Skip to content

Learnable Spectral Frequency Filtering

Home / Concepts / Learnable Spectral Frequency Filtering

Learnable Spectral Frequency Filtering

Auto-generated stub. Edit this file to add more details.

A technique for integrating learnable spectral filters into the embedding layer of transformer models to mitigate low-frequency bias in time-series data.

Why It Matters

Addresses the structural high-frequency bias of self-attention in time-series forecasting, providing a lightweight plug-and-play mechanism for diverse transformer architectures.

Evidence

applying learnable frequency filters to embedded signals enhance the performance of transformer-based forecasters across a number of architectures with negligible increases in memory and compute.

Metadata & Links