Research map / Time series / Causal and generative time series
FlowPath: Learning Data-Driven Manifolds with Invertible Flows for Robust Irregularly-sampled Time Series Classification
YongKyung Oh, Dong-Young Lim, Sungil Kim
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 |
|---|---|---|---|
| replace the fixed spline path in Neural CDE with a learned non-invertible MLP path | Neural CDE · 18 UEA/UCR datasets (average) | accuracy: 0.667 -> 0.689 | Improved |
| make the learned control path an invertible flow instead of an MLP (ablation) | Neural CDE with learned path · 18 UEA/UCR datasets (average) | accuracy: 0.689 -> 0.730 | Improved |
| make the learned path invertible instead of an MLP, at 70% missing (ablation) | Neural CDE with learned path · 18 UEA/UCR datasets (70% missing) | accuracy: 0.655 -> 0.718 | Improved |
| use a fixed spline path versus FlowPath's learned invertible path under 70% missing data | Neural CDE · 18 UEA/UCR datasets (70% missing) | accuracy: 0.652 -> 0.718 | Improved |
| use the per-dataset best flow versus a fixed ResNet flow (ablation) | FlowPath · 18 UEA/UCR datasets (average) | accuracy: 0.705 -> 0.730 | Mixed |
| FlowPath versus strongest baseline DualDynamics, pairwise over dataset-scenarios | DualDynamics · 18 UEA/UCR datasets (all settings) | wins/ties/losses: 40 wins, 6 ties, 26 losses | Improved |
| FlowPath versus DualDynamics with no missing data | DualDynamics · 18 UEA/UCR datasets (regular) | wins/ties/losses: 8 wins, 2 ties, 8 losses | No clear effect |
| use GRU flow instead of spline path in Neural CDE on PAMAP2 without sensor dropout | Neural CDE · PAMAP2 (0% sensor dropout) | F1 score: 95.0 -> 95.6 | Improved |
| use ResNet flow instead of spline path at 50% sensor dropout | Neural CDE · PAMAP2 (50% sensor dropout) | F1 score: 51.6 -> 57.9 | Improved |
| use a learned MLP path instead of spline in Neural CDE on PAMAP2 without dropout (ablation) | Neural CDE · PAMAP2 (0% sensor dropout) | F1 score: 95.0 -> 93.9 | Hurt |
| choose ResNet flow over GRU flow at heavy sensor dropout (ablation) | FlowPath · PAMAP2 (40% sensor dropout) | F1 score: 59.6 -> 65.9 | Improved |
| use FlowPath instead of Neural CDE under 30% sensor dropout on PAMAP2 | Neural CDE · PAMAP2 (30% sensor dropout) | accuracy: 65.5 -> 67.3 | Improved |
Later papers that compared with this method
1 later accepted paper reports FlowPath as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- One-Step Graph-Structured Neural Flows for Irregular Multivariate Time Series Classification (ICML 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DualDynamics — DualDynamics: Synergizing Implicit and Explicit Methods for Robust Irregular Time Series Analysis (AAAI 2025) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Faithful Embeddings of Irregular and Asynchronous Data for Online Log-NCDEs (NeurIPS 2026 · similarity 0.85)
- PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation (NeurIPS 2026 · similarity 0.85)
- Neural Stochastic Flows: Solver-Free Modelling and Inference for SDE Solutions (NeurIPS 2025 · similarity 0.85)
- Elucidating Flow Matching ODE Dynamics via Data Geometry and Denoisers (ICML 2025 · similarity 0.85)
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.85)
- Trajectory Flow Matching with Applications to Clinical Time Series Modelling (NeurIPS 2024 · 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.
