Time-Series Foundation Models

Overview

Time-series foundation models learn reusable patterns from large collections of sequences, aiming to support tasks such as forecasting, classification, imputation, and anomaly detection across domains. Common approaches include Transformer models with patch or value tokenization, masked and contrastive pretraining, next-token prediction, and newer recurrent or state-space architectures; training data may mix real series with synthetic ones. Models are used zero-shot or adapted through fine-tuning, in-context examples, covariates, retrieval, and lightweight adapters, with benchmarks such as GIFT-Eval and UCR/UEA testing different capabilities. Transfer can reduce the need for task-specific training, but results vary with data and operating conditions: specialists may still win, and robustness to distribution shifts, long contexts, uncertainty, and fair evaluation remain open challenges.

Latest papers

345 papers54 papers in the last four weeks, down 17% on the four weeks before. 0.4% of all new papers.

Nov 18, 2026

Recurrent Domain Incremental Adaptation of Time-Series Foundation Models

Pretrained Time-Series Foundation Models (TSFMs) provide strong representations for downstream classification tasks. Yet adapting a TSFM under changing operating conditions remains challenging. A model must improve on each newly observed condition without sacrificing conditions it has already learned, often from only a small labeled calibration set and without knowing the operating condition at inference time. We find that this difficulty has an important geometric structure. Under an operating-condition shift, the embedding space of a TSFM does not move uniformly: some local regions change substantially, while others remain nearly unchanged. Consequently, a global adaptation can unnecessarily move embeddings that were already correctly classified. To address this, we propose the novel PACT (Piecewise Adaptation via Chart Transport). PACT partitions the frozen source embedding space into local charts and learns lightweight residual transports for the regions affected by each new condition. An explicit identity path is always retained, and a learned gate softly selects among the identity and accumulated transports for each input. The pretrained TSFM encoder, classifier head, and previously learned transports remain fixed. With only 5% labeled calibration data per arriving condition and no condition identity at test time, PACT maintains an average macro-F1 above 0.95 across a four-condition bearing-fault stream. Under homogeneous tasks and label spaces, similar retention-adaptation behavior is observed across public benchmarks. Code is available at https://github.com/Sachithx/PACT.
Nov 16, 2026

Severe but Harmless: Stress-Testing Shift Severity for Frozen Time-Series Forecasters

Zero-shot time-series foundation models (TSFMs) have become attractive alternatives to task-specific models because they handle diverse downstream tasks, including anomaly detection and forecasting, without per-series training. However, time-series data undergo distribution shifts, and the dissimilarity between two regimes is commonly scored with statistical measures such as Wasserstein-1, where a higher score is read as a more severe shift. In systematic experiments using three established severity measures on nine regime shifts from three standard traffic datasets, we show that these measures do not reliably predict degradation for instance-normalized TSFMs. The regimes rated most severe are not the regimes that degrade the models most, and degradation reaches 29% in the lowest-scoring regime. In most cases, the audit shows no positive relationship between measured severity and model degradation. We provide an equivariance-based stress-testing analysis that explains why severity readings can become decoupled from model degradation.
Nov 16, 2026

Breaking feature collapse in self-supervised time series encoders

General-purpose time series encoders promise a single backbone that generalises across tasks and domains; however, they currently do not scale. We trace this down to a critical point of failure: under the predominant masked data modelling supervision, the encoder's output tokens partially collapse. This is the natural consequence of supervising encoders only in the input space, which leaves no constraint against redundant token dimensions in the feature space. We introduce OTISv2, which suppresses the collapse by complementing the reconstruction objective in input space with two objectives in the feature space: a self-distillation loss against an exponential-moving-average teacher, and a KoLeo regulariser. Under matched training, the resulting encoder reduces the cosine similarity and lifts the effective rank of its output tokens relative to its predecessor, and it scales monotonically from 7.1 M to 116 M parameters where its predecessor declines. It matches or surpasses existing general-purpose time series encoders across 143 uni- and multi-variate datasets. To foster further research, we release our code and pre-trained weights.
Nov 13, 2026

SWIFT: Sample-Wise Instance Fine-Tuning of Time-Series Foundation Models for Forecasting

