Online Planner Selection with Graph Neural Networks and Adaptive Scheduling

Abstract

Automated planning is one of the foundational areas of AI. Since no single planner can work well for all tasks and domains, portfolio-based techniques have become increasingly popular in recent years. In particular, deep learning emerges as a promising methodology for online planner selection. Owing to the recent development of structural graph representations of planning tasks, we propose a graph neural network (GNN) approach to selecting candidate planners. GNNs are advantageous over a straightforward alternative, the convolutional neural networks, in that they are invariant to node permutations and that they incorporate node labels for better inference.Additionally, for cost-optimal planning, we propose a two-stage adaptive scheduling method to further improve the likelihood that a given task is solved in time. The scheduler may switch at halftime to a different planner, conditioned on the observed performance of the first one. Experimental results validate the effectiveness of the proposed method against strong baselines, both deep learning and non-deep learning based.The code is available at https://github.com/matenure/GNN_planner.

Paradigm Shift: The Promise of Deep Learning in Molecular Systems Engineering and Design

“…(Dai et al, 2020). It is worth noting that graph neural networks have also been extended to several chemical engineering applications, such as control (Wang et al, 2018), process scheduling (Ma et al, 2020b), fault diagnosis (Wu and Zhao, 2021), among others. In the realm of molecular systems, applications of GNNs have reaped numerous successes.…”

Planning with Learned Object Importance in Large Problem Instances using Graph Neural Networks

“…Relational Representations for Learning to Plan. Our work uses graph neural networks (GNNs) (Scarselli et al 2008;Kipf and Welling 2016;Battaglia et al 2018), an increasingly popular choice for relational machine learning with applications to planning (Wu et al 2020;Ma et al 2020;Shen, Trevizan, and Thiébaux 2020;Rivlin, Hazan, and Karpas 2020). One advantage of GNNs over logical representations (Muggleton 1991;Lavrac and Dzeroski 1994;Džeroski, De Raedt, and Driessens 2001) is that GNNs natively support continuous object-level and relational prop-erties.…”

Explainable Planner Selection for Classical Planning

“…The first category of portfolio approaches from the literature learns a schedule for multiple planners offline (e.g., for optimal planning Ma et al (2020) predict which planner has the highest chance of solving a task and which second planner is most likely to solve it if the first planner fails.…”

“…The Thirty-Sixth AAAI Conference on Artificial Intelligence the resulting portfolio selector suggests that the remaining information in the image is sufficient for planner selection. In a follow-up paper, Ma et al (2020) eliminate the lossy transformation from graphs to images by directly feeding the graphs into graph convolutional networks (GCN; Kipf and Welling 2017). This causes a modest performance improvement and implies that the images already contain enough information for good predictions.…”