Research map / Probabilistic and scientific modeling / Variational and generative sampling
Generative Modeling of Discrete Latent Structures via Dynamic Policy Gradients
Stefan Ivanovic, Ge Liu, Mohammed El-Kebir
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 |
|---|---|---|---|
| add dynamic reward rescaling by 1/Pr(X_i|theta) (ablation) | naive policy gradient · Process graph inference | F1: 0 -> 0.891 (k=10) | Improved |
| add dynamic reward rescaling by 1/Pr(X_i|theta) (ablation) | naive policy gradient · Subset inference (|U|=1000) | median F1: 0.849 -> 0.938 | Improved |
| replace local search with GReinSS (shared latent model) | local search · Subset inference (|U|=1000) | median F1: 0.869 -> 0.938 | Improved |
| use GReinSS instead of GEM-based VAE, autoregressive, diffusion | GEM baselines · Subset inference (|U|=1000) | median F1: < 0.4 -> 0.938 | Improved |
| use GReinSS instead of baselines with very few random walks | baselines · Process graph inference (k=10) | median F1: below 0.55 -> 0.891 | Improved |
| use GReinSS instead of GFlowNets | GFlowNets · Process graph inference | median F1: below 0.55 -> 0.891 | Improved |
| increase number of random walks k | most methods · Process graph inference | F1 | Improved |
| use a different off-policy sampling proposal biased toward Pr(S|X,theta) (ablation) | GReinSS · Subset inference | F1 | No clear effect |
| use GReinSS instead of RSEM | RSEM · RNA isoform reconstruction (MBD2, 61 samples) | isoform prediction error: 0.471 -> 0.0069 | Improved |
| use GReinSS instead of RSEM across genes | RSEM · RNA isoform reconstruction (14,390 genes) | isoform prediction error: median difference -0.0405 | Improved |
| use GReinSS instead of GEM baselines at tiny universe size | GEM baselines · Subset inference (|U|=10) | median F1: 1.0 | No clear effect |
Closest papers by meaning
- Latent Traversals in Generative Models as Potential Flows (ICML 2023 · similarity 0.86)
- $f$-Trajectory Balance: A Loss Family for Tuning GFlowNets, Generative Models, and LLMs with Off- and On-Policy Data (ICML 2026 · similarity 0.85)
- Structural Inference of Dynamical Systems with Conjoined State Space Models (NeurIPS 2024 · similarity 0.84)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · similarity 0.83)
- On Divergence Measures for Training GFlowNets (NeurIPS 2024 · similarity 0.83)
- Guided Structural Inference: Leveraging Priors with Soft Gating Mechanisms (ICML 2025 · 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.
