Please enable JavaScript.

Coggle requires JavaScript to display documents.

PMLDL, Semantic Segmentation, Recommendation Systems - Coggle Diagram

- - - - Covariate
        
        изменилось распределение данных
      - Label
        
        меняется распределение для конкретного label
        :smiley:
      - Concept
        
        изменились зависимости которые мы estimated
- - - - Back Propagation Through Time
        
        Vanishing of the gradients
      - LSTM
        
        hidden state
        
        cell state
      - GRU
        
        hidden state
  - - - Encoder-Decoder Attention
      - Self-Attention
        
        Multihead Self-attention
        
        Masked Multihead Self Attention
        
        Query, Key, Value
        
        In encoder equal
- - - - Inception Module
    - - Skip Layers
  - - - for multiple objects
        
        Instance segmentation
      - RCNN
        Region CNN
        
        Fast RCNN
        
        Faster RCNN
  - - - Denoising Autoencoders
      - Sparsity Loss
      - Contractive autoencoder
      - Stacked convolutional autoencoder
      - Winner-Take-All Autoencoder
      - Adversarial Autoencoder
      - Variational Autoencoder
  - - - Conditional (with class)
      - Semi-Supervised
      - InfoGAN
    - - Noise
      - AdaIN
- - - - Node-level, does not change nodes, only their dimensions
    - - Graph-level, can change everything
        generates smaller graph
  - - - Many parameters
      - Can not use node attributes
      - Are inherently transductive
  - - - Deep Walk
        
        wanna
        predict
        neighbours by features
      - Node2vec
        
        simulate n biased
        random walks of fixed length starting from every node
  - - - Message
      - Aggregation
    - - Pros:
        
        Shift-invariance
        
        Locality (Local similarity)
        
        Hierarchy (can be learned)
      - Operations
        
        Convolution
        
        Pooling