The paper introduces a dynamic multi‑step decoding approach for time‑series foundation models, arguing that the best decoding strategy varies by instance rather than being fixed. A lightweight selector network ranks candidate rollout schedules using the backbone’s hidden states, trained on historical windows and evaluated via dynamic programming to keep complexity quadratic. At deployment, multiple selectors vote to choose a schedule, defaulting to one‑shot decoding when they disagree, and the method shows statistically significant improvements in point‑forecast accuracy on the fev‑bench benchmark with frozen Chronos‑2 and Moirai‑1.1‑R‑small backbones.
arXiv:2607. 24892v1 Announce Type: cross Abstract: Text-conditioned time-series forecasting predicts a series from both its numerical history and natural-language context, allowing forecasts to account for events and constraints that the past alone cannot reveal.
Text-conditioned time-series forecasting predicts a series from both its numerical history and natural-language context, allowing forecasts to account for events and constraints that the past alone cannot reveal. This requires both reliable numerical forecasting and the ability to interpret contextual information.
arXiv:2511. 18191v2 Announce Type: replace Abstract: Time series forecasting drives operational decisions under tight latency budgets, and autoregressive time series foundation models (TSFMs) increasingly deliver the most accurate forecasts.
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...
arXiv:2607. 19659v1 Announce Type: new Abstract: Time-series foundation models can forecast across heterogeneous domains without task-specific training, but their forecasts are fixed once produced and cannot directly incorporate task-specific expert feedback.
arXiv:2606. 04342v1 Announce Type: cross Abstract: Multi-step time series forecasting (MSF) is commonly evaluated using point-wise error metrics such as mean squared error (MSE), implicitly treating the conditional mean as a sufficient target.
arXiv:2606. 27711v1 Announce Type: cross Abstract: We introduce a neural network-based framework for learning time series estimators through a process we term decision-theoretic pretraining.
arXiv:2608. 07420v1 Announce Type: new Abstract: World models are expected to support imagination over extended temporal horizons, yet most are still trained through local few-step prediction objectives and deployed by recursively rolling out their own predictions.
arXiv:2608. 05571v1 Announce Type: new Abstract: Retrieval-augmented forecasting promises to adapt frozen Time Series Foundation Models (TSFMs) to new domains without fine-tuning, but recent methods typically rely on learned fusion modules, i.
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.
The paper introduces a hybrid attention model that learns a unified time‑aware patch representation for irregular multivariate time series (IMTS) forecasting. It employs a time‑aware patch encoding to embed variable‑length intra‑patch timestamps, a time bias attention mechanism to adjust for temporal misalignment and asynchronous cross‑channel dependencies, and a hybrid causal mask on a decoder‑only Transformer to balance historical context with autoregressive forecasting. The authors also curate VersaTSA, a 30 B‑observation dataset preserving native sampling sparsity, and demonstrate state‑of‑the‑art zero‑shot performance on three IMTS benchmarks while remaining competitive on regular MTS tasks.
ROAR is a Retrieval Opportunity-Aware Refinement framework designed for zero-shot time series forecasting. It enhances predictions by weighting queries according to base-forecast difficulty and the improvement potential of retrieved historical candidates, then learning a gate to control correction strength. The method jointly calibrates the forecasting module and gate, anchoring the final prediction to an initial refined output, and achieves the lowest average MSE across seven benchmarks compared to other methods.