Posts tagged coursera

1x1 Convolutions and Network in Network

Why a 1x1 filter is not a no-op, how it becomes a per-pixel fully connected layer across channels, and the bottleneck it makes possible.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
31 Mar 2023 · 3 min read

Why ResNets Work

Four explanations for the skip connection — easy identity, gradient highways, a smoother loss surface, and implicit ensembles — and what each one gets right.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
30 Mar 2023 · 4 min read

ResNets and the Residual Block

The degradation problem that deeper plain networks hit, the skip connection that fixes it, and the bottleneck block that makes 152 layers affordable.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
29 Mar 2023 · 4 min read

Classic Networks: LeNet-5, AlexNet and VGG-16

Three architectures seventeen years apart, traced layer by layer, and the 3x3 stacking argument that made VGG's uniformity worth copying.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
28 Mar 2023 · 4 min read

A Simple Convolutional Network Example

A complete ConvNet for 39x39 RGB images traced layer by layer, with every shape derived from the formula and every parameter counted.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
27 Mar 2023 · 4 min read

Unleashing the Power of Convolutional Neural Networks

The 14-million-parameter comparison that justifies convolution, worked in full: parameter sharing, sparse connectivity, and what each one buys.

• cnn
• computer-vision
• coursera
• deep-learning
• Written in EN

Daniel Du
07 Mar 2023 · 4 min read