Wide residual networks

How this idea evolved

Suggested from citations. No curator has recorded what this idea builds on yet. These ideas from the same theme come from papers that Wide ResNet cites, directly or one step removed. Citation is a fact; the connection between the ideas is not verified.

• Very deep 3×3 ConvNetsVGG · 2014cited directly

  Stack many small 3×3 convolutions; depth alone drives accuracy.

• Residual connectionResNet · 2015cited directly

  Add a layer's input to its output so very deep networks stay trainable.

• Inception moduleGoogLeNet (Inception) · 2014cited directly

  Parallel convolutions of several sizes in one block, to go deeper and wider at fixed cost.