence, such as reasoning, perception, or decision-making.
Iteration, Batch, and Epoch
In an iteration, the model's weights are updated. During an iteration, the model can be trained on a single data point or a set of data points. This set is known as a batch. Finally, an epoch is equivalent to one full cycle in which the model has seen the entire training dataset.
Confusion Matrix
A tool that evaluates the performance of a classification model by showing correct and incorrect predictions against the actual and predicted classes.
Generative Models
Networks that can create new content, such as images, text, or audio. Notable examples include GANs (Generative Adversarial Networks) and Large Language Models (LLMs).
Neurons, Layers, and Weights
Neurons process information; layers group neurons at different levels of abstraction; and weights determine the influence of each connection on the model's final output.
Overfitting
This occurs when a model learns the training data too well, including the noise, and loses its ability to generalize to new data. It is like memorizing an exam word-for-word without actually understanding the material.
Natural Language Processing (NLP)
A branch of AI that works with natural language. It is primarily used for chatbots, machine translation, authorship detection, sentiment analysis, and general text processing.
Neural Network
A model inspired by the human brain, composed of interconnected units called neurons. Each connection has a weight that is adjusted during training to minimize errors.
Regularization
A set of techniques (such as dropout or L2 regularization) that prevent overfitting by penalizing excessive model complexity.
Transfer Learning
A technique that leverages a model previously trained on one task to improve performance on a similar, new task. It saves time and resources, and typically achieves higher performance because it utilizes external knowledge beyond the specific problem being solved.
Transformers
An architecture that uses attention mechanisms to process entire sequences of data simultaneously. They are the foundation of models like GPT, BERT, or T5 for Language Modeling.
DL as the Starting Point to Master Artificial Intelligence
Deep Learning has evolved from an experimental field into the core of modern artificial intelligence.
Understanding its terminology is essential for anyone looking to dive into this field or apply its benefits to business or research.
This glossary is just the starting point. Want to keep learning about Deep Learning?