Inception-v1 / GoogLeNet (2014) | one minute summary
Machine learning inspired by the “we need to go deeper” meme
Published in
1 min readJun 17, 2021
The 2014 paper: “Going deeper with convolutions” from Google introduced the Inception module architecture, which has come to be known as Inception-v1 or GoogLeNet (which was the team-name when they won ILSVRC 2014).
- Why? The most straightforward way to increase CNN performance is to increase their size, but bigger sized models are prone to overfitting and require more computational resources.
- What? The Inception-v1 model is an efficient architecture for computer vision that introduced a few fancy techniques to achieve a “deeper” model while keeping a reasonable number of parameters (i.e. computational resources)
- How? Inception-v1 uses a combination of: 1. repeated inception modules that use convolution filters of different sizes in parallel to attend to multiple scales at the same time (along with 1x1 convolutions to reduce the number of parameters), 2. global average pooling to replace fully connected layers at the end of typical CNNs, and 3. auxiliary classifiers (for training only) to combat the vanishing gradient problem and help with regularization
Related articles: