02 Chapter

Deep Learning

Neural architectures that learn hierarchical representations from large datasets.

Deep learning models learn layered representations directly from high-dimensional data like images, text, and audio. The families below dominate modern AI, from Transformers in language to CNNs in vision and GNNs over relational structures.

  • Use Transformers for text, code, documents, multimodal AI, and modern recommendation/search pipelines.
  • Use CNNs for classic image tasks and GNNs when relationships between entities matter.
#AlgorithmBest forCommon fields
1Transformers Language, vision-language, sequence modeling
  • NLP
  • LLMs
  • search
  • chatbots
  • code
  • document AI
2Convolutional Neural Networks Images, spatial data
  • Computer vision
  • medical imaging
  • defect detection
  • satellite imagery
3Recurrent Neural Networks / LSTM / GRU Sequential data, older NLP/time-series systems
  • Speech
  • time series
  • sensor data
  • finance
4Autoencoders Compression, denoising, anomaly detection
  • Cybersecurity
  • manufacturing
  • medical imaging
5Diffusion Models Generative images/audio/video
  • Image generation
  • design
  • media
  • simulation
6Graph Neural Networks Graph-structured data
  • Drug discovery
  • social networks
  • fraud rings
  • recommendations
7Siamese / Contrastive Networks Similarity learning
  • Face verification
  • product matching
  • semantic search
8Neural Collaborative Filtering Recommender systems
  • E-commerce
  • streaming
  • ads
  • marketplaces