While current few-shot adaptation strategies for time series forecasting foundation models are predominantly dataset-centric, the necessity of obtaining an additional dataset for the finetuning process poses a significant hurdle for adoption. To address this constraint, we propose Sample-Wise Instance Fine-Tuning (SWIFT), a novel forecasting adaptation framework that achieves fine-grained specialization by leveraging structural priors from the target instance's own history. This shift allows the fine-tuning process to be entirely transparent to the end-user, which significantly enhances the ease of adoption. To implement this framework, SWIFT synergizes Self-Retrieval for historical mining and Frequency-Aware Loss for spectral consistency, while employing Deviation-Aware Dynamic Fusion to balance adaptation with general knowledge preservation. Extensive experiments across four TSFM backbones and three long-history benchmark subsets validate SWIFT, yielding average MASE and WQL reductions of 6.41% and 4.83% across 24 combinations of four backbones and six Chronos Benchmark II datasets. To the best of our knowledge, SWIFT is the first framework to investigate sample-wise instance fine-tuning of time-series foundation models, specializing the model for an individual prediction window using only the corresponding forecasting sample's own observed history.
Nov 13, 2026

FlexCast: Flexible Context Learning for Time Series Foundation Models

Time series foundation models (TSFMs) based on autoregressive (AR) next-token prediction entangle two factors: variable-context conditioning and prediction-unit factorization with autoregressive rollout. Strong AR performance therefore does not reveal whether generalization stems from the flexible training distribution, the local prediction target, the left-to-right inference process, or their combination. We propose FlexCast, a flexible context-target learning framework that disentangles these factors. FlexCast reformulates forecasting as variable-length masked prediction over time series patches: during pretraining, it dynamically samples visible context and masked target patches, exposing the model to diverse context-target configurations without sequential decoding. At inference time, the same pretrained model supports one-step forecasting, confidence-based iterative refinement, and next-token-style generation through a unified mask-and-predict interface, enabling controlled comparison of inference schedules under identical architecture, data, and objective. Evaluated on BLAST and multiple zero-shot benchmarks, FlexCast achieves accuracy competitive with or better than strong TSFM baselines while substantially reducing inference latency. Ablations confirm that context diversity and contiguous target masking drive performance, while left-to-right decoding does not yield consistent gains once flexible context-target relationships have been learned, suggesting that the key benefit of AR-style TSFMs lies in flexible training rather than sequential inference. The code is available at https://github.com/BenchCouncil/FlexCast.
Nov 12, 2026

Universal Infectious Disease Forecasting with Mantis 2

Public health agencies need to anticipate how outbreaks will unfold, when they will peak, and how long demand on health services will remain elevated. We previously developed a foundation model, Mantis, which was trained entirely on simulated outbreaks---enabling us to generate the diversity and scale of training data needed to support forecasting across diseases. However, our previous work left unresolved when simulation improves predictions and how it should complement observational data. Here, we varied the balance of simulated and observed training data while holding forecasting tasks fixed and varying the available surveillance history. We found that simulation-rich training was particularly valuable with limited local history, and its relative advantage diminished as more observations became available. We also found that simulation-based training reduced exceptionally large forecast errors, suggesting that mechanistic knowledge can help compensate for limited local evidence. Building on these findings, we developed Mantis 2, which combines simulated and observational training to jointly forecast up to six variables sixteen weeks ahead. Across 30 forecasting tasks, Mantis 2 achieved the best aggregate probabilistic accuracy over short, medium, and long horizons among the machine learning, statistical, and baseline models evaluated. It missed roughly 30% fewer outbreak peaks than the strongest comparison model, with successfully anticipated peaks identified more than six weeks in advance on average. We publicly release Mantis 2, enabling public health agencies and researchers to forecast disease outcomes without developing a separate model for each disease or location.
Nov 12, 2026

Time Flies When You Parallelise: Pareto-efficient Time Series Foundation Models with Linear Recurrences

Time series foundation models have demonstrated strong zero-shot forecasting performance, but computational efficiency remains a central architectural challenge. Temporal self-attention incurs quadratic cost in the number of temporal tokens, while nonlinear recurrent state transitions restrict parallel computation along the temporal dimension. We introduce ci-proto, a family of time series foundation models based on Mamba-3, whose linear state recurrences support chunk-wise parallel-in-time computation during training, and inference costs that scale linearly with sequence length. The architecture alternates recurrent temporal processing with cross-variate attention, supporting multivariate probabilistic forecasting with past-only and future-known covariates. Across fev-bench's 100 forecasting tasks, ci-proto XL sets a new state of the art among time series foundation models. The five ci-proto variants also make up the entire Pareto frontier for mean scaled quantile loss and mean absolute scaled error against both parameter count and median runtime.
Nov 11, 2026

From Forecasting to World Modeling: Can Temporal Foundation Models Recover Causal Response?

