Polimi · 2024/25
Deep Learning
From a single perceptron to convolutional backbones, recurrent nets and word embeddings — the full Politecnico di Milano Deep Learning course, rebuilt as an interactive, exam-focused study guide on learned, hierarchical features.
12 chapters~11 h reading 49 past-exam questions
After this course you can
- Build a multi-layer perceptron and derive backpropagation from the chain rule
- Control overfitting with dropout, batch normalisation, weight decay and learning-rate schedules
- Compute convolutional layer geometry and compare the AlexNet, VGG, ResNet and MobileNet design choices
- Choose between feature extraction, fine-tuning and training from scratch for a new dataset
- Explain LSTM and GRU gating and the vanishing-gradient limit it answers
- Contrast segmentation and detection architectures and evaluate detectors with IoU and mAP
Syllabus
-
ML vs Deep Learning
By the end you can place any problem in the right learning paradigm, state precisely what "generalization" means, explain what deep learning actually changed — learned features instead of hand-crafted ones — and say honestly why it took off when it did.
low exam weight. 40 minutes reading. 1 past-exam question. -
From Perceptrons to Neural Networks
By the end you can explain why a single perceptron can't learn XOR, trace backprop through a small net, and derive — not memorise — why regression uses MSE and classification uses cross-entropy.
medium exam weight. 55 minutes reading. 3 past-exam questions. -
Neural Networks Training & Overfitting
By the end you can diagnose under- vs over-fitting, choose a validation scheme, and sort every training trick into the two buckets the exam keeps asking about: does it improve GENERALIZATION, or training PERFORMANCE?
high exam weight. 60 minutes reading. 7 past-exam questions. -
Recurrent Neural Networks
By the end you can explain why a vanilla RNN forgets, walk an LSTM cell gate-by-gate, and reason about which sequence architecture (stacked, bidirectional, seq2seq) fits a task — the exact judgements the exams test.
high exam weight. 65 minutes reading. 9 past-exam questions. -
Word Embeddings
By the end you can explain how an autoencoder learns a representation with no labels, why one-hot words are useless for meaning, what task word2vec actually trains on (self-supervised but NOT an autoencoder), and where its analogy arithmetic comes from.
high exam weight. 50 minutes reading. 4 past-exam questions. -
Image Classification
By the end you can explain why a linear classifier on raw pixels is literally template matching by correlation, why that's too weak, and why pixel distance fails — the motivation that makes CNNs necessary.
medium exam weight. 45 minutes reading. 2 past-exam questions. -
Convolutional Neural Networks
By the end you can read any CNN's model.summary() — predict every layer's output shape and parameter count — and explain why convolution, not a dense layer, is the right tool for images.
high exam weight. 55 minutes reading. 9 past-exam questions. -
CNNs — Transfer Learning & Data Scarcity
By the end you can decide — given how much data you have and how similar it is to ImageNet — whether to freeze, fine-tune, or train from scratch, and at what learning rate, plus how augmentation and receptive fields fit in.
low exam weight. 50 minutes reading. 1 past-exam question. -
Famous CNN Architectures
By the end you can read each landmark architecture as a DESIGN LESSON — depth (VGG), parameter collapse (GAP), and efficiency (MobileNet) — rather than memorising specs.
low exam weight. 50 minutes reading. 1 past-exam question. -
WSL, Explanations & Advanced Architectures
By the end you can read BatchNorm's parameters off a model.summary, explain how a classifier localises objects it was never taught to localise (CAM), and reason about why residual skips ADD while U-Net skips concatenate.
medium exam weight. 55 minutes reading. 2 past-exam questions. -
Semantic Segmentation
By the end you can explain how a classifier becomes a dense predictor (FC → 1×1 conv), how transpose convolution learns to upsample, and exactly why U-Net concatenates its skips — the facts the segmentation questions hinge on.
high exam weight. 50 minutes reading. 5 past-exam questions. -
Localization and Object Detection
By the end you can read an output head and name the task (regression vs detection), compute IoU, and trace the R-CNN family's evolution — why each version moved more of the pipeline into the network.
high exam weight. 55 minutes reading. 5 past-exam questions.