Research map / Label-efficient and robust learning / Semi-supervised learning
Flow Contrastive Estimation of Energy-Based Models
Ruiqi Gao, Erik Nijkamp, Diederik P. Kingma, Zhen Xu, Andrew M. Dai, Ying Nian Wu
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 |
|---|---|---|---|
| jointly train Glow and an EBM with the shared FCE objective instead of training Glow by MLE | Glow · SVHN | FID: 41.70 -> 20.19 | Improved |
| jointly train Glow and an EBM with the shared FCE objective instead of training Glow by MLE | Glow · CIFAR-10 | FID: 45.99 -> 37.30 | Improved |
| jointly train Glow and an EBM with the shared FCE objective instead of training Glow by MLE | Glow · CelebA | FID: 23.32 -> 12.21 | Improved |
| use FCE rather than MLE to train Glow | Glow · SVHN test | bits per dimension: 2.17 -> 2.25 | Hurt |
| use FCE rather than MLE to train Glow | Glow · CIFAR-10 test | bits per dimension: 3.35 -> 3.45 | Hurt |
| use FCE rather than MLE to train Glow | Glow · CelebA test | bits per dimension: 3.49 -> 3.54 | Hurt |
| add joint FCE training after the EBM-only initialization stage (ablation) | Conv-small EBM on SVHN · SVHN, 500 labels | classification error (%): 9.42 -> 7.05 | Improved |
| add joint FCE training after the EBM-only initialization stage (ablation) | Conv-small EBM on SVHN · SVHN, 1000 labels | classification error (%): 8.50 -> 6.35 | Improved |
| add joint FCE training after the EBM-only initialization stage (ablation) | Conv-large EBM on SVHN · SVHN, 500 labels | classification error (%): 8.86 -> 6.86 | Improved |
| add joint FCE training after the EBM-only initialization stage (ablation) | Conv-large EBM on SVHN · SVHN, 1000 labels | classification error (%): 7.60 -> 5.54 | Improved |
| add VAT regularization (ablation) | FCE with Conv-large EBM · SVHN, 500 labels | classification error (%): 6.86 -> 4.47 | Improved |
| add VAT regularization (ablation) | FCE with Conv-large EBM · SVHN, 1000 labels | classification error (%): 5.54 -> 3.87 | Improved |
Closest papers by meaning
- A Unified Contrastive Energy-based Model for Understanding the Generative Ability of Adversarial Training (ICLR 2022 · similarity 0.87)
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.85)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (NeurIPS 2025 · similarity 0.85)
- Learning Energy Decompositions for Partial Inference in GFlowNets (ICLR 2024 · similarity 0.84)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.84)
- Self-Corrected Flow Distillation for Consistent One-Step and Few-Step Image Generation (AAAI 2025 · similarity 0.84)
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.
