Skip to main content
Glite Dictionary
🚶 A quitter 🤖 A training method
dropout

dropout

2 5.5
A method used in training neural networks that randomly ignores some units
  • noun
  • /ˈdrɒpˌaʊt/
  • Specialized
translation icon : dropout
  • dropout rate

Examples

  • And the two really new things algorithmically are dropout and rectified linear units.

  • Future research should focus on empirically validating dropout prevention strategies.

    Academic text (2006)

  • Within this group, 3,196 student records were matched and 23 dropouts (24 events) were recorded, yielding a dropout rate of 0.75%.

    Academic text (2006)

  • Srivastava et al. (43) suggested a simple regularization technique named "Dropout" to solve the overfitting problem and to improve the performance of the NN.

    Academic text (2018)

  • During training, the dropout technique helps to prevent overfitting by randomly ignoring some neurons.

  • Using a dropout layer in a neural network can improve its performance on test data.

  • The model's accuracy increased once we applied a dropout method to reduce its complexity.

Surface Forms

dropout singular
dropouts plural

Morphology

dropout = drop (opaque) = out + drop

Although the term is historically based on the idea of units being 'dropped out' during training, the machine-learning sense is technical and not predictable for an average learner from the parts alone. [Postprocessing: moved 'out' from base_words to prefix]

Etymology

Dropout in machine learning borrows the same idea from drop and out, which mean 'to leave' or be 'left out', because during training the model temporarily 'leaves out' some small parts of the network to help it learn more generally.