AI.info
Deep architectures
Convolution, recurrence and attention read as one question: how information is routed through a network.
- Deep Learning Architectures as Information Routing
- Locality, Weight Sharing, and Equivariance
- Designing Convolutional Stages
- Bottlenecks, Depthwise Convolution, and Efficient CNN Blocks
- Residual, Dense, and Skip Connectivity
- Multi-Scale Hierarchies and Feature Pyramids
- Encoder–Decoder and U-Net Architectures
- Recurrent Networks as State Machines
- LSTM, GRU, and Gated Memory
- Sequence-to-Sequence Architectures
- Attention as Content-Based Routing
- Anatomy of a Transformer Block
- Position, Masking, and Sequence Geometry
- Encoder, Decoder, and Encoder–Decoder Transformers
- Efficient Attention and Long-Context Architectures
- State-Space and Selective Sequence Models
- Vision Transformers and Hierarchical Visual Attention
- Autoencoders, Bottlenecks, and Reconstruction
- Variational Autoencoders and Probabilistic Latent Variables
- Autoregressive Generative Architectures
- Normalizing Flows and Invertible Networks
- Generative Adversarial Networks and Learned Critics
- Diffusion and Score-Based Generative Architectures
- Latent Diffusion, Conditioning, and Guidance
- Graph Neural Networks and Message Passing
- Graph Attention, Graph Transformers, and Structural Encoding
- Multimodal Dual Encoders and Shared Embedding Spaces
- Multimodal Fusion and Cross-Attention Architectures
- Mixture-of-Experts and Sparse Conditional Computation
- External Memory and Memory-Augmented Networks
- Continuous-Depth and Implicit Neural Networks
- Architecture Evaluation, Ablation, and Scaling Evidence
- Deep Architecture Capstone: Design, Defend, and De-Risk a System