ResearchBenchmarksManifold Representation Learning on Swiss Roll, dSprites, and MNIST combined averageFollow9k-NN RecallGAE1.5123.4565.47.344Apr 8, 2026Evaluation ResultsMethodMethodLinksk-NN RecallKL Divergence (0.01)KL Divergence (0.1)KL Divergence (1)MSEAccuracyGAE2026.0496.56.5683Vanilla AE2026.047.89.28.88.27.29GGAE2026.047.58.2987.510IRAE2026.047.26.87.27.278CAE2026.045.877.575.52DN2026.044.565.55.25.24GRAE2026.0444.53.54.24.56TAE2026.043.82.52.83.247SPAE2026.043.233.24.24.85BLAE2026.041.81111.21