Please enable JavaScript.

Coggle requires JavaScript to display documents.

SotA AI (Static neurons (activation functions (sigmoid, tanh, relu, leaky…

- - - - ImageNet Classification with Deep Convolutional
        Neural Networks
      - dropout regularization
  - - - https://github.com/lengstrom/fast-style-transfer
        
        checkpoints
    - - Howto
        
        https://medium.freecodecamp.org/how-to-play-quidditch-using-the-tensorflow-object-detection-api-b0742b99065d
        
        https://medium.com/nanonets/how-to-automate-surveillance-easily-with-deep-learning-4eb4fa0cd68d
- - - - regularization
  - - - train set
      - dev set
      - test set
    - - high bias = underfit
        
        adding features reduces high bias
        
        adding polynomial features fixes high bias
        
        decreasing lambda (less regularization) reduces high bias
      - high variance = overfit
        
        regularization reduces overfitting
        
        more training data reduces overfitting
        
        reducing the number of features fixes overfitting
        
        increasing lambda (more regularization) recudes high variance
      - start with a simple algorithm, analyse, then decide what to change
  - - - use of kernels when dimensions are small to average
  - - - F1 score