86 Quantifying and Learning Linear Symmetry-Based Disentanglement (LSBD) Table 4.6: Scores for the Arrow dataset. MODEL BETA ↑ FACTOR ↑ SAP ↑ DCI ↑ MIG↑ MOD↑ DLSBD↓ VAE 1.000±.000 .646±.032 .017±.004 .009±.003 .013±.004 .961±.012 1.316±.193 β-VAE .999±.002 .588±.045 .018±.004 .008±.002 .015±.005 .898±.032 1.178±.065 CC-VAE .982±.056 .707±.102 .019±.004 .011±.005 .016±.004 .980±.038 1.013±.096 FACTOR 1.000±.000 .659±.028 .017±.003 .008±.003 .014±.002 .935±.037 1.526±.125 DIP-I 1.000±.000 .624±.042 .020±.004 .008±.002 .012±.003 .967±.027 1.521±.113 DIP-II 1.000±.000 .644±.064 .020±.004 .009±.003 .013±.004 .973±.011 1.616±.102 ADAGVAE 1.000±.000 .656±.137 .016±.005 .020±.009 .009±.004 .973±.042 1.620±.147 ADAMLVAE .997±.008 .706±.168 .017±.007 .019±.009 .011±.004 .943±.111 1.395±.117 QUESSARD 1.000±.000 .596±.032 .016±.006 .008±.004 .017±.008 .999±.000 1.183±.412 LSBD-VAE 1.000±.001 .664±.105 .016±.002 .009±.004 .019±.005 .897±.108 1.627±.104 /0 LSBD-VAE 1.000±.000 .662±.046 .017±.005 .009±.004 .020±.005 .963±.010 1.475±.121 /256 LSBD-VAE 1.000±.000 .956±.119 .021±.006 .297±.157 .023±.003 .967±.092 .245±.474 /512 LSBD-VAE 1.000±.000 1.000±.000 .022±.006 .390±.022 .026±.003 .999±.000 .000±.000 /768 LSBD-VAE 1.000±.000 1.000±.000 .022±.003 .396±.026 .026±.006 .999±.000 .000±.000 /1024 LSBD-VAE 1.000±.000 1.000±.000 .019±.005 .401±.018 .026±.004 .999±.000 .000±.000 /1280 LSBD-VAE 1.000±.000 1.000±.000 .019±.005 .397±.017 .026±.007 .999±.000 .000±.000 /1536 LSBD-VAE 1.000±.000 1.000±.000 .020±.004 .399±.018 .026±.004 .999±.000 .000±.000 /1792 LSBD-VAE 1.000±.000 1.000±.000 .020±.006 .444±.186 .027±.004 .999±.000 .000±.000 /FULL LSBD-VAE .016±.006 /PATHS
RkJQdWJsaXNoZXIy MjY0ODMw