Research map / Probabilistic and scientific modeling / Variational and generative sampling
SESaMo: Symmetry-Enforcing Stochastic Modulation for Normalizing Flows
Janik Kreit, Dominic Schuh, Kim Andrea Nicoli, Lena Funcke
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 |
|---|---|---|---|
| replace canonicalization with stochastic modulation | RealNVP · GMM 2x1 (broken Z8) | ESS: 0.747 -> 0.9947 | Improved |
| replace canonicalization with stochastic modulation | RealNVP · GMM 2x1 (exact Z8) | KL divergence: 0.013 -> 0.0008 | Improved |
| use stochastic modulation instead of FAB / VMoNF | RealNVP · Complex phi^4 8x8 | ESS: 0.26/0.22 -> 0.9472 (exact U(1)) | Improved |
| use stochastic modulation with learnable breaking parameter | RealNVP · Hubbard 2x1 (broken Z4) | ESS: 0.839 -> 0.996 vs canonicalization | Improved |
| scale to a large lattice with SESaMo | RealNVP · Hubbard 18x100 | ESS: 0.024 -> 0.74 vs canonicalization | Improved |
| use VMoNF (mixture of flows) with reverse KL | VMoNF · GMM 2x1 Z8 | mode coverage | Hurt |
| make the modulation probability learnable (broken symmetry) (ablation) | SESaMo · phi^4 and Hubbard | symmetry breaking recovery | Improved |
| add bijectivity penalty term to reverse KL (ablation) | SESaMo · all | bijectivity | Improved |
| move modes closer together (smaller radius R) | SESaMo · GMM | ESS | Hurt |
Closest papers by meaning
- SymDrift: One-Shot Generative Modeling under Symmetries (NeurIPS 2026 · similarity 0.89)
- SymDiff: Equivariant Diffusion via Stochastic Symmetrisation (ICLR 2025 · similarity 0.87)
- Parameter Symmetry and Noise Equilibrium of Stochastic Gradient Descent (NeurIPS 2024 · similarity 0.86)
- Symmetries in Overparametrized Neural Networks: A Mean Field View (NeurIPS 2024 · similarity 0.86)
- Improving Equivariant Networks with Probabilistic Symmetry Breaking (ICLR 2025 · similarity 0.85)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.85)
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.
