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.https://www.polistudy.me/enhttps://www.polistudy.me/deep-learning/ml-vs-deep-learning/https://www.polistudy.me/deep-learning/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.machine learningsupervised / unsupervised / reinforcement learninggeneralizationempirical risk minimizationbias–variance trade-offrepresentation learninghierarchical featuresuniversal approximationhttps://www.polistudy.me/deep-learning/perceptrons-to-neural-networks/https://www.polistudy.me/deep-learning/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.perceptronlinear separabilitymulti-layer perceptronuniversal approximationbackpropagationactivation functionscross-entropymaximum likelihood estimationgradient descenthttps://www.polistudy.me/deep-learning/training-and-overfitting/https://www.polistudy.me/deep-learning/training-and-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?generalizationoverfittingbias–variance trade-offcross-validationearly stoppingregularizationdropoutbatch normalizationweight initializationoptimizerslearning ratehttps://www.polistudy.me/deep-learning/recurrent-neural-networks/https://www.polistudy.me/deep-learning/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.recurrent neural networksequential databackpropagation through timevanishing gradientLSTMGRUbidirectional RNNseq2seqteacher forcingbeam searchattentionhttps://www.polistudy.me/deep-learning/word-embeddings/https://www.polistudy.me/deep-learning/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.autoencoderPCArepresentation learningone-hot encodingword embeddingsword2vecCBOWskip-gramnegative samplingGloVeanalogy arithmetichttps://www.polistudy.me/deep-learning/image-classification/https://www.polistudy.me/deep-learning/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.computer visiondigital image tensorcorrelationtemplate matchingtranslation invariancelinear classifierhinge losscross-entropynearest neighborhttps://www.polistudy.me/deep-learning/convolutional-neural-networks/https://www.polistudy.me/deep-learning/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.convolutionweight sharingsparse connectivityfeature mapCNN arithmeticpaddingstridepoolingreceptive fieldLeNetlatent embeddinghttps://www.polistudy.me/deep-learning/transfer-learning/https://www.polistudy.me/deep-learning/transfer-learning/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.receptive fielddata augmentationmixuptransfer learningfine-tuningfeature extractioncatastrophic forgettingconfusion matrixprecision and recallROCAUCequivariance vs invariancehttps://www.polistudy.me/deep-learning/famous-cnn-architectures/https://www.polistudy.me/deep-learning/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.AlexNetVGGNetwork in Networkglobal average poolingMobileNetdepthwise separable convolutionResNetInceptionDenseNetEfficientNetCNN visualizationhttps://www.polistudy.me/deep-learning/wsl-explanations/https://www.polistudy.me/deep-learning/wsl-explanations/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.batch normalizationdata preprocessingPCA whiteningmulti-label classificationsigmoid vs softmaxclass activation mappingGrad-CAMsaliency mapweakly supervised localizationResNetresidual learningskip connectionsInception1x1 convolutionhttps://www.polistudy.me/deep-learning/semantic-segmentation/https://www.polistudy.me/deep-learning/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.semantic segmentationinstance segmentationdense predictionfully convolutional network1x1 convolutiontranspose convolutionupsamplingdilated convolutionU-Netskip connectionsper-pixel lossweakly supervised learninghttps://www.polistudy.me/deep-learning/localization-object-detection/https://www.polistudy.me/deep-learning/localization-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.localizationbounding box regressionmulti-task learningpose estimationobject detectionIoUnon-maximum suppressionR-CNNFast R-CNNFaster R-CNNregion proposal networkanchorsYOLOSSDMask R-CNNfeature pyramid network