A model that forecasts sales accurately can still mislead a retailer about how sales respond to a price change. We evaluate released tabular and time-series foundation models, as well as a foundation model pretrained specifically to estimate causal effects. We report three findings. First, forecast accuracy does not indicate the accuracy of the predicted effect of price, both on simulated data in which the true effect is known and on real retail sales records. Second, for the tabular foundation models, additional context improves forecasts but not the predicted effect of price. Third, instrumental variables, the standard remedy in econometrics, reduce this error only when supplied in a specific form and increase it for every model under confounding when supplied as raw instruments.
Nov 11, 2026

A Cardiorespiratory Foundation Model Pretrained on Ten Physiological Signals for Perioperative and Critical Care

Arterial, venous, pulmonary artery and intracranial pressures and ventilator waveforms guide decisions in operating rooms and intensive care units, yet foundation models for physiological waveforms have been built almost entirely on the electrocardiogram and photoplethysmogram. We present CARMEN, a single encoder pretrained without labels on ten cardiorespiratory waveforms (2.27 million modality-hours). CARMEN accepts any combination of co-recorded signals from 10 s to 5 min, and a conditioning mechanism, LSCNorm, restores the absolute signal level that normalization removes; without it, the AUROC for intraoperative hypotension fell from 0.894 to 0.780. Across 19 tasks from Korean and US cohorts, the frozen encoder outperformed supervised models on 12 of 15 and three foundation models, one general-purpose and two for biosignals, on 14 to 17 of 19, with none exceeding it on any task. Given the electrocardiogram or photoplethysmogram alone, it matched or exceeded signal-specific foundation models in most comparisons, and it retained its performance when channels were missing.
Nov 10, 2026

TimeSeek: Multi-Task and Multi-Domain Pretraining for Generalized Time Series Analytics

Time series analytics has been widely applied across various domains. While existing time series foundation models (TSFMs) have demonstrated promising results, they still face two major challenges: 1) Most TSFMs mainly rely on stacking Transformer layers to increase parameter size, lacking the ability to adaptively model complex time series patterns from multi-domain data. 2) Current TSFMs are primarily pre-trained for specific tasks, e.g., forecasting or anomaly detection, and a few models support multiple tasks, but they fail to fully leverage task specificity and commonality, resulting in suboptimal performance. To address these challenges, we propose TimeSeek, a novel generalized time series analytics model that adaptively models time series patterns from multi-domain data and unifies multiple tasks in pre-training. Conceptually, TimeSeek is guided by selective information sharing under domain and task heterogeneity. From the perspective of multi-domain data, we propose a Multi-Domain Feature Extractor that adaptively captures temporal dependencies and channel correlations based on domain-specific characteristics, including progressive multi-scale modeling and adaptive slot channel modeling. From the perspective of the diversity of tasks, we propose multi-task collaborative training, which leverages task specificity and task commonality through task customization and cross-task training, effectively unifying multiple tasks in pre-training and better adapting to diverse downstream scenarios. After pre-training on large-scale time series data, TimeSeek achieves state-of-the-art performance in forecasting, anomaly detection, and classification tasks across both few-shot learning and zero-shot situations, together with favorable inference efficiency.
Nov 10, 2026

Quantifying the Sampling Frequency Invariance of Time-Series Foundation Models for Sensor Data Applications

Time-Series Foundation Models (TSFMs) are increasingly used in sensor applications, where sampling frequency (SF) varies across sensors, devices, and operating modes (e.g., power saving). Ideally, TSFMs must remain effective during such SF changes, provided the SF retains task-critical bandwidth. In this paper, we stress-test 4 prominent TSFMs: MOMENT, Chronos-2, Moirai-MoE, and TimesFM, to quantify their SF-invariance and investigate whether their behavior tracks the resulting loss of information bandwidth rather than the SF value itself. Our stress-test probes 3 dimensions: (i) the shift in embeddings, (ii) the degradation in forecasting performance, and (iii) the degradation in downstream performance, quantified by three novel metrics: FSI, FDI, and CDI, respectively. To our knowledge, this is the first study to place SF-invariance at the core of TSFM evaluations. Our results characterize: (i) how SF-invariance depends on the margin between signal's native-SF and the bandwidth its task requires, (ii) how much of that invariance the choice of TSFM explains, and (iii) how much of TSFM's performance degradation under a lower SF reflects information loss rather than a limitation of their abilities. These findings reframe SF-invariance as a measurable, signal-aware property of the (signal X TSFM) pair, and guide practitioners on when TSFMs display SF-invariance.
Nov 10, 2026

The Unused Input: The Initial State as a Conditioning Interface for Recurrent TSFMs

