Start Simple

How to start simple with deep learning models?
Start Simple - Troubleshooting

Summary

  • Choose simple architecture:

    • LeNet/ResNet for images.

    • LSTM for sequences.

    • Fully-connected network with one hidden layer for all other tasks.

  • Use sensible hyper-parameter defaults:

    • Adam optimizer with a “magic” learning rate value of 3e-4.

    • ReLU activation for fully-connected and convolutional models and TanH activation for LSTM models.

    • He initialization for ReLU and Glorot initialization for TanH.

    • No regularization and data normalization.

  • Normalize data inputs: subtracting the mean and dividing by the variance.

  • Simplify the problem:

    • Working with a small training set around 10,000 examples.

    • Using a fixed number of objects, classes, input size, etc.

    • Creating a simpler synthetic training set like in research labs.