Research map / Probabilistic and scientific modeling / Variational and generative sampling
Evidential Sparsification of Multimodal Latent Spaces in Conditional Variational Autoencoders
Masha Itkina, Boris Ivanovic, Ransalu Senanayake, Mykel J Kochenderfer, Marco Pavone
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 |
|---|---|---|---|
| use evidential sparse distribution instead of softmax | VQ-VAE PixelCNN prior · miniImageNet classification of samples | accuracy (%): 20.688 -> 19.937 | No clear effect |
| use sparsemax instead of softmax | VQ-VAE PixelCNN prior · miniImageNet classification of samples | accuracy (%): 20.688 -> 6.125 | Hurt |
| use evidential sparse distribution instead of sparsemax | VQ-VAE PixelCNN prior · miniImageNet classification of samples | top-5 accuracy (%): 17.500 -> 47.875 | Improved |
| use evidential sparse distribution instead of softmax | Trajectron++ · ETH pedestrian, direct sampling | NLL: 4.698 -> 4.686 | No clear effect |
| use evidential sparse distribution instead of softmax | Trajectron++ · ETH pedestrian, top-5 sampling | NLL: 3.951 -> 3.862 | Improved |
| use sparsemax instead of softmax | Trajectron++ · ETH pedestrian, top-5 sampling | NLL: 3.951 -> 4.360 | Hurt |
| use sparsemax instead of softmax | Trajectron++ · ETH pedestrian, top-5 sampling | FDE: 0.757 -> 0.802 | Hurt |
| use evidential sparse distribution instead of softmax | Trajectron++ · ETH pedestrian, top-5 sampling | FDE: 0.757 -> 0.753 | No clear effect |
| use evidential sparse distribution instead of softmax | CVAE · MNIST | Bhattacharyya distance: 22% | Improved |
| use evidential sparse distribution instead of sparsemax | CVAE · MNIST | Bhattacharyya distance: 10% | Improved |
| use sparsemax instead of softmax | CVAE · MNIST odd query | distance to target | Hurt |
| train on reduced fractions of MNIST (0.1, 0.5, 0.75, 1.0) (ablation) | evidential sparse distribution · MNIST | Wasserstein and Bhattacharyya | Improved |
Closest papers by meaning
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.89)
- Exploring the Latent Space of Autoencoders with Interventional Assays (NeurIPS 2022 · similarity 0.87)
- Aggregation of Dependent Expert Distributions in Multimodal Variational Autoencoders (ICML 2025 · similarity 0.87)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.86)
- Balancing Multimodal Training Through Game-Theoretic Regularization (NeurIPS 2025 · similarity 0.86)
- Aligning Multimodal Representations through an Information Bottleneck (ICML 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.
