$\color{black}\rule{365px}{3px}$

“ImageNet Classification with Deep Convolutional Neural Networks” - 2012

Link to Paper:

proceedings.neurips.cc

Table of Contents

1. Introduction

$\color{black}\rule{365px}{3px}$

Contributions

• 1st Place in ImageNet Large Scale Visual Recognition Challenge(LSVRC) in 2012

• Activation function

  First model that introduced ReLU as an activation function. RELU had around 6 times faster performance than Tank and Sigmoid function to reach 25% error rate.

• Overfitting

  1. Data augmentation : flip augmentation and Crop augmentation along with adjusting RGB data(jittering) to increase the datasets.

  2. Dropout: rate of 0.5

     Dropout

  3. Local Response Normalization

     Local Response Normalization(LRN)

2. Dataset

$\color{black}\rule{365px}{3px}$