Research map / Time series / Causal and generative time series
Multi-Task Dynamical Systems
Alex Bird, Christopher K. I. Williams, Christopher Hawthorne
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 |
|---|---|---|---|
| generate all RNN weights from latent z instead of only the bias (ablation) | MTDS · Mocap data efficiency (7% of training set) | MSE: 0.44 -> 0.29 | Improved |
| generate all RNN weights from latent z instead of only the bias (ablation) | MTDS · Mocap style transfer (Child) | classifier probability: 0.59 -> 0.95 | Improved |
| generate all RNN weights from latent z instead of only the bias (ablation) | MTDS · Mocap style transfer (Old) | classifier probability: 0.55 -> 0.86 | Improved |
| generate all RNN weights from latent z instead of only the bias (ablation) | MTDS (k=7) · Mocap novel styles | MSE: 0.48 -> 0.46 | Improved |
| use multi-task MTDS instead of single-task GRU | MTDS (k=5) · Mocap data efficiency (3% of training set) | MSE: 1.11 -> 0.53 | Improved |
| use MTDS instead of pooled GRU (open loop) | MTDS (k=5) · Mocap data efficiency (7% of training set) | MSE: 0.52 -> 0.29 | Improved |
| use MTDS instead of pooled GRU (open loop) with large training data | MTDS (k=7) · Mocap data efficiency (97% of training set) | MSE: 0.16 -> 0.18 | No clear effect |
| train with open-loop criterion instead of closed loop (teacher forcing) (ablation) | Pooled GRU · Mocap data efficiency (97% of training set) | MSE: 1.21 -> 0.16 | Improved |
| use MTDS instead of pooled GRU (closed loop) at long horizon | MTDS (k=7) · Mocap novel styles (tau=200) | MSE: 7.28 -> 0.61 | Improved |
| increase latent dimension k from 3 to 5 (ablation) | MTDS · Mocap data efficiency (3% of training set) | MSE: 0.62 -> 0.53 | Improved |
| use 2-layer instead of 1-layer pooled GRU (ablation) | Pooled GRU (open loop) · Mocap novel styles (tau=10) | MSE: 0.56 -> 0.55 | No clear effect |
Closest papers by meaning
- UniTS: A Unified Multi-Task Time Series Model (NeurIPS 2024 · similarity 0.86)
- MAESTRO : Adaptive Sparse Attention and Robust Learning for Multimodal Dynamic Time Series (NeurIPS 2025 · similarity 0.86)
- MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification (AAAI 2025 · similarity 0.85)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.85)
- Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data (ICLR 2025 · similarity 0.85)
- Meta-Dynamical State Space Models for Integrative Neural Data Analysis (ICLR 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.
