1: \begin{abstract}

2:   This paper builds theoretical foundations for the recovery of a

3:   newly proposed class of smooth graph signals, approximately

4:   bandlimited graph signals, under three sampling strategies: uniform

5:   sampling, experimentally designed sampling and active sampling. We

6:   then state minimax lower bounds on the maximum risk for the

7:   approximately bandlimited class under these three sampling

8:   strategies and show that active sampling cannot fundamentally

9:   outperform experimentally designed sampling. We propose a recovery

10:   strategy to compare uniform sampling with experimentally designed

11:   sampling. As the proposed recovery strategy lends itself well to

12:   statistical analysis, we derive the exact mean square error for each

13:   sampling strategy. To study convergence rates, we introduce two

14:   types of graphs and find that (1) the proposed recovery strategy

15:   achieves the optimal rates; and (2) the experimentally designed

16:   sampling fundamentally outperforms uniform sampling for Type-2 class

17:   of graphs. To validate our proposed recovery strategy, we test it on

18:   five specific graphs: a ring graph with $k$ nearest neighbors, an

19:   Erd\H{o}s-R\'enyi graph, a random geometric graph, a small-world

20:   graph and a power-law graph and find that experimental results match

21:   the proposed theory well. This work also presents a comprehensive

22:   explanation for when and why sampling for semi-supervised learning

23:   with graphs works.

24: \end{abstract}

25: