Install

git clone https://github.com/mpiorczynski/maml-from-scratch.git
cd maml-from-scratch
conda create -n maml-from-scratch python=3.9
conda activate maml-from-scratch
pip install -r requirements.txt

Stronly inspired by higher package MAML example.

Usage

./run.sh maml.py \
  --task-name 'sinusoid' \
  --k-shot 10 --k-query 10 \
  --meta-batch-size 16 --num-epochs 10 \
  --meta-optimizer 'adam' --meta-learning-rate 1e-3 \
  --inner-steps 1 --inner-learning-rate 0.01 \
  --seed 42 \
  --use-wandb \
  --log-interval 100 \
  --run-name 'sinusoid_10-shot_steps=1'

./run.sh maml.py \
  --task-name 'omniglot' \
  --n-way 20 --k-shot 5 --k-query 5 \
  --meta-batch-size 16 --num-epochs 10 \
  --meta-optimizer 'adam' --meta-learning-rate 1e-3 \
  --inner-steps 5 --inner-learning-rate 0.1 \
  --seed 42 \
  --use-wandb \
  --log-interval 100 \
  --run-name 'omniglot_20-way_5-shot_steps=5'

Citations

@inproceedings{finn2017model,
  title={Model-agnostic meta-learning for fast adaptation of deep networks},
  author={Finn, Chelsea and Abbeel, Pieter and Levine, Sergey},
  booktitle={International conference on machine learning},
  pages={1126--1135},
  year={2017},
  organization={PMLR}
}