Research map / Video / Video segmentation and tracking
A Disentangled Recognition and Nonlinear Dynamics Model for Unsupervised Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use KVAE with a CNN encoder-decoder instead of DVBF | Pendulum · Pendulum test | ELBO: 798.56 -> 810.08 | Helped |
| Use KVAE with a CNN encoder-decoder instead of DMM | Pendulum · Pendulum test | ELBO: 784.70 -> 810.08 | Helped |
| Use the CNN encoder-decoder rather than the MLP encoder-decoder (ablation) | KVAE · Pendulum test | ELBO: 807.02 -> 810.08 | Helped |
| Use Kalman smoothing rather than filtering for missing-frame imputation | KVAE · Bouncing ball, random missing frames | fraction of incorrectly guessed pixels | Helped |
| Use Kalman smoothing to incorporate future observations when imputing a consecutive gap | KVAE · Bouncing ball, consecutive missing frames | fraction of incorrectly guessed pixels | Helped |
| Replace the LSTM dynamics parameter network with an MLP (ablation) | KVAE · Bouncing ball | imputation performance | Hurt |
| Replace the LSTM dynamics parameter network with an RNN (ablation) | KVAE · Bouncing ball | imputation performance | Hurt |
| Replace the LSTM dynamics parameter network with a FIFO MLP (ablation) | KVAE · Bouncing ball | imputation performance | Hurt |
| Train with 30% or 40% missing data rather than complete training sequences (ablation) | KVAE · Bouncing ball | imputation accuracy | Hurt |
| Use smoothing after training with 30% or 40% missing data rather than filtering with all training data available | KVAE · Bouncing ball | imputation accuracy | Helped |
Closest papers by meaning
- Maximum-Likelihood Learning of Latent Dynamics Without Reconstruction (ICML 2026 · similarity 0.87)
- Sequential Disentanglement by Extracting Static Information From A Single Sequence Element (ICML 2024 · similarity 0.87)
- Continuity-Preserving Convolutional Autoencoders for Learning Continuous Latent Dynamical Models from Images (ICLR 2025 · similarity 0.86)
- Identifying latent state transitions in non-linear dynamical systems (ICLR 2025 · similarity 0.86)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.85)
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.85)
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.