Research map / Probabilistic and scientific modeling / Molecules and proteins
Neural Expectation Maximization
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 |
|---|---|---|---|
| Replace the EM-constrained updates with a flexible recurrent update | N-EM on Static Shapes · Static Shapes | AMI: 0.475 ± 0.043 -> 0.826 ± 0.005 | Helped |
| Train in stages using 20, 500, 50,000 digits (ablation) | RNN-EM on Flying MNIST · Flying MNIST, 2 digits | AMI: 0.819 ± 0.022 -> 0.917 ± 0.005 | Helped |
| Increase K from 3 to 5 for 3-object training and test (ablation) | RNN-EM · Flying Shapes, 3 objects | AMI: 0.970 ± 0.005 -> 0.997 ± 0.002 | Helped |
| Use K=3 rather than K=5 for 5-object training and test (ablation) | RNN-EM · Flying Shapes, 5 objects | AMI: 0.878 ± 0.003 -> 0.614 ± 0.003 | Hurt |
| Choose K larger than the number of objects during training (ablation) | RNN-EM · Flying Shapes | AMI: 0.970 ± 0.005 -> 0.997 ± 0.002 | Helped |
| Evaluate with K=3 after training with K=5 (ablation) | RNN-EM trained on 3-object Flying Shapes · Flying Shapes, test generalization | AMI: 0.997 ± 0.002 -> 0.914 ± 0.015 | Hurt |
| Evaluate on three digits without further training (ablation) | RNN-EM trained on two-digit Flying MNIST · Flying MNIST, 3 digits | AMI: 0.917 ± 0.005 -> 0.838 ± 0.008 | Hurt |
| Transfer grouping dynamics to static images (ablation) | RNN-EM trained on Flying MNIST · Static MNIST, 2 digits | AMI: 0.619 ± 0.023; stage-wise: 0.772 ± 0.008 | Mixed |
| Increase the number of objects from 3 to 5 while retaining K=5 (ablation) | RNN-EM on Flying Shapes · Flying Shapes | AMI: 0.997 ± 0.002 -> 0.878 ± 0.003 | Hurt |
Closest papers by meaning
- Bridging Neural and Symbolic Representations with Transitional Dictionary Learning (ICLR 2024 · similarity 0.85)
- Object-Centric Learning with Slot Mixture Module (ICLR 2024 · similarity 0.83)
- Deciphering 'What' and 'Where' Visual Pathways from Spectral Clustering of Layer-Distributed Neural Representations (CVPR 2024 · similarity 0.83)
- Easing Concept Bleeding in Diffusion via Entity Localization and Anchoring (ICML 2024 · similarity 0.83)
- Attention-based clustering (NeurIPS 2025 · similarity 0.83)
- Perceptual Group Tokenizer: Building Perception with Iterative Grouping (ICLR 2024 · 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.