Research map / Probabilistic and scientific modeling / Variational and generative sampling
Overcoming Limitations of Mixture Density Networks: A Sampling and Fitting Framework for Multimodal Future Prediction
Check 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 the two-stage Evolving WTA sampling and mixture-fitting framework instead of MDN | Multimodal future prediction · CPI | EMD: 1.83 -> 1.57 | Helped |
| Use the two-stage Evolving WTA sampling and mixture-fitting framework instead of MDN | Multimodal future prediction · Stanford Drone Dataset | NLL: 9.71 -> 9.33 | Helped |
| Use Evolving WTA instead of WTA or Relaxed WTA for hypothesis generation | CPI hypothesis sampling · CPI hypotheses | EMD: 3.94 / 2.82 -> 1.89 | Helped |
| Use distribution hypotheses with NLL and include their uncertainty in mixture fitting rather than point hypotheses with Euclidean loss (ablation) | Sampling-fitting framework · CPI | EMD: 1.70 -> 1.57 | Helped |
| Use distribution hypotheses with NLL and include their uncertainty in mixture fitting rather than point hypotheses with Euclidean loss (ablation) | Sampling-fitting framework · Stanford Drone Dataset | NLL: 9.56 -> 9.33 | Helped |
| Add one history frame rather than using no history (ablation) | EWTAD-MDF · Stanford Drone Dataset | NLL: 15.30 -> 9.13 | Helped |
| Add a second history frame rather than using one (ablation) | EWTAD-MDF · CPI | EMD: 2.13 -> 1.63 | Helped |
| Add a second history frame rather than using one (ablation) | EWTAD-MDF · Stanford Drone Dataset | NLL: 9.13 -> 9.30 | Hurt |
| Predict farther into the future (ablation) | EWTAD-MDF · CPI | EMD: 1.30 -> 1.74 -> 1.84 | Hurt |
| Predict farther into the future (ablation) | EWTAD-MDF · Stanford Drone Dataset | NLL: 6.94 -> 8.46 -> 12.21 | Hurt |
| Increase the number of hypotheses from 20 to 40 (ablation) | EWTAD-MDF · CPI | EMD: 1.63 -> 1.57 | Helped |
| Increase the number of hypotheses from 40 to 80 (ablation) | EWTAD-MDF · CPI | EMD: 1.57 -> 1.65 | Hurt |
Closest papers by meaning
- Predictive Dynamic Fusion (ICML 2024 · similarity 0.86)
- Annealing in variational inference mitigates mode collapse: a theoretical study on Gaussian mixtures (NeurIPS 2026 · similarity 0.85)
- End-to-end Learning of Gaussian Mixture Priors for Diffusion Sampler (ICLR 2025 · similarity 0.85)
- FutureOmni: Evaluating Future Forecasting from Omni-Modal Context for Multimodal LLMs (ICML 2026 · similarity 0.85)
- When to Align, When to Predict: A Phase Diagram for Multimodal Learning (NeurIPS 2026 · similarity 0.84)
- Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers (CVPR 2025 · similarity 0.83)
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.