Tag
The paper identifies that existing time series tokenization strategies (patch and phase tokenization) introduce spectral distortions such as spectral leakage and aliasing, and proposes TimeFT, a parameter-free, nearly distortion-free frequency-based tokenizer that serves as a drop-in replacement and improves performance across forecasting, classification, and anomaly detection tasks.
SpecKCast 是一种免训练的扩散 Transformer 推理加速方法，在频域中用逐频带 Kalman 滤波器预测模型输出并自适应调度去噪步，在 FLUX.1-dev、Qwen-Image、Wan2.1-14B 和 HunyuanVideo 上最高实现 6 倍加速，额外内存开销不足 0.05 GiB。
CoRe proposes a model-agnostic learning objective for multivariate time-series forecasting that uses frequency coherence and relational graph losses to improve prediction accuracy over standard methods.
The paper introduces CoSPOT, an LLM-based framework for online time series forecasting that uses compositional spectral prompts to efficiently adapt to non-stationary data and unseen patterns.
This paper proposes a zero-shot time-series anomaly detection framework that enhances LLM inputs with frequency-domain evidence from FFT to better capture structural anomalies.
DSTFView is a dual-input spatio-temporal-frequency multi-view framework for cloud-edge workload forecasting, jointly modeling closeness and period dependencies with an adaptive fusion mechanism to capture abrupt changes.
This paper introduces FourierQK, a method that applies FFT-based frequency-domain preprocessing to learned query and key projections in transformer attention, achieving significant validation loss reductions on character-level language modelling. The approach preserves the full attention score structure and demonstrates reproducible gains over standard dot-product attention.
This paper introduces FRESCO, an Echo State Network architecture operating entirely in the frequency domain to achieve O(N) complexity for dense recurrent updates, matching state-of-the-art performance on benchmarks while reducing computational costs.