EasyDeepLearn

What is the one-cycle policy?

medium

Answer

  • Smith's 'super-convergence' recipe: linearly ramp LR up from LRmax/25\mathrm{LR}_{\mathrm{max}} / 25 to LRmax\mathrm{LR}_{\mathrm{max}} over the first ~30% of training, then linearly decay back down.
  • Simultaneously modulate momentum in the opposite direction (high when LR is low, low when LR is high).
  • Enables very fast training with large max LR — regularizes via the 'LR-as-annealing-temperature' effect.
Check yourself — multiple choice
  • Constant LR throughout
  • Ramp LR up then down over one epoch cycle; momentum modulates inversely — very fast convergence
  • Same as cosine decay
  • Only used for RNNs

One-cycle: LR triangle with inverse momentum → super-convergence.

#schedules#learning-rate#training

Practise Deep Learning

214 interview questions in this topic.

Related questions