Research map / Probabilistic and scientific modeling / Molecules and proteins
Set Prediction without Imposing Structure as Conditional Density Estimation
David W Zhang, Gertjan J. Burghouts, Cees G. M. Snoek
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 conditional density estimation instead of assignment-based set-loss training | Set prediction · Polygons | Hungarian loss [10^-5]: 1195.0 ± 6.8 / 14.8 ± 5.3 -> 8.8 ± 1.5 | Improved |
| Use conditional density estimation instead of assignment-based set-loss training | Set prediction · Digits | Chamfer loss [10^-5]: 2.4 ± 0.3 / 48.0 ± 2.3 -> 1.8 ± 0.1 | Improved |
| Use DESP rather than DSPN trained with Chamfer loss | Point-cloud auto-encoding · set MNIST | Set size RMSE: 0.300 ± 0.130 / 0.800 ± 0.080 -> 0.002 ± 0.003 | Improved |
| Use DESP rather than DSPN trained with assignment loss | Point-cloud auto-encoding · set MNIST | Chamfer loss [10^-4]: 0.9 ± 0.1 / 2.8 ± 0.8 -> 0.6 ± 0.1 | Improved |
| Use DESP rather than DSPN trained with assignment loss | Point-cloud auto-encoding · set MNIST | Hungarian loss [10^-4]: 1600.2 ± 162.7 / 3.2 ± 0.8 -> 0.6 ± 0.1 | Improved |
| Use DESP rather than DSPN with Hungarian training | Object set prediction · CLEVR | AP50: 94.0 ± 0.4 -> 96.2 ± 0.5 | Improved |
| Use DESP rather than DSPN with Hungarian training | Object set prediction · CLEVR | AP70: 82.2 ± 1.4 -> 80.9 ± 1.8 | Hurt |
| Use DESP rather than DSPN with Hungarian training | Object set prediction · CLEVR | AP80: 58.9 ± 3.4 -> 54.3 ± 3.4 | Hurt |
| Use DESP rather than the element-wise DeepSets baseline | Subset anomaly detection · CelebA ambiguous | F1: 0.47 ± 0.01 -> 0.66 ± 0.00 | Improved |
| Use DESP rather than the element-wise DeepSets baseline | Subset anomaly detection · CelebA unambiguous + ambiguous | F1: 0.63 ± 0.01 -> 0.76 ± 0.01 | Improved |
| Add stochastic steps to prediction instead of using fully deterministic gradient descent (ablation) | DESP · set MNIST | Reconstruction performance | Improved |
| Set the stochastic-step proportion to 0.8 rather than fully deterministic prediction (ablation) | DESP · CelebA subset anomaly detection | F1 | Improved |
Closest papers by meaning
- Multiset-Equivariant Set Prediction with Approximate Implicit Differentiation (ICLR 2022 · similarity 0.84)
- Arbitrary Conditional Distributions with Energy (NeurIPS 2021 · similarity 0.84)
- Volume Optimality in Conformal Prediction with Structured Prediction Sets (ICML 2025 · similarity 0.83)
- Joint Learning of Energy-based Models and their Partition Function (ICML 2025 · similarity 0.82)
- Learning Energy-Based Models from Stochastic Interpolants using Spatiotemporal Differences (NeurIPS 2026 · similarity 0.82)
- Learning Set Functions with Implicit Differentiation (AAAI 2025 · similarity 0.82)
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.
