Generative Augmented Flow Networks

Authors: Ling Pan, Dinghuai Zhang, Aaron Courville, Longbo Huang, Yoshua Bengio

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Based on extensive experiments on the Grid World task, we demonstrate the effectiveness and efficiency of GAFlow Net in terms of convergence, performance, and diversity of solutions. We further show that GAFlow Net is scalable to a more complex and large-scale molecule generation domain, where it achieves consistent and significant performance improvement.
Researcher Affiliation Academia Ling Pan1,2, Dinghuai Zhang1,2, Aaron Courville1,2,4, Longbo Huang3, Yoshua Bengio1,2,4 1Mila Qu ebec AI Institute 2Universit e de Montr eal 3Tsinghua University 4CIFAR AI Chair
Pseudocode Yes Algorithm 1 Generative Augmented Flow Networks.
Open Source Code Yes The code is publicly available at https://github.com/ling-pan/GAFN.
Open Datasets Yes We adopt a pretrained proxy model for the reward, which is trained on a dataset of 300, 000 molecules that are randomly generated as provided in (Bengio et al., 2021a).
Dataset Splits No The paper uses standard tasks like Grid World and Molecule Generation, but it does not provide specific percentages or counts for training, validation, and test splits for these datasets. It refers to a "pretrained proxy model" for the molecule generation task but does not detail its splits.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or cloud computing instances used for running the experiments.
Software Dependencies No The paper mentions software like Adam optimizer and Message Passing Neural Networks, but it does not specify version numbers for general software dependencies or libraries (e.g., Python, PyTorch, etc.).
Experiment Setup Yes A detailed description of the hyperparameters and setup can be found in Appendix C.1. For GAFlow Net, the only hyperparameter that requires tuning is the coefficient α of intrinsic rewards, where we use a same value for state-based and edge-based augmentation. We tune α in {0.001, 0.005, 0.01, 0.05, 0.1, 0.5} with grid search. Specifically, α = 0.001 for Grid World with all values of horizon except for H {16, 64}, where we set α to be 0.005. For the molecule generation task, α is set to be 0.1.