Baysian Model-Agnostic Meta-Learning

This repository contains implementations of the paper, Bayesian Model-Agnostic Meta-Learning (Jaesik Yoon and Taesup Kim et al., NuerIPS 2018). It includes code for running the sinusoid regression task described in the paper.

To comparison with MAML and Ensemble MAML, we implemented emaml_main.py and emaml.py. With the setting, num_particle=1, you can run MAML on emaml_main.py.

For the reinforcement learning experiments, plese see this repository.

Quick run

The instructions for the experiments described in the paper are at the top of each main script as followed.

• Examples of BMAML running script

# 5 shot sinusoid regression (|T|=100) with 10 particles
python bmaml_main.py --finite=True --train_total_num_tasks=100 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=5 --val_k_shot=5 --num_epochs=10000

# 10 shot sinusoid regression (|T|=100) with 10 particles
python bmaml_main.py --finite=True --train_total_num_tasks=100 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=10 --val_k_shot=10 --num_epochs=10000

# 5 shot sinusoid regression (|T|=1000) with 10 particles
python bmaml_main.py --finite=True --train_total_num_tasks=1000 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=5 --val_k_shot=5 --num_epochs=1000

# 10 shot sinusoid regression (|T|=1000) with 10 particles
python bmaml_main.py --finite=True --train_total_num_tasks=1000 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=10 --val_k_shot=10 --num_epochs=1000

• Examples of EMAML running script

# 5 shot sinusoid regression (|T|=100) with 10 particles
python emaml_main.py --finite=True --train_total_num_tasks=100 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=5 --val_k_shot=5 --num_epochs=10000

# 10 shot sinusoid regression (|T|=100) with 10 particles
python emaml_main.py --finite=True --train_total_num_tasks=100 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=10 --val_k_shot=10 --num_epochs=10000

# 5 shot sinusoid regression (|T|=1000) with 10 particles
python emaml_main.py --finite=True --train_total_num_tasks=1000 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=5 --val_k_shot=5 --num_epochs=1000

# 10 shot sinusoid regression (|T|=1000) with 10 particles
python emaml_main.py --finite=True --train_total_num_tasks=1000 --test_total_num_tasks=100 --num_particles=10 --num_tasks=10 --few_k_shot=10 --val_k_shot=10 --num_epochs=1000

We tested this code on the following versions of libraries (not every ones).

• TensorFlow v.1.14
• Numpy v.1.15.4

You can get figures similar to that in the paper with plotting.py by setting configurations.

Contact

Any feedback is welcome! Please open an issue on this repository or send email to Jaesik Yoon ([email protected]), Taesup Kim ([email protected]) or Sungjin Ahn ([email protected]).