Monthly
288 pp. per issue
6 x 9, illustrated
ISSN
0899-7667
E-ISSN
1530-888X
2014 Impact factor:
2.21

Neural Computation

June 2014, Vol. 26, No. 6, Pages 1236-1237
(doi: 10.1162/NECO_c_00597)
@ 2014 Massachusetts Institute of Technology
Errata to “Bayesian Community Detection” (Neural Computation, Sept. 2012, Vol. 24, No. 9:2434–2456)
Article PDF (37.56 KB)
Abstract

Due to bugs in the MCMC sampler for the IHM, IDM, IRM, and IRMCB models, the inference and predictions for these models were not correct. We have therefore revised the sampler in the toolbox accompanying the article and rerun the experiments of Tables 1, 2, and 3.

The new results show that the BCD outperforms the existing methods for about half of the considered networks, but not for the majority, as we originally stated (see the updated Table 2). In terms of the number of components extracted by IRM and BCD, it still holds in general that the BCD extracts more clusters; however, the new results show that it is not the case for all the considered networks (see the updated Table 1).

Table 1: Network Properties and Results of Analysis.

  Network Properties Clusters   AUC [%]
         
  N N+ WGH. DIR. IRM BCD γ IRM BCD
USAir97 332 2126     14.4(2) 19.0(3) 1.00(0) 95.2(14) 94.5(15)
USPowerGrid 4941 6594     16.6(5) 37.8(32) 0.04(0) 81.4(10) 77.0(16)
Football 115 613     11.0(0) 13.6(4) 0.10(0) 88.0(43) 87.9(35)
Celegans 306 2345 36.8(10) 29.8(9) 1.00(0) 88.7(8) 89.1(2)
yeast 2361 6646     25.0(0) 32.6(12) 1.00(0) 88.7(5) 87.5(7)
lesmis 77 254   12.6(4) 12.4(4) 0.99(1) 93.0(39) 94.2(27)
Geom 7343 11898   71.2(14) 76.8(9) 0.98(1) 89.4(8) 89.4(10)
netscience 1589 2742   8.2(2) 8.4(2) 0.89(4) 59.5(21) 65.7(24)
cond-mat 16726 47594   53.0(7) 53.8(41) 0.31(11) 71.5(4) 75.9(8)
SciMet 3084 10413   19.0(3) 23.0(7) 1.00(0) 90.7(9) 89.1(8)
smaGri 1059 4919   13.8(5) 19.0(4) 1.00(0) 92.4(7) 89.2(8)
smallW 396 994   8.8(2) 11.4(5) 1.00(0) 99.1(3) 98.3(5)
NIPS 234 598     8.0(0) 30.6(15) 0.03(0) 89.1(26) 89.4(25)
NIPSCW 2865 4733   32.0(4) 85.6(23) 0.01(0) 90.4(12) 89.7(5)

Table 2: Area Under Curve (AUC) [%] Link Prediction Score.

          BCD BCD
  IHW IDM IRM IRMCB Shared γ Separate γ
USAir97 80.3(25) 85.2(16) 95.2(14) 94.4(13)  94.5(15)  95.3(12)
USPowerGrid 75.8(11) 74.4(19) 81.4(10) 82.9(13)  77.0(16)  80.7(17)
Football 84.4(41) 84.7(38) 88.0(43) 88.2(36)  87.9(35)  88.6(41)
Celegans 57.9(21) 57.6(27) 88.7(8) 89.2(6)  89.1(2)  88.9(5)
yeast 71.7(27) 83.8(4) 88.7(5) 88.6(7)  87.5(7)  87.4(7)
lesmis 60.0(25) 79.7(62) 93.0(39) 97.1(29)  94.2(27)  96.1(23)
Geom 70.8(20) 73.4(5) 89.4(8) 88.8(4)  89.4(10)  90.1(7)
netscience 67.0(21) 53.8(19) 59.5(21) 58.4(23)  65.7(24)  68.2(17)
cond-mat 65.4(8) 64.1(5) 71.5(4) 68.5(3)  75.9(8)  74.5(16)
SciMet 74.0(6) 80.9(5) 90.7(9) 90.5(6)  89.1(8)  89.6(6)
smaGri 71.4(9) 82.8(10) 92.4(7) 92.4(5)  89.2(8)  89.3(5)
smallW 84.0(16) 90.8(10) 99.1(3) 99.0(3)  98.3(5)  97.7(6)
NIPS 89.7(25) 92.5(28) 89.1(26) 87.4(45)  89.4(25)  94.5(17)
NIPSCW 68.5(37) 83.7(14) 90.4(12) 89.9(5)  89.7(5)  91.5(6)

Table 3: Average per Iteration CPU Time.

          BCD BCD
  IHW IDM IRM IRMCB Shared γ Separate γ
USAir97 0.23(4) 0.29(0) 0.29(2) 0.40(6) 2.68(21) 3.09(25)
USPowerGrid 1.83(3) 2.42(3) 3.96(38) 5.34(38) 64.13(376) 69.01(479)
Football 0.04(0) 0.06(0) 0.07(0) 0.08(0) 0.74(4) 1.11(6)
Celegans 0.23(4) 0.18(1) 0.37(4) 0.55(7) 4.63(44) 5.32(43)
yeast 1.92(29) 2.60(18) 2.92(21) 5.48(22) 30.81(180) 40.74(269)
lesmis 0.06(1) 0.04(0) 0.06(0) 0.08(1) 0.50(3) 0.64(3)
Geom 9.73(120) 9.44(141) 16.92(383) 31.09(742) 188.23(1146) 320.29(705)
netscience 0.62(5) 1.25(5) 2.09(9) 2.54(15) 15.92(93) 17.72(99)
cond-mat 45.89(1566) 46.60(1797) 50.21(2068) 72.85(187) 245.70(2914) 739.99(7486)
SciMet 2.30(32) 2.58(4) 3.82(5) 6.87(42) 57.39(399) 63.27(367)
smaGri 0.86(16) 1.07(3) 1.15(2) 2.01(23) 14.83(76) 14.47(37)
smallW 0.35(5) 0.37(0) 0.51(3) 0.83(15) 3.93(10) 4.40(20)
NIPS 0.10(1) 0.11(1) 0.21(1) 0.30(5) 1.74(12) 2.29(13)
NIPSCW 2.33(31) 1.43(4) 2.28(19) 3.69(35) 32.13(350) 39.62(371)