Transferring Knowledge across Learning Processes

[Blog post] [Paper]

Original PyTorch implementation of the Leap meta-learner (https://arxiv.org/abs/1812.01054) along with code for running the Omniglot experiment presented in the paper.

License

This library is licensed under the Apache 2.0 License.

Authors

Sebastian Flennerhag

Install

This repository was developed against PyTorch v0.4 on Ubuntu 16.04 using Python 3.6. To install Leap, clone the repo and install the source code:

git clone https://github.com/amazon/pytorch-leap
cd pytorch-leap/src/leap
pip install -e .

This installs the leap package and the Leap meta-learner class. The meta-learner can be used with any torch.nn.Module class as follows:

Require: criterion, model, tasks, opt_cls, meta_opt_cls, opt_kwargs, meta_opt_kwargs

leap = Leap(model)
mopt = meta_opt_cls(leap.parameters(), **meta_opt_kwargs)
for meta_steps:
    meta_batch = tasks.sample()
    for task in meta_batch:
        leap.init_task()
        leap.to(model)
        opt = opt_cls(model.parameters(), **opt_kwargs)

        for x, y in task:
            loss = criterion(model(x), y)
            loss.backward()

            leap.update(loss, model)

            opt.step()
            opt.zero_grad()  # MUST come after leap.update
    ###
    leap.normalize()
    meta_optimizer.step()
    meta_optimizer.zero_grad()

Omniglot

To run the Omniglot experiment, first prepare the dataset using the make_omniglot.sh script in the root directory. The p flag downloads the dataset, d installs dependencies and l creates log directories.

bash make_omniglot.sh -pdl

To train a meta-learner, use the main.py script. To replicate experiments in the paper select a meta learner and number of pretraining tasks. For instance, to train Leap using 20 meta-training tasks, execute

python main.py --meta_model leap --num_pretrain 20 --suffix myrun

Logged results can be inspected and visualised using the monitor.FileHandler class. For all runtime options see

python main.py -h

Meta-learners available:

• leap (requires the src/leap package)
• reptile
• fomaml
• maml (requires the src/maml package)
• ft (multi-headed finetuning)
• no (no meta-training)