Time series foundation models (TSFMs) forecast unseen series without task-specific training. Increasingly they are built on recurrent backbones, whose constant per-step cost makes long-context inference cheap, and which, unlike transformers, carry a state. A recurrent backbone computes its prediction from two inputs: the observed context window and the initial recurrent state. In practice only the first is utilised and the initial state is unused by convention. We ask whether the initial state can be utilised in recurrent TSFMs. Using the xLSTM-based zero-shot forecaster TiRex as a case study, we freeze the backbone and train only a small head that writes a context vector into the initial state, so that any change in accuracy is attributable to the injected state rather than to adapted weights. A spectral summary of the input window, learned once and applied unchanged to every series, lowers mean GiftEval MASE by 0.175%, with every seed improving, which neither a parameter-matched LoRA adapter nor a learnable static state reproduces. An embedding of text attached to each series, evaluated on six MoTime datasets, reveals two roles for the state. As a per-dataset adaptation surface, a static state lowers MASE by 0.04-6.96% of the zero-shot baseline (8-70% of the gain of full fine-tuning). As a channel for exogenous information, the text-conditioned state lowers it by a further 0.15-0.41% of the zero-shot baseline on four of six datasets and, combined with a LoRA adapter, exceeds full fine-tuning, which has no access to the text, on one of them. In TiRex, the effect concentrates in the mLSTM blocks and is strongest at short context. The results are preliminary and the improvements small, but they point to the initial state as a direction for further research.
Nov 9, 2026

Do Time-Series Foundation Models Forecast Index-Futures Returns? A Preregistered Evaluation of Models, Metrics, and Input Representations

We ask whether the apparent forecasting skill of frozen time-series foundation models (TSFMs) on intraday index-futures returns is incremental skill. In a pre-registered evaluation on E-mini Nasdaq-100 (NQ) and E-mini S&P 500 (ES) futures, a common panel of five zero-shot models (Chronos-Bolt, Chronos-2, TimesFM 2.5, Moirai 2.0 and TiRex), plus separate experiments with Kronos, is compared with a no-change forecast at 5-120 minutes, using native prices, log returns, volatility-standardised returns and a volume clock as inputs. On a blind set opened once, 0 of 160 model-representation-horizon-market cells beat no-change, and a Hansen SPA test gives p = 1.000 on both markets; a later retrospective hold-out, scored as a descriptive check, gives 0 of 120. Input representations change forecast magnitude but not accuracy, and a transformation-aware ridge regression does as well. Several standard measures nevertheless suggest skill. Price-level correlation exceeds 0.9996, but no-change scores higher. Within-day RankIC across forecast origins rewards a trivial revert-to-mean rule (+0.35 at 30 minutes); we derive the expected within-day covariance under a random walk in closed form, and a sign-randomised null reproduces about 95% of the observed value. One model beats HAR for 5-minute volatility, but not the EWMA its input already contains. On strongly planted synthetic signals the models recover part of the signal at 5 minutes, and recovery deteriorates at 30 minutes. Benchmark-level success of TSFMs should therefore not be assumed to carry over to low-signal domains such as intraday returns.
Nov 8, 2026

Means vs. Foundation Models for Emergency Call Forecasting

This study investigates whether generalized means can achieve forecasting performance comparable to that of TimeGPT for daily emergency call volume prediction. Specifically, two families of generalized means, the Hölder means (HM) and the Lehmer means (LM), are evaluated against fine-tuned TimeGPT across forecasting horizons ranging from one to seven days. Forecasting performance is assessed on a real-world emergency call dataset using multiple error metrics, and Diebold-Mariano test. Experimental results show that HM and LM achieve nearly identical predictive performance across all forecasting horizons. For short-term forecasting, both generalized means outperform TimeGPT, with reductions of mean absolute error (MAE) approximately 1.24% and 2.48% at horizons of one and three days, respectively. Conversely, TimeGPT achieves a lower error at the seven-day horizon, with an descresing of MAE by 1.98%. However, Diebold-Mariano tests reveal no statistically significant differences between the methods at the evaluated horizons. These findings suggest that generalized means can provide competitive forecasting performance compared with a energyvore foundation model.
Nov 8, 2026

Daily Emergency Call Volume Forecasting using Univariate and Multivariate TimeGPT

