ResNet
Also called Residual Network
The network that let us stack layers really deep by giving each one a shortcut around itself.
Think of it like
Like an express lane on a highway — if a stretch of road adds nothing, traffic can just skip past it untouched.
Example
ResNet-50 is still a default image backbone in production vision systems years after its 2015 debut.
How it actually works
Very deep networks used to get worse, not better, because the training signal faded on its way back through dozens of layers. ResNet added “residual” shortcuts that let each layer learn only the change it needs, with the input passed through untouched by default. That single trick unlocked networks hundreds of layers deep and reshaped everything after.
For product teams
A durable, battle-tested vision backbone you can still ship on today.
For engineers
A CNN with identity skip connections so layers learn residual functions, easing gradient flow.
Related
- Residual Connection — The shortcut idea at its heart.
- Convolutional Neural Network — The image-model family it belongs to.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome