Please enable JavaScript.

Coggle requires JavaScript to display documents.

Deep Learning, Graphs, Input: Network, Big part of the class will be about…

- - - - grids
        
        e.g.
        
        images
        
        are represented as
        
        fixed sized grids
        
        fixed sized tensors
      - sequences
        
        e.g.
        
        speech
        
        text
- - - - complex
        
        Arbitrary size and complex topology
        
        No spatial locality as in images or text
        
        no notion of up, down, left or right
        
        No reference point - no fixed node ordering
        
        Often dynamic and multimodal features
- - - - Research
        
        in
        
        Representation Learning
- - - - Prediction
        
        at the level of
        
        new generated graph
        
        or
        
        level of property of whole graph
        
        Nodes
        
        Links
      - Activation Functions
      - Regularization e.g. dropouts
      - Graph Convolutions
    - - automatically learns the
        
        features
        
        in the
      - maps
        
        nodes
        
        to
        
        d-dimensional embeddings
        
        such that
        
        similar nodes
        
        1 more item...
      - Given a node u in a graph
        
        Goal is to learn the function f that takes the nodes and maps then into a d dimensional real valued vector
        
        called as
        
        feature representation
        
        of a given
        
        link
        
        graph
        
        node
        
        embedding
        
        representation
        
        such that
        
        similar nodes in the network
        
        are
        
        embedded close together