Fig. 4From: Understanding the limitations of network online learningCumulative reward results on a Twitter social network constructed from mutual interactions. The budget in the experiment was 50k queries. Twitter data is typically collected via repeated API access, so represents a natural use case for NOL-HTR. Inset: Magnification showing that NOL-HTR and the high degree baseline methods outperform NOL in the first few thousand probes of the experimentBack to article page