Delving Deeper into Machine Learning - What is Deep Learning?
As artificial intelligence becomes the foundational technology across all industries, the term deep learning
is frequently mentioned.
Deep learning is a subset of machine learning, evolving from networks with more profound and complex connections.
The Core of Deep Learning: Deep Neural Networks
Deep learning is based on the structure known as the Deep Neural Network
.
Traditional neural networks typically had about 1-2 layers, but deep learning consists of tens or even hundreds of layers, allowing for deeper and more precise data analysis.
Each layer transforms the input information into more abstract and complex features, ultimately deriving accurate results.
For instance, in image recognition, the first layer might identify simple features of the image like lines or colors, while the second layer might detect more complex features like edges or patterns.
Layers deeper within the network can then identify high-dimensional information such as human faces or specific objects.
Thanks to this deeper structured system, deep learning models can learn subtle features and abstract concepts from highly complex data, enabling them to make more sophisticated judgments.
Technologies Enabling Deep Learning
Deep learning requires vast amounts of data and high-performance hardware.
By vast amounts of data
, we mean large datasets, often comprising tens of thousands to millions of images in fields like image recognition.
For example, the ImageNet dataset, commonly used for image classification tasks, contains approximately 14 million images.
By high-performance hardware
, we refer to powerful computing capabilities, primarily achieved using GPUs (graphics processing units).
Notably, the famous natural language processing model, GPT-3, utilized around 10,000 NVIDIA Tesla V100 GPUs in parallel during its training, and the supercomputer equipped with this setup amounted to a cost exceeding $5 million.
Recent large-scale models like GPT-4.5 are reported to have employed thousands of NVIDIA's latest GPUs in parallel, with hardware costs alone estimated to reach hundreds of millions.
Nevertheless, using open-source libraries like Google's TensorFlow
or Meta's PyTorch
, meaningful deep learning models can be built even on general GPU environments owned by individuals.
These libraries have made it easier for anyone to engage in deep learning at a certain level without the need for high-end equipment.
In the next lesson, we will briefly summarize the differences between machine learning and deep learning.
Want to learn more?
Join CodeFriends Plus membership or enroll in a course to start your journey.