Concise exam notes — straight from the slides
Distilled from the actual lecture decks: only what you need to pass — definitions, formulas, the key intuition, and likely exam questions. No fluff. Each note also links its original slide PDF.
Why Deep Learning?
⏱ 5 min read · ⚡ concise
Hand-crafted features vs learned features, universal approximation, representation learning.
Perceptron to MLP
⏱ 5 min read · ⚡ concise
Neuron, activation functions, multi-layer perceptron, forward pass, the XOR problem.
Backpropagation & Computational Graphs
⏱ 7 min read · ⚡ concise
Chain rule, computational graphs, local gradients, the backprop algorithm, gradient flow.
Loss Functions & Optimization
⏱ 6 min read · ⚡ concise
MSE, cross-entropy, SGD, momentum, Adam, learning-rate schedules.
Regularization
⏱ 5 min read · ⚡ concise
Overfitting, L1/L2 weight decay, dropout, early stopping, augmentation, batch norm.
Initialization, Normalization & Debugging
⏱ 7 min read · ⚡ concise
Xavier/He init, vanishing/exploding gradients, BatchNorm/LayerNorm, gradient checking.
Convolutions
⏱ 5 min read · ⚡ concise
Local connectivity, filters, padding, stride, receptive field, pooling, output-size formula.
CNN Architectures
⏱ 7 min read · ⚡ concise
LeNet, AlexNet, VGG, Inception, ResNet + skip connections — and each one's key idea.
Transfer Learning & CNN Applications
⏱ 5 min read · ⚡ concise
Feature extraction vs fine-tuning, when to use which, detection, segmentation.
Sequence Modeling: RNNs
⏱ 5 min read · ⚡ concise
Hidden-state recurrence, BPTT, vanishing/exploding gradients, sequence task types.
LSTMs, GRUs & Gating
⏱ 5 min read · ⚡ concise
Cell state, forget/input/output gates, how gating fixes vanishing gradients, GRUs, BiRNNs.
Seq2Seq & Attention
⏱ 5 min read · ⚡ concise
Encoder-decoder, the information bottleneck, attention, alignment scores, context vector.
The Transformer & Self-Attention
⏱ 6 min read · ⚡ concise
Self-attention Q/K/V, scaled dot-product, multi-head, positional encoding, why it beats RNNs.
Transformers in Practice — BERT & GPT
⏱ 7 min read · ⚡ concise
BERT (masked LM, encoder-only), GPT (autoregressive, decoder-only), pre-training, BPE/WordPiece.