Research map / Probabilistic and scientific modeling / Variational and generative sampling
Disentangling Multiple Features in Video Sequences using Gaussian Processes in Variational Autoencoders
Sarthak Bhagat, Shagun Uppal, Zhuyun Yin, Nengli Lim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add geodesic loss (ablation) | MGP-VAE prediction network · Moving MNIST (k=1) | MSE: 25.4 -> 18.5 | Improved |
| add geodesic loss (ablation) | MGP-VAE prediction network · Moving MNIST (k=2) | MSE: 72.2 -> 69.2 | Improved |
| add geodesic loss (ablation) | MGP-VAE prediction network · Coloured dSprites (k=1) | MSE: 6.1 -> 4.5 | Improved |
| add geodesic loss (ablation) | MGP-VAE prediction network · Sprites (k=1) | MSE: 68.8 -> 61.6 | Improved |
| add geodesic loss (ablation) | MGP-VAE prediction network · Coloured dSprites (k=1) | BCE: 85.2 -> 70.3 | Improved |
| use MGP-VAE latent instead of DDPAE | prediction MLP · Moving MNIST (k=1) | MSE: 35.2 -> 25.4 | Improved |
| use MGP-VAE representation instead of DDPAE | kNN attribute classification · Coloured dSprites | average mAP (%): 80.8 -> 83.4 | Improved |
| use MGP-VAE representation instead of DDPAE | kNN attribute classification · Sprites | average mAP (%): 76.9 -> 78.5 | Improved |
| set fBM Hurst H=0.9 rather than H=0.5 for motion channel (ablation) | MGP-VAE · Moving MNIST | motion capture (qualitative) | Improved |
| include more channels than factors of variation (ablation) | MGP-VAE · Coloured dSprites and Sprites | disentanglement (qualitative) | Improved |
| replace fBM mixtures with 4-5 Brownian bridge channels (ablation) | MGP-VAE · Coloured dSprites and Sprites | disentanglement (qualitative) | No clear effect |
| use Ornstein-Ulenbeck process as latent prior (ablation) | MGP-VAE · video datasets | disentanglement (qualitative) | Hurt |
Closest papers by meaning
- Disentangling Disentanglement in Variational Autoencoders (ICML 2019 · similarity 0.89)
- Learning to Decompose and Disentangle Representations for Video Prediction (NeurIPS 2018 · similarity 0.89)
- Gaussian Process Prior Variational Autoencoders (NeurIPS 2018 · similarity 0.88)
- Disentangling Learning Representations with Density Estimation (ICLR 2023 · similarity 0.88)
- Orthogonality-Enforced Latent Space in Autoencoders: An Approach to Learning Disentangled Representations (ICML 2023 · similarity 0.87)
- Neighbour-Driven Gaussian Process Variational Autoencoders for Scalable Structured Latent Modelling (ICML 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
