Disentangling by Factorising
About
We define and address the problem of unsupervised learning of disentangled representations on data generated from independent factors of variation. We propose FactorVAE, a method that disentangles by encouraging the distribution of representations to be factorial and hence independent across the dimensions. We show that it improves upon $\beta$-VAE by providing a better trade-off between disentanglement and reconstruction quality. Moreover, we highlight the problems of a commonly used disentanglement metric and introduce a new metric that does not suffer from them.
Hyunjik Kim, Andriy Mnih• 2018
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Time Series Forecasting | ETTh1 | MSE0.484 | 601 | |
| Time Series Forecasting | ETTm1 | MSE2.326 | 334 | |
| Time Series Forecasting | ETTh1 (test) | MSE0.87 | 262 | |
| Time Series Forecasting | ETTm1 (test) | MSE2.326 | 196 | |
| Time Series Forecasting | Traffic (test) | MSE0.982 | 192 | |
| Time Series Forecasting | Traffic | MSE0.936 | 145 | |
| Time Series Forecasting | Weather (test) | MSE0.53 | 110 | |
| Time Series Forecasting | Electricity (test) | MSE0.807 | 72 | |
| Image Reconstruction | CelebA-HQ (test) | FID (Reconstruction)87.19 | 50 | |
| Disentangled Representation Learning | Cars3D | FactorVAE0.906 | 35 |
Showing 10 of 48 rows