reptile-nichol-2018-first-order-approximation-maml
IN premise — summaries/2026/08/24/wiki-Meta-learning_computer_science.md
Created 2026-08-25T02:58:49+00:00
Reptile (Nichol et al., 2018) is a first-order approximation of MAML that avoids second-order derivatives, making it cheaper but less accurate.
Summary
Reptile is a practical shortcut for meta-learning: instead of computing the expensive "gradient of the gradient" that full MAML requires, it just steps in the direction of improvement using a single derivative, which cuts compute and memory costs substantially. The trade-off is that this simplification loses some precision, so it is best used when the system can tolerate a slightly less accurate adaptation in exchange for speed.