… learning framework to discover effective network sampling heuristics by leveraging automatically
learnt node and graph … the nodes selected from this sampled subgraph can effectively …

… By carefully examining sampling policies, we find that their influence on the learning
performance, in terms of training speed and accuracy, cannot be described using precise closed-…

… learning structured policies for continuous control. In traditional reinforcement learning, policies
… Random We also include the random policy which is uniformly sampled from the action …

… network with reinforcement learning to learn the optimal query strategy. By jointly training
on several source graphs with full labels, we learn a transferable active learning policy which …

… To enable learning the representation on the large-scale graph data … sampling strategies
to facilitate the training process. Herein, we propose an adaptive Graph Policy-driven Sampling …

… sampling strategy that optimizes a sampling policy … , our sampling policy better generalizes
across various graphs. We show how random sampling complements importance sampling …

… (b) GCPN conducts message passing to encode the state as node embeddings then
produce a policy πθ. (c) An action at with 4 components is sampled from the policy. (d) The …

… must be a state in the sampling subset Ss. Notice that Q-learning asymptotically achieves Q
… For the sampled Q-learning, since we focus on the policy, as long as correct estimation can …

… graph and local action representations for the individual nodes; capturing information at both
scales allows our agent to learn nuanced policies. … depending on the state (graph). We also …

… -wise sampling, which is a common type of graph sampling … the previous layer using the
sampling policy q. To make this … our sampling policy q and training methods for the sampling …