Posts tagged deep-learning
Strided Convolutions
How stride changes the way a filter moves across an image, the floor in the output-size formula, and when striding beats pooling for downsampling.
27 Feb 2023 · 3 min read
Pooling Layers in Convolutional Neural Networks
How max and average pooling downsample a representation, why they have no weights, and why modern architectures increasingly do without them.
27 Feb 2023 · 4 min read
One Layer of a Convolutional Network
A single convolutional layer written out as an equation, mapped onto the dense-layer form it replaces, with the parameter count it actually costs.
27 Feb 2023 · 3 min read
Convolutions Over Volume
How convolution extends from flat images to 3D volumes, why a filter always spans every input channel, and what actually makes a layer deep.
27 Feb 2023 · 3 min read
Deep Learning Computer Vision Advancements and Exciting Applications
Why fully connected layers cannot read images, what ImageNet error rates did between 2010 and 2015, and the two ideas convolution replaces them with.
26 Feb 2023 · 4 min read
Deep Q-Learning (DQN)
Why tabular Q-learning breaks down in continuous state spaces, and how DQN replaces the Q-table with a neural network and a replay buffer.
22 Feb 2023 · 1 min read