EPOC (Endpoint‑Preserving Online Correction) is a method for multi‑horizon time‑series forecasting that compresses residual information using low‑order discrete cosine transform (DCT) coefficients and the final residual value. It applies channel‑wise online ridge regressions to blend a DCT‑based correction with a base forecast, achieving significant reductions in mean squared error (15.40%) and mean absolute error (9.35%) across 96 fixed‑base conditions while keeping auxiliary state small (median 6,352 B). The approach outperforms several existing adapters and offers a lightweight, endpoint‑focused correction strategy.
arXiv:2609.23686v1 Announce Type: new Abstract: Patch-based autoregressive time-series forecasting often ties input representation, learned transitions, and recursive execution to one patch length. W...
RATL is a plug‑in method for multivariate time‑series forecasting that uses a frozen base forecaster to build a memory of its historical forecast residuals. During inference, RATL retrieves residual trajectories from similar past contexts and employs a set‑aware router to combine them, providing learned feedback correction. Experiments demonstrate that this residual‑retrieval approach improves the performance of the base forecaster across various benchmarks and backbones.
arXiv:2610.07834v1 Announce Type: new Abstract: Retrieval-augmented time-series forecasting uses the continuations of historical segments similar to the current context as references for a forecaster...
arXiv:2609.23701v1 Announce Type: cross Abstract: Probabilistic forecasting requires accurate central predictions and calibrated uncertainty estimates. This paper presents TEMPER, the Temporal Encode...
arXiv:2606. 01339v1 Announce Type: cross Abstract: Long-term time-series forecasting needs models that are accurate yet efficient enough for commodity hardware.
SACQ introduces a structured prediction head for long‑horizon time‑series forecasting that replaces the traditional flatten readout. It uses a two‑stage decoding process: a coarse patch‑grid scaffold is first generated, then each future position is refined via cross‑attention over historical memory, with a learned gate merging the corrections. A batch‑adaptive scaled log‑cosh loss further stabilizes training under long horizons and noisy labels, leading to superior MSE/MAE performance across multiple backbones with minimal overhead.
Probabilistic forecasting requires accurate central predictions and calibrated uncertainty estimates. This paper presents TEMPER, the Temporal Encoder-Masked Probabilistic Ensemble Regressor, a univar...
arXiv:2609. 20193v1 Announce Type: new Abstract: Retrieval plug-ins supply a deep forecaster with information its lookback window cannot carry.
The paper investigates whether entire Transformer blocks can be skipped during inference in two time‑series foundation models, Chronos‑Bolt and TimesFM 2.5. By applying block‑level masks selected on calibration windows, the authors show that moderate sparsity (e.g., 7/11 blocks for Chronos‑Bolt and 16/20 for TimesFM) preserves forecasting quality while reducing latency by roughly 11–13%. However, more aggressive skipping degrades performance on certain datasets, and dynamic, input‑dependent routing remains unproven across architectures.
arXiv:2610.02248v1 Announce Type: new Abstract: Operational land surface forecasting systems built on Mamba-family Structured State Space Models absorb non-stationary confounding events (unrecorded i...
The paper investigates whether simplifying a forecasting target via a statistical first stage improves neural network performance in short‑horizon volatility prediction. Using a signal‑forecast‑system diagnostic framework on five liquid U.S. assets, a volatility‑aligned HAR‑style model outperforms AR, MA, and ARIMA, and its residuals exhibit markedly lower variance and weak autocorrelation. However, augmenting a residual‑LSTM with these simplified targets actually worsens mean squared error compared to a pure LSTM, revealing a forecaster‑preconditioner asymmetry where statistical simplification does not translate into neural preconditioning benefits.
Target-Constrained Pooling (TCP) is a method for improving trajectory extrapolation by separating the finite-horizon endpoint from the continuation shape and estimating the endpoint from two complementary source signals—uniform pooling for final response levels and matched-increment pooling for remaining change. TCP selects representative sources, projects endpoint estimates onto a target-consistent region, applies corrections to a free forecast, and calibrates their influence with strength calibration and degradation-aware weighting. Across 25,245 targets from five domains, TCP reduces mean RMSE by 35.8% to 42.1% for all ten target-only forecasters and improves all 100 forecaster, dataset, and direction comparisons, while using far fewer donor inputs and matching distance computations.