Research map / Detection and segmentation / Open-set and few-shot detection
Efficient Iterative Amortized Inference for Learning Symmetric and Disentangled Multi-Object Representations
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 |
|---|---|---|---|
| Remove GECO from training objective (ablation) | EfficientMORL · CLEVR6 | ARI: 82.4 -> 79.9 ARI (within noise), KL drops from 692.7 to 177.5 | Mixed |
| Remove GECO from training objective (ablation) | EfficientMORL · Tetrominoes | ARI: 97.9 -> 82.2 ARI | Hurt |
| Use hierarchical prior variant with reversed prior++ instead of bottom-up prior (ablation) | EfficientMORL · Tetrominoes | ARI: 85.7 -> 97.9 ARI | Helped |
| Use reversed prior++ (refinement KL uses p(z^1|z^2)) instead of plain reversed prior (ablation) | EfficientMORL · Tetrominoes | ARI: 86.9 -> 97.9 ARI | Helped |
| Replace DualGRU with single standard GRU of hidden dim 2D (ablation) | EfficientMORL · CLEVR6 | ARI: 82.4 -> 79.7 ARI, KL rises 692.7 -> 1357.4 | Hurt |
| Use IAI refinement with test-time steps I=0 instead of I=1 (ablation) | EfficientMORL · CLEVR6 | ARI and MSE: zero steps achieves 99.1% of refined ARI and MSE | No clear effect |
| Add iterative refinement (test-time IAI) relative to zero steps (ablation) | EfficientMORL · CLEVR6 KL | KL: larger KL gap between I=0 and I=1 | Helped |
| Use refinement steps during training (I>0) vs no refinement (I=0) (ablation) | EfficientMORL HVAE · CLEVR6 training ELBO | negative ELBO / convergence | Helped |
| Use EfficientMORL instead of Slot Attention (disentanglement) | EfficientMORL · CLEVR6 | DCI disentanglement: 0.46 -> 0.63 | Helped |
| Use Gaussian likelihood instead of mixture of Gaussians for sprite datasets | EfficientMORL · Multi-dSprites and Tetrominoes | ARI | No clear effect |
| Reduce refinement steps I from 3 to 1 during training (curriculum) | EfficientMORL · CLEVR6 and Multi-dSprites | training wall-clock time | Helped |
| Use EfficientMORL (I=0) instead of IODINE for forward pass | EfficientMORL · 128x128 forward pass timing | speed: 10x faster forward pass | Helped |
Closest papers by meaning
- Tripod: Three Complementary Inductive Biases for Disentangled Representation Learning (ICML 2024 · similarity 0.87)
- Object centric architectures enable efficient causal representation learning (ICLR 2024 · similarity 0.87)
- RePercENT: Scaling Disentangled Representation Learning Beyond Two Modalities (NeurIPS 2026 · similarity 0.87)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.86)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.86)
- Disentanglement Beyond Static vs. Dynamic: A Benchmark and Evaluation Framework for Multi-Factor Sequential Representations (NeurIPS 2025 · similarity 0.86)
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.