This study investigates the potential of TimeGPT, a pretrained foundation model for timeseries, for forecasting daily emergency call volumes. Real-world emergency call data, comprising 2,101 daily observations, are used to evaluate both univariate and multivariate forecasting settings under one-step-ahead and seven-day multi-step-ahead scenarios. The multivariate forecasting incorporates temporal, meteorological, demographic, and socioeconomic variables. A chronological training, validation, and test split is adopted to prevent information leakage, while hyperparameters are selected using the validation set through grid search and early stopping when applicable. Forecasting performance is evaluated using multiple error metrics. Its forecasting performance is compared with several established forecasting approaches, including Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), Poisson regression (PR), Negative Binomial regression (NBR), Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM). The results show that SARIMAX achieves the best overall performance in the univariate setting. In the multivariate setting, NBR and PR provide best performance. These findings indicate that the large-scale pretraining of TimeGPT do not necessarily translate into superior forecasting accuracy for emergency call volumes. Frugal methods trained solely on the available data remain better for this application.
Nov 8, 2026

StatPrompt-LLM: A Lightweight Statistical Refinement for Foundation-Style Time-Series Forecasting

Time-series forecasting has recently seen a paradigm shift with the introduction of foundation-style models that adapt large pretrained language architectures to sequential data. However, effectively exposing fundamental temporal structures, such as trend, scale, and temporal dependence, to these models remains an open challenge, particularly when the available observations are limited. In this paper, we conduct an exploratory empirical study across different generations of forecasting architectures and propose StatPrompt-LLM, a lightweight statistical refinement of Time-LLM. StatPrompt-LLM augments the existing Time-LLM prompting mechanism with dataset-level statistical information and introduces a zero-initialised autoregressive bypass as a complementary linear forecasting pathway. We evaluate our framework on electricity and meteorological time-series benchmarks, including ETTh1, ETTh2, and Weather, under full-prefix and reduced-prefix configurations. Under the 10% reduced-prefix configuration, StatPrompt-LLM achieved lower in-sample prediction errors than Time-LLM, with unweighted average relative reductions across datasets of 40.2% in MSE and 21.4% in MAE. Full-prefix results were more dataset-dependent. Ablation analyses revealed relatively modest and dataset-dependent changes when selected additional statistical descriptor groups were masked within the combined architecture. Computational profiling indicated additional model parameters and runtime overhead relative to Time-LLM.
Nov 6, 2026

Can Time Series Foundation Models Serve as Surrogates for Power Systems Dynamic Simulation?

High-fidelity dynamic simulation is essential for scientific and engineering applications, but its computational cost limits large-scale scenario screening. We investigate the use of pretrained time-series foundation models (TSFMs) as computationally efficient surrogates to address this limitation. In-context learning and post-training strategies, including retrieval-augmented reference sample selection and physics-informed regularization, are considered in the investigation. Results are benchmarked against a traditional AI model representative of the state-of-the-art in the domain. Experiments using an open-source TSFM and data from a simulated 9-bus power system indicate promising results. Retrieval augmentation improves in-context learning performance, and fine-tuning leads to superior results compared to the traditional AI model, especially when including physics-informed regularization. The fine-tuned TSFMs achieve lower errors and improved computational costs compared to the traditional AI model, and a 563×\times speedup compared to the original physics-based simulation.
Nov 6, 2026

TimeOmni-Live: Continuous & Live Time Series Understanding with Long-Term Memory

Existing time series language models predominantly follow an offline, one-shot paradigm: they process fixed-length inputs independently and do not retain past states. This static approach fundamentally misaligns with the continuous nature of real-world streaming applications and fails to maintain the historical context necessary for cross-window reasoning. To bridge this gap, we introduce Continuous & Live Time Series Understanding (CL-TSU), a novel paradigm that shifts the focus toward maintaining a persistent, stateful understanding over open-ended data streams. To systematically evaluate this paradigm, we present LiveTS-Suite, the first comprehensive data suite tailored for CL-TSU, supporting data generation, model training, and systematic evaluation across 18 tasks spanning six core capabilities. It comprises over 5K streams totaling 2.2B timesteps, annotated with 15K specific response events. We further design an evaluation protocol that jointly assesses response timeliness, content accuracy, and computational efficiency. Building on this foundation, we propose TimeOmni-Live, the first live time series understanding model. It integrates a hybrid time series encoding for natively temporal input, a stream-persistent hybrid memory mechanism to effectively preserve exact historical evidence and long-term task states, and an elastic stream-inference synchronization strategy to reduce processing lag. Evaluations show that TimeOmni-Live establishes a strong baseline for CL-TSU, outperforming frontier LLMs (90.0 vs. 79.3 for Opus 5) with only 4B parameters and achieving 206×\times lower p95 latency than its backbone. Deployment on a Jetson edge device further shows its ability to support real-time continuous temporal understanding locally at the edge.
Nov 6, 2026

TailDART: Compute-Efficient Foundation Models for Financial Time-Series Forecasting

