Research map / Time series / Causal and generative time series
Scalable Bayesian dynamic covariance modeling with variational Wishart and inverse Wishart processes
Creighton Heaukulani, Mark van der Wilk
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 |
|---|---|---|---|
| use an additive white noise parameterization (ablation) | Wishart process · Dow 30 | test loglikelihood | Improved |
| use an additive white noise parameterization (ablation) | inverse Wishart process · Dow 30 | test loglikelihood: 164.09 ± 26.47 -> 164.49 ± 19.82 | Improved |
| use an additive white noise parameterization (ablation) | inverse Wishart process · FX | test loglikelihood: 81.42 ± 4.12 -> 82.10 ± 3.72 | Improved |
| replace the full covariance model with a 10-factor model (ablation) | Wishart process with additive white noise · Dow 30 | test loglikelihood: 165.98 ± 23.23 -> 165.39 ± 30.89 | Mixed |
| replace the full covariance model with a 10-factor model (ablation) | Wishart process with additive white noise · FX | test loglikelihood: 82.69 ± 4.15 -> 81.12 ± 3.59 | Hurt |
| use a 10-factor rather than 30-factor model (ablation) | Wishart process with additive white noise · S&P 500 | test loglikelihood: 1438.40 ± 130.14 -> 1423.31 ± 132.19 | Hurt |
| use an inverse Wishart process instead of a Wishart process | 10-factor model · FX | test loglikelihood: 81.12 ± 3.59 vs 77.76 ± 3.94 | Hurt |
| use an inverse Wishart process instead of a Wishart process | 30-factor model · S&P 500 | test loglikelihood: 1438.40 ± 130.14 vs 1047.73 ± 1436.16 | Hurt |
| use an inverse Wishart process instead of a Wishart process | full covariance model with additive white noise · FX | test loglikelihood: 82.69 ± 4.15 vs 82.10 ± 3.72 | Hurt |
| use a 10-factor inverse Wishart model rather than the ARCH baseline | S&P 500 · S&P 500 | test loglikelihood: 1358.23 ± 355.12 vs 1275.27 ± 264.48 | Hurt |
| use a 10-factor Wishart model rather than the ARCH baseline | S&P 500 · S&P 500 | test loglikelihood: 1358.23 ± 355.12 -> 1423.31 ± 132.19 | Improved |
Closest papers by meaning
- Stochastic Differential Equations with Variational Wishart Diffusions (ICML 2020 · similarity 0.87)
- Free-Form Variational Inference for Gaussian Process State-Space Models (ICML 2023 · similarity 0.86)
- Volatility Based Kernels and Moving Average Means for Accurate Forecasting with Gaussian Processes (ICML 2022 · similarity 0.86)
- Variational Inference for Latent Variables and Uncertain Inputs in Gaussian Processes (JMLR 2016 · similarity 0.86)
- A Set-Sequence Model for Time Series (NeurIPS 2026 · similarity 0.85)
- Least squares variational inference (NeurIPS 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.
