Research map / Probabilistic and scientific modeling / Dynamical systems
WeightFlow: Learning Stochastic Dynamics via Evolving Weight of Neural Network
Ruikun Li, Jiazhen Liu, Huandong Wang, Qingmin Liao, Yong Li
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 |
|---|---|---|---|
| integrate along a latent control path (CDE) instead of directly over time (ODE) (ablation) | WeightFlow · five simulated systems | Wasserstein distance: Toggle Switch W 0.90 -> 0.82 | Improved |
| use autoencoder-based path projection instead of other projection algorithms (ablation) | WeightFlow · simulated systems | Wasserstein distance | Improved |
| add sequential aligning (warm-start from previous snapshot weights) in warm-up (ablation) | WeightFlow · five simulated systems | Wasserstein distance | Improved |
| reduce training data to 20% (ablation) | WeightFlow · five simulated systems | Wasserstein distance | No clear effect |
| use a 1-dim latent path instead of higher dimensions (ablation) | WeightFlow · five simulated systems | Wasserstein distance | No clear effect |
| shuffle the autoregressive order randomly (ablation) | WeightFlow · five simulated systems | Wasserstein distance | No clear effect |
| replace GRU backbone with a Transformer backbone (ablation) | WeightFlow · five simulated systems | Wasserstein distance | No clear effect |
| increase backbone hidden dimension beyond 8 (ablation) | WeightFlow · five simulated systems | Wasserstein distance | No clear effect |
| model dynamics in weight space (WeightFlow) instead of DeepRUOT | stochastic dynamics models · Epidemic | Wasserstein distance: 1.78 -> 1.10 | Improved |
| model dynamics in weight space (WeightFlow) instead of DeepRUOT | stochastic dynamics models · Signalling Cascade 1 and 2 | Wasserstein distance: 0.52 -> 0.48; 0.51 -> 0.49 | No clear effect |
| model dynamics in weight space (WeightFlow) instead of NLSB | stochastic dynamics models · Embryoid body | Wasserstein distance: 14.39 -> 14.18 | No clear effect |
| model dynamics in weight space (WeightFlow) instead of RUOT | stochastic dynamics models · beta-cell | MMD: 0.06 -> 0.02 | Improved |
Closest papers by meaning
- Neural Network Weights Do Not Converge to Stationary Points: An Invariant Measure Perspective (ICML 2022 · similarity 0.87)
- Better Training using Weight-Constrained Stochastic Dynamics (ICML 2021 · similarity 0.87)
- Neural Stochastic Flows: Solver-Free Modelling and Inference for SDE Solutions (NeurIPS 2025 · similarity 0.86)
- From Lazy to Rich: Exact Learning Dynamics in Deep Linear Networks (ICLR 2025 · similarity 0.85)
- Learning invariant representations of time-homogeneous stochastic dynamical systems (ICLR 2024 · similarity 0.85)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · 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.
