ImageNet Classification with
Deep Convolutional Neural
Networks
Feb 13, 2021
HeeDae Kwon
Contents
1. Introduction
2. The dataset
3. Methods
4. result
Introduction
- Collect large date
=> =>
- Learn model powerfoul models
=> Methods
- Use better techniques
=> Methods
The Dataset
Method
Overall structure
Method
- faster than sigmoid , tanH
ReLU
Method
GTX580 x 2
1.2millions
Local Response Normalization
Training on Multiple GPU
Method
• Overlapping Pooling
Method
256x256 => 224x224
Altering intensity RGB Channels
Data augmentation
Method
Dropout
Details of learning
Batch size = 128 exmples
Momentum = 0.9
Weight decay = 0.0005
Learning rate = 0.01
Results
• 감사합니다

Image net classification with deep convolutional neural networks