Generative Modeling on Gaussian-mixture Boltzmann distribution (test)
0.0754Mean L2 DistanceGaussian kernel drifting
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Gaussian kernel driftingLatent dimension=32, Hidden width=256, Hidden layers=3, Batch size=1024, Learning rate=10^−3, Kernel bandwidth (sigma)=0.22, Drift step (eta)=0.22, Local perturbations per sample=256, Number of generated samples=5000, Number of reference samples=50002026.03 | 0.0754 | 0.0425 | 0.002 | 1.0045 | 1.0263 |