언어 선택 / Language Selection:
This repository contains implementations of various foundational deep learning models using PyTorch. The purpose of this project is twofold:
- Master PyTorch: Develop and refine skills in building deep learning models from scratch using PyTorch.
- Understand Core Architectures: Gain an in-depth understanding of CNN (Convolutional Neural Network) and Transformer architectures, which form the foundation of modern deep learning.
| Model | Colab |
|---|---|
| LeNet One of the earliest convolutional neural networks introduced by Yann LeCun. A foundational model for image classification tasks. |
 |
| VGG Network Known for its simplicity and use of very deep networks composed of 3x3 convolution layers. Explores the role of network depth in improving accuracy. |
|
| GoogLeNet (Inception) Introduced the Inception module, allowing for efficient computation with reduced parameters. Demonstrates the importance of designing models with computational efficiency in mind. |
|
| ResNet Introduced the concept of residual connections, addressing the vanishing gradient problem in deep networks. Paved the way for very deep architectures with hundreds of layers. |
|
| EfficientNet Scales models effectively across width, depth, and resolution using a compound scaling method. Combines efficiency and performance, achieving state-of-the-art results. |
|
| Transformer Revolutionized deep learning by replacing RNNs with self-attention mechanisms. Forms the backbone of many modern models like BERT and GPT. |
|
These models were selected to cover the evolution of deep learning architectures:
- LeNet introduces CNN basics.
- VGG and GoogLeNet explore architectural innovations for improving depth and efficiency.
- ResNet solves the challenges of training very deep networks.
- EfficientNet optimizes model scaling.
- Transformer breaks away from traditional CNNs and RNNs, showcasing the versatility of attention mechanisms.
- PyTorch Official Documentation: https://pytorch.org/docs/
- Original Papers: