Research map / Architectures and efficiency / RNNs and normalization
Multiset-Equivariant Set Prediction with Approximate Implicit Differentiation
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 iDSPN instead of Slot Attention† | CLEVR object property prediction · CLEVR object property prediction, 128x128 | AP0.125: 7.9% → 76.9% | Helped |
| Use approximate implicit differentiation instead of backpropagating through the optimizer (ablation) | DSPN · Random multiset autoencoding | reconstruction loss | No clear effect |
| Add Nesterov momentum to iDSPN (ablation) | iDSPN · Random multiset autoencoding | reconstruction loss | Helped |
| Compare at roughly equal computation rather than equal iteration counts | iDSPN versus DSPN · Random multiset autoencoding | reconstruction loss | Helped |
| Use iDSPN with Nesterov momentum when n × d approaches latent dimensionality 512 (ablation) | DSPN · Random multiset autoencoding | reconstruction loss | Hurt |
| Use exclusive multiset-equivariance rather than set-equivariance (ablation) | iDSPN · Random multiset autoencoding | reconstruction loss: lower loss by a factor of >10 | Helped |
| Replace FSPool with sum or mean pooling (ablation) | DSPN · Random multiset autoencoding | reconstruction loss | Hurt |
| Use exclusive multiset-equivariance rather than set-equivariance (ablation) | iDSPN · CLEVR object property prediction | average precision | Helped |
| Use iDSPN rather than DSPN with the same iteration count and optimizer (ablation) | DSPN · CLEVR object property prediction | average precision | Helped |
| Use set-equivariant models for class-specific numbering of equal elements | Class-specific numbering models · Class-specific numbering | accuracy: 0.0 | Hurt |
| Use non-equivariant BiLSTM or positional encoding instead of iDSPN | Class-specific numbering models · Class-specific numbering | sample efficiency: 100× | Hurt |
| Use iDSPN instead of Slot Attention† | CLEVR object property prediction · CLEVR object property prediction, 128x128 | AP∞: 96.4% → 98.8% | Helped |
Closest papers by meaning
- Set Norm and Equivariant Skip Connections: Putting the Deep in Deep Sets (ICML 2022 · similarity 0.87)
- Set-based Neural Network Encoding Without Weight Tying (NeurIPS 2024 · similarity 0.84)
- Polynomial Width is Sufficient for Set Representation with High-dimensional Features (ICLR 2024 · similarity 0.84)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.84)
- Improving Equivariant Model Training via Constraint Relaxation (NeurIPS 2024 · similarity 0.83)
- Unlocking Better Closed-Set Alignment Based on Neural Collapse for Open-Set Recognition (AAAI 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.