A Supplementary Material: Learning Compositional Rules via Neural Program Synthesis

A.1 Experimental and computational details All models were implemented in PyTorch. All testing and training was performed on one Nvidia GTX 1080 Ti GPU. For all models, we used LSTM embedding and hidden sizes of 200, and trained using the Adam optimizer [1] with a learning rate of 1e-3. Training and testing runs used a batch size of 128. For all experiments, we report standard error below.