Researchers Release SHADE Heavy-Tailed Time Series Forecasting Paper
Get up to speed
Jiahao Weng and Jiyu Liu have released SHADE, an alpha-stable Lévy diffusion predictor. Through nonzero deep readout, it enables residual stable kernels to achieve distribution-level shrinkage, addressing the problem of Gaussian diffusion models producing overly narrow, overconfident predictive distributions when forecasting heavy-tailed time series. The model reverts to Gaussian diffusion or full denoising mode on near-light-tailed or high signal-to-noise ratio data.
AI-generated from reports · 3 hr agoUpdated
Coverage timeline
Follow the coverage to understand different sides of the event.
- arXiv q-finSHADE: A Pause-Alpha Stable Denoising-Based Shrinkage Method for Heavy-Tailed Probabilistic Time Series Forecasting
The study proposes an Alpha-stable Lévy diffusion predictor called SHADE, which achieves distribution-level shrinkage through a residual stable kernel by reading outputs at non-zero depths, addressing the problem of overly narrow predictive distributions and overconfidence when Gaussian diffusion models forecast heavy-tailed time series.
Trend of this event
Not enough consecutive observations yet; trend is not drawn.