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

“Going deeper with convolutions” - 2014

Link to Paper:

arxiv.org

Table of Contents

1. Introduction

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

Contributions

• 1st Place in ImageNet Large Scale Visual Recognition Challenge(LSVRC) in 2014

  (top-5 test error of 6.67%)

• Around 6-million parameters used which is 9x fewer than AlexNet of 2012 and 20x fewer than the second place model VGG-16

• New architecture in seek of reducing parameters by introducing inception modules with

  $1 \times 1$ conv blocks, still making the network significantly deeper.

<aside> <img src="/icons/light-bulb_blue.svg" alt="/icons/light-bulb_blue.svg" width="40px" /> Inception Module-v1

</aside>

<aside> <img src="/icons/light-bulb_blue.svg" alt="/icons/light-bulb_blue.svg" width="40px" /> Auxiliary Classifier

</aside>