Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Deep Learning

Overview

Deep learning uses artificial neural networks with multiple layers to learn hierarchical representations of data. Unlike classical ML where features are hand-engineered, deep learning automatically discovers the representations needed for detection or classification.

Why Deep Learning?

graph TD
    A[Classical ML] --> B[Manual Feature Engineering]
    B --> C[Domain Expertise Required]
    C --> D[Limited to Known Features]
    
    E[Deep Learning] --> F[Automatic Feature Learning]
    F --> G[Learns from Raw Data]
    G --> H[Discovers Unknown Patterns]
AspectClassical MLDeep Learning
FeaturesManual engineeringAutomatic learning
Data requirementsSmall to mediumLarge
ComputeCPUGPU/TPU
InterpretabilityHighLow
Performance on unstructured dataLimitedState-of-the-art

Topics in This Section

TopicKey ConceptsInterview Frequency
Neural Network BasicsPerceptron, MLP, universal approximation⭐⭐⭐⭐⭐
BackpropagationChain rule, computational graphs⭐⭐⭐⭐⭐
Activation FunctionsReLU, sigmoid, GELU, Swish⭐⭐⭐⭐
CNNsConvolution, pooling, ResNet⭐⭐⭐⭐⭐
RNNs & LSTMsVanilla RNN, LSTM, GRU⭐⭐⭐⭐
Batch NormalizationLayer norm, group norm⭐⭐⭐⭐
DropoutTraining vs inference⭐⭐⭐⭐
OptimizersAdam, AdamW, learning rate schedules⭐⭐⭐⭐⭐
Transfer LearningFine-tuning, feature extraction⭐⭐⭐⭐
Attention MechanismSelf-attention, multi-head attention⭐⭐⭐⭐⭐

The Deep Learning Revolution

graph LR
    A[1950s: Perceptron] --> B[1980s: Backpropagation]
    B --> C[1990s: SVMs dominate]
    C --> D[2012: AlexNet - Deep Learning revolution]
    D --> E[2014: GANs, VAEs]
    E --> F[2017: Transformers]
    F --> G[2020s: LLMs, Foundation Models]

Key breakthroughs:

  • 2012: AlexNet wins ImageNet (GPU training, ReLU, dropout)
  • 2014: GANs (generative models), VAEs
  • 2015: ResNet (skip connections), Batch Normalization
  • 2017: Transformers (“Attention is All You Need”)
  • 2018-2020: BERT, GPT-2/3 (pre-training revolution)
  • 2022+: ChatGPT, GPT-4 (LLMs as general-purpose AI)

Cross-References