Although time-series foundation models (TSFMs) achieve strong zero-shot transferability across high-information-density physical systems, applying overparameterized TSFMs to quantitative finance is fundamentally constrained by low signal-to-noise ratios, non-stationarity, and prohibitive autoregressive decoding latency. To reconcile representation learning with institutional operational constraints, this study presents TailDART, a dense 2.68 million-parameter foundation model engineered for financial time-series forecasting. By integrating grouped-query attention into a lightweight backbone, the architecture drastically reduces Key-Value cache memory bandwidth overhead, achieving major inference acceleration (up to 48.1%) over prevailing state-of-the-art TSFMs to enable low-latency, high-throughput streaming execution. Furthermore, multi-asset benchmarks across varying temporal resolutions show that a compact model can remain competitive for market forecasting: TailDART matches the predictive fidelity of substantially larger baselines while demonstrating superior error resilience in high-volatility regimes.
Nov 6, 2026

PhysioWave-T: Token-Efficient Wavelet Foundation Models for Biosignals

We present PhysioWave-T, a token-efficient wavelet foundation model family for EEG, ECG, and sEMG. Dynamic scale folding recombines J+1 wavelet subbands prior to tokenization, compressing sequence length from (J+1)CS to CS tokens for J levels, C channels, and S temporal patches. Route-specific front ends and learned channel-name embeddings allow one Transformer per modality to handle diverse montages and sampling rates, while unseen setups reuse the backbone via lightweight adapters. We pre-train a 30M-parameter EEG model on over 15M windows (19--128 channels), alongside ECG (30M) and sEMG (12M) encoders. Across four EEG routes, scale folding cuts Transformer latency by 4.0--7.2x and peak memory by 6.8--12.7x over scale expansion without accuracy loss. Our EEG encoder leads 8 of 9 metrics against BENDR, BIOT, LaBraM, and EEGPT, and surpasses LUNA-L on wearable seizure detection with fewer parameters. Reliability-aware latent fusion (RALF) further integrates summary tokens from independently pre-trained encoders via sample-adaptive weights. On SeizeIT2, RALF raises event sensitivity at 11 false alarms/h from 61.4% to 74.4% and window AUPRC from 0.321 to 0.399 over the best unimodal baseline, outperforming weighted-logit and concatenation fusion across all variants. PhysioWave-T also achieves state-of-the-art sEMG accuracies and top macro-F1 on PTB-XL Super and Chapman-Shaoxing. Code and models: https://github.com/ForeverBlue816/PhysioWave.
Nov 5, 2026

EnergyBench: a live, gate-aligned benchmark of day-ahead forecasts for European power markets

Energy forecasting models are usually compared on backtests and fixed public datasets. Backtests can use data that did not yet exist when the forecast would have been made, and a fixed test set, reused across studies and possibly contained in the pretraining data of foundation models, invites overfitting to the benchmark. EnergyBench evaluates every entry live instead. Each day at 11:30 CET/CEST, thirty minutes before the day-ahead (SDAC) gate closes, every model forecasts the next day's price, wind and solar generation and load on the 15-minute grid for six bidding zones (BE, NL, FR, DE-LU, DK1, DK2). Unlike existing live benchmarks, EnergyBench runs the models itself, on one information set that it freezes at the cut-off and publishes, including three numerical weather prediction models, so that foundation models can be compared with and without the same weather covariates under the actual market deadline. Errors are pooled per series against a seasonal-naive baseline, and models are ranked by a Bradley-Terry rating on the Elo scale with a day-bootstrap interval. The field spans baselines, linear and gradient-boosted models on the frozen weather, eighteen time-series foundation models from eleven families, ensembles, and the grid operators' own forecasts. Over the first thirty-two live days, the covariates reduced the error of each of the ten foundation models run with and without them by 37-51% on matched races (64-73% on wind, 48-61% on solar). Used zero-shot, these models match or outperform a LightGBM model trained on the same covariates. Model size, from 3M to 331M parameters, has a much smaller effect, and median ensembles of the covariate models rank highest. The grid operators' pre-gate forecasts remain ahead on wind and solar, where the best single foundation models have 16-22% more error; on load the gap is 4-6%, within the intervals.
Nov 5, 2026

