Deep Learning

Look inside the architectures behind modern AI. Slide convolution kernels over images, unroll recurrent networks through time, and watch attention decide which words matter.

Key concepts

Convolutional Networks

Shared kernels slide over images to detect local patterns, building up from edges to objects

Recurrent Networks

A hidden state carries information from one step of a sequence to the next

Attention

Each token builds its representation as a weighted mix of every other token

Depth

Stacking layers lets networks build complex features out of simple ones

Visualizations

Convolutional Neural Networks

Intermediate

Step a kernel across an image to build feature maps, then pool and stack layers

ConvolutionFeature MapsPoolingStride & Padding
Open

Recurrent Neural Networks

Intermediate

Unroll an RNN through time and watch the hidden state and vanishing gradients

Hidden StateUnrollingVanishing GradientsLSTM
Open

Self-Attention & Transformers

Advanced

See queries, keys and values combine into an attention heatmap for a sentence

Self-AttentionQuery/Key/ValueSoftmaxMulti-Head
Open