Research map / Probabilistic and scientific modeling / Variational and generative sampling
ICE-BeeM: Identifiable Conditional Energy-Based Deep Models Based on Nonlinear ICA
Ilyes Khemakhem, Ricardo Monti, Diederik Kingma, Aapo Hyvarinen
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 ICE-BeeM pretrained features and fit conditioning coefficients for unseen classes instead of training an unconditional EBM on those classes | ConvMLP-50 on MNIST · MNIST unseen classes 8-9 | CDSM objective: 3.64 ± 0.10 -> 2.95 ± 0.02 | Improved |
| Use ICE-BeeM pretrained features and fit conditioning coefficients for unseen classes instead of training an unconditional EBM on those classes | ConvMLP-90 on CIFAR10 · CIFAR10 unseen classes 8-9 | CDSM objective: 8.16 ± 0.05 -> 8.03 ± 0.01 | Improved |
| Train an identifiable conditional EBM rather than an unconditional EBM | ConvMLP-50 on FMNIST · FMNIST unseen classes 8-9 | classification accuracy: 56.33 ± 3.18 -> 77.07 ± 1.39 | Improved |
| Train an identifiable conditional EBM rather than an unconditional EBM | ConvMLP-50p on CIFAR10 · CIFAR10 unseen classes 8-9 | classification accuracy: 51.88 ± 1.33 -> 64.42 ± 1.09 | Improved |
| Add positive output features (ablation) | ICE-BeeM ConvMLP-50 on MNIST · MNIST unseen classes 8-9 | classification accuracy: 76.98 ± 1.61 -> 88.46 ± 1.14 | Improved |
| Add positive output features (ablation) | ICE-BeeM ConvMLP-50 on FMNIST · FMNIST unseen classes 8-9 | classification accuracy: 77.07 ± 1.39 -> 71.67 ± 1.85 | Hurt |
| Add positive output features (ablation) | ICE-BeeM ConvMLP-50 on CIFAR10 · CIFAR10 unseen classes 8-9 | classification accuracy: 69.36 ± 2.23 -> 64.42 ± 1.09 | Hurt |
| Use augmented features (ablation) | ICE-BeeM Unet on MNIST · MNIST unseen classes 8-9 | classification accuracy: 97.79 ± 0.34 -> 97.18 ± 0.5 | No clear effect |
| Increase feature dimension from 50 to 200 (ablation) | ICE-BeeM ConvMLP on CIFAR10 · CIFAR10 unseen classes 8-9 | classification accuracy: 69.36 ± 2.23 -> 64.73 ± 1.36 | Hurt |
Closest papers by meaning
- Robustness of Nonlinear Representation Learning (ICML 2024 · similarity 0.86)
- Beyond ICA: Identifiability by Symmetry Breaking (NeurIPS 2026 · similarity 0.86)
- Multi-View Causal Representation Learning with Partial Observability (ICLR 2024 · similarity 0.85)
- Feature learning from non-Gaussian inputs: the case of Independent Component Analysis in high dimensions (ICML 2025 · similarity 0.84)
- On the Identifiability of Hybrid Deep Generative Models: Meta-Learning as a Solution (NeurIPS 2024 · similarity 0.83)
- Bridging Mini-Batch and Asymptotic Analysis in Contrastive Learning: From InfoNCE to Kernel-Based Losses (ICML 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.