IMUDINO: Self-distilled Foundation Model for Inertial Sensing

We present IMUDINO, the first foundation model (FM) for inertial sensing to achieve competitive performance across diverse sensing tasks including human activity recognition (HAR), health, motion capture, and odometry. Prior inertial FMs often focus on one subfield of inertial sensing and sometimes underperform state-of-the-art supervised methods (e.g., in HAR), making their scope considerably narrower than FMs for vision and natural language. Moreover, they often require fixed input length and sampling rate, and generalize poorly to unseen sensor locations. To build a general-purpose foundation model for inertial sensing, we adapt the DINO self-supervised learning (SSL) paradigm to inertial data through a set of semantics-preserving motion transformations that simulate variations in motion pattern and sensor setup. For model architecture, we use a plain transformer encoder with frequency-aware positional encoding that natively supports varying input length and sampling rate. This simple setup proved remarkably effective for learning global and dense features generalizable across users, sensor setups, tasks, and, in the case of robot odometry, embodiments. We hope IMUDINO can serve as a backbone for inertial sensing and kickstart the development of general-purpose inertial FMs.
Nov 4, 2026

Big Models, Small Gains: Frozen Forecasters as Classifiers on MONSTER

Forecasting foundation models are increasingly used as frozen feature extractors for time series classification, but it remains unclear whether their pretrained representations offer an advantage over strong classifiers that do not require pretraining. This raises an important question: do these representations provide useful information beyond what the model architecture alone can capture, particularly when labelled data are scarce? We present a controlled benchmarking of two forecasting models, Reverso and Chronos-Bolt, on six large datasets from the MONSTER benchmark, using 1%, 10%, and 100% of the available training labels. We compare their frozen representations against four established time series classification methods and random-weight versions of the same forecasting models. Our results show that pretraining improves classification accuracy over random-weight counterparts on five of the six datasets, with gains of up to 23 percentage points. However, the frozen representations do not reliably outperform the strongest non-pretrained baseline in any setting: they perform worse in 15 of 17 settings and tie in the remaining two. Further experiments show that restoring amplitude information recovers part of the performance gap, while fine-tuning and forecast-error features do not close it. These findings highlight the importance of strong task specific baselines and random weight controls when evaluating forecasting foundation models for classification. Our code is available at https://github.com/Rachet2006/frozen-forecasters-monster.
Nov 4, 2026

Decouple to Reuse: Fine-Grained Knowledge Augmentation for Time Series Foundation Models

Time series forecasting is critical for demand prediction and resource management in modern Web services, yet applying pretrained Time Series Foundation Models (TSFMs) to downstream tasks remains challenging due to domain shifts. Existing retrieval-augmented approaches typically retrieve coarse-grained window contexts and may introduce context-specific noise. To address this problem, we argue that temporal patterns should be reused at a decomposition level, as effective trend and seasonal dynamics may originate from different contexts. We propose \method, a two-stage Fine-grained Knowledge Augmentation framework that decouples and consolidates pre-training knowledge into compact, task-shared residual codebooks to enhance frozen TSFMs. Specifically, Stage I constructs dual-branch residual codebooks for decoupled trend and seasonal signals, while Stage II dynamically retrieves and merges these fine-grained patterns into target representations via task-adaptive aggregation and gate fusion. Extensive experiments across eight real-world benchmarks with four representative TSFM backbones demonstrate the superior effectiveness and computational efficiency of \method, while further validating the strong transferability of the residual codebooks across different backbones. The source code is available at the anonymous repository: https://github.com/yuppielqx/TSFM-Component-Token-Adapter
Nov 3, 2026

TSFM Meets LLM: Context as Covariates

Textual information is ubiquitous in real-world time series datasets, appearing as news, reports, and annotations that hold key predictive signals not deducible from numerical history alone. Large Language Models (LLMs) interpret text but lack numerical precision, while Time Series Foundation Models (TSFMs) are quantitatively accurate but cannot reason about the effects of text. Moreover, the lack of large-scale paired multimodal datasets limits progress in bridging this gap. In this work, we address this challenge by providing a scalable approach for generating realistic multimodal datasets by annotating real-world time series with LLMs. Building on this, we propose Context as Covariates (CoCo), a forecasting framework in which an LLM distills its textual reasoning into numerically grounded forecast and confidence covariates that guide a TSFM backbone, thereby exploiting the strengths of both LLMs and TSFMs. We align the two models in stages, first fine-tuning the LLM via Group Relative Policy Optimization (GRPO) with a reward tied to covariate-induced improvement in forecasting performance, and then fine-tuning the TSFM on covariates generated by the aligned LLM. Our experiments show that models trained on our synthetic multimodal corpus generalize to real benchmarks spanning finance, economics, weather, traffic, and security. Notably, CoCo with Qwen3-4B-Instruct-2507 as the LLM and Chronos2 as the TSFM achieves up to 12% MAE improvement over state-of-the-art TSFMs and LLM forecasters, such as GPT-4o. Code, checkpoints, and the evaluation datasets are publicly available.
Nov 2, 2026

Evaluating Uncertainty Calibration in Probabilistic Time Series Foundation Models

