Robust Scheduling with GFlowNets
Authors: David W Zhang, Corrado Rainone, Markus Peschl, Roberto Bondesan
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate different aspects of our generative scheduling approach by incrementally adding complexity to the computation graph dataset. First, we restrict training and evaluation to a single computation graph, which corresponds to the same unconditional setting considered by previous works on GFlow Nets (Bengio et al., 2021a; Deleu et al., 2022; Jain et al., 2022). Next, we train with multiple computation graphs and evaluate on unseen ones. |
| Researcher Affiliation | Collaboration | David W. Zhang University of Amsterdam Corrado Rainone Markus Peschl Roberto Bondesan Qualcomm AI Research |
| Pseudocode | No | The paper does not contain any sections or figures explicitly labeled 'Pseudocode' or 'Algorithm'. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing code or provide links to a code repository for their methodology. |
| Open Datasets | Yes | For training, we use 1000 different computation graphs, with equally many sampled from the two random graph distributions: Erd os R enyi (Erd os et al., 1960), and Layered Graphs (Gagrani et al., 2022). ... The computation graphs in this dataset originate from a diverse set of neural network architectures with different applications... |
| Dataset Splits | No | The paper mentions training and evaluation on different sets of graphs ('train on graphs... and evaluate on unseen ones'), but does not specify explicit train/validation/test splits, percentages, or absolute counts for dataset partitioning. |
| Hardware Specification | No | The paper refers to 'target hardware' and 'four homogenous devices' but does not specify any particular CPU, GPU, or other hardware model numbers used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'pymoo' and 'Adam optimizer' but does not provide specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | We train GFlow Nets conditioned on a temperature randomly sampled between 0.01 and 1. At inference, we use 0.005 for the temperature in all experiments. We use the Adam optimizer with default hyperparameters to optimize the parameters. We compute the gradients at each update step based on a minibatch that consists of 100 sampled trajectories for a single computation graph and use a single temperature value to compute their rewards. |