Deep learning (DL) is a machine learning approach that uses neural networks with multiple layers to learn the features and rules contained in data.

The neural networks used in deep learning are called deep neural networks (DNNs).

Deep learning is used in a wide range of fields that handle data such as images, text, and speech.

1.5.1 Relationship Between Deep Learning and Machine Learning

Deep learning is a subset of machine learning and is built on artificial neural networks.

In conventional machine learning, people may decide in advance which properties of the data to focus on, and then feed those features into the model.

In deep learning, by contrast, the representations of the features needed for prediction are also learned from the data, as part of the process of computing outputs from inputs.

However, the loss function must still be chosen according to the task and the characteristics of the data.

1.5.2 Structure of Deep Neural Networks

A deep neural network consists of an input layer that receives data, hidden layers that transform the data step by step, and an output layer that produces the final result.

The word “deep” refers to the presence of multiple hidden layers between the input and output layers, through which the data passes in multiple stages of processing.

In the hidden layers, computations and transformations are applied to the incoming data, and features and patterns are extracted.

1.5.3 Learning Features Step by Step

One characteristic of deep learning is that it can represent the features of data step by step through multiple layers.

The early layers process relatively simple features close to the input, and the later layers that receive their results form more complex representations by combining multiple features.

In this way, the ability to learn everything from the input data to the feature representations needed for prediction as a single model is an important property of deep learning.

1.5.4 Training and Inference

Deep learning shares the same basic flow of model training and inference as other machine learning models.

In training, the discrepancy between the model’s output and the desired result is expressed as a loss, and the parameters are adjusted so that this loss becomes smaller.

Inputting new data into a trained model and computing the output is called inference.

When actually building a model, the process involves examining the data and preprocessing it, building and training the model, and then validating the results after training.

1.5.5 Representative Networks

Deep learning uses neural networks with different structures depending on the data and the purpose.

Convolutional neural networks (CNNs) are used to handle spatial features, such as those in images.

Recurrent neural networks (RNNs) are structures for handling data with an order, such as text and time series.

The Transformer is a structure that uses an attention mechanism to handle the relationships between parts of the input, and it is used in natural language processing and other fields.

1.5.6 Points to Note When Using Deep Learning

While deep learning can learn complex features, training a model may require large amounts of data and computational resources.

In addition, because processing is carried out through many layers and parameters, it can be difficult to understand why a model produced a particular result.

Deep learning is not always better suited than other machine learning methods. The method should be chosen according to the amount and type of data, the required performance, the computing environment, and other factors.

Summary

Deep learning is a machine learning approach that uses deep neural networks with multiple layers to learn features and rules from data.

Input data is transformed step by step through multiple hidden layers, and more complex feature representations are formed from the relatively simple features handled in the early layers.

The model adjusts its parameters through training, and after training it is used for inference on new inputs.

About This Article

References

• Simon J. D. Prince (author), Toshiaki Kurokawa (translator), Shinsō Gakushū o Rikai suru I: Kyōshi-ari Gakushū to Moderu no Kiso [Understanding Deep Learning I: Supervised Learning and Model Fundamentals] (Japanese edition)
• Mohammad Mustafa Taye, “Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Future Directions”
• Ibomoiye Domor Mienye, Theo G. Swart, “A Comprehensive Review of Deep Learning: Architectures, Recent Advances, and Applications”

(Last accessed: October 14, 2026)


*This article was compiled based on the author’s understanding, with reference to the literature listed above.