Time series foundation models (TSFMs) are increasingly used for probabilistic forecasting across diverse domains, yet the reliability of their uncertainty estimates remains poorly understood. In this work, we study uncertainty calibration in probabilistic TSFMs from a broader methodological and empirical perspective. We introduce a taxonomy of how TSFMs represent predictive uncertainty and develop a fine-grained evaluation framework for calibration in multi-horizon forecasting that distinguishes pooled, horizon-wise, and stronger conditional notions of calibration. Using this framework, we conduct a comprehensive empirical evaluation of popular TSFMs across diverse datasets and show that current models exhibit distinct and systematic patterns of miscalibration that standard metrics often obscure. We further analyze common post-hoc recalibration methods and characterize which forms of calibration they can improve and which tend to persist. Our results provide practical guidance for evaluating and improving uncertainty calibration in TSFMs and highlight important limitations of current practice.
Nov 2, 2026

Control Foundation Models across Changing Block Diagrams: Generalizing Time-Series Prediction to Dynamic Graphs

In the prediction and control of physical systems, combining records from different experiments and operations is expected to supply information that no single experiment provides on its own. Time-series foundation models, which learn a shared predictor from diverse time series, offer a framework for treating such records as a common training resource. Moreover, query-based prediction models, which switch between known and target signals within the same time series, make it possible to handle response prediction and control-input generation in a unified manner. However, in control-experiment data, the connections between signals change with the experimental stage, such as system identification, filtering, and feedback control, which calls for a formulation that handles signal values and connection structures dynamically and in a unified way. We formulate a control foundation model that defines a data space of dynamic graphs containing connections, signal values, and missing values, together with an operation on this space that specifies the known signals and the prediction targets. This allows response prediction and control-input generation to be described as a common masked prediction problem, even for experimental data with different connections and recorded signals. The Block Diagram Transformer (BDT), which implements this formulation, combines attention along the time axis with a graph neural network that follows the block-diagram connections, and can be initialized with the pretrained parameters of existing time-series foundation models. Trained on control-experiment data that mix multiple experimental settings, the proposed model stabilized an unstable system through feedback control, which is difficult to achieve by fine-tuning existing time-series foundation models. Ongoing work applying this approach to a wide range of target systems is also yielding promising results.
Nov 2, 2026

Zero-shot time-series foundation models against the Khepri v1.5 baselines for 96-hour carbon-intensity forecasts in nine Nordic bidding zones: a preregistered comparison

We compare three zero-shot time-series foundation models (Timestral-5M, Timestral-209M and Chronos-2, 8192-hour context, no covariates) with the Khepri v1.5 SARIMA and GBM baselines, with their code and hyperparameters unchanged, for 96-hour forecasts of hourly carbon intensity in the Nordic bidding zones NO1-NO5 and SE1-SE4. The design, the analysis code and a minimum detectable effect (MDE) were timestamped before any foundation-model forecast by an OpenTimestamps proof (Bitcoin-confirmed) and the GPG-signed tag prereg-v1; the proofs are archived in Zenodo (DOI 10.5281/zenodo.23881315). The preregistration was also submitted to OSF Registries 2026-11-02 07:49 UTC (osf.io/hncz5; embargo requested, pending OSF approval). On the 2025 test (361 daily origins, contamination not ruled out) all three models have a lower zone-averaged MAPE than both baselines, by 0.952 to 1.347 percentage points (Diebold-Mariano with the Harvey-Leybourne-Newbold correction, all p < 0.0001), and each beats SARIMA in 9 of 9 zones. On a post-release holdout of 14 origins from 2026-10-16 the point estimates have the same sign (0.729 to 1.326 percentage points lower), but all lie within the preregistered MDE of 2.989 points and are reported as not determined. For calibration (RQ4) the preregistration locked a Diebold-Mariano-HLN test on CRPS; the MDE applies to MAPE only. Holdout CRPS point estimates (n = 14) are 1.1114 (Timestral-5M), 1.0733 (Timestral-209M) and 1.1198 (Chronos-2) against 1.4165 gCO2-eq/kWh for SARIMA (each p < 0.0001). The SARIMA quantiles are constructed from the fitted model's Gaussian predictive distribution (deviation D-03), not taken from a native v1.5 forecast, so the difference may partly reflect the SARIMA intervals. The contamination indicator does not exceed its preregistered threshold for any comparison; with 14 holdout origins it has low power (HAC standard errors 0.124 to 0.360 pp).
Nov 2, 2026

SynForecast: Controlled Synthetic Time Series for Forecasting Research

Synthetic time series constitute a substantial share of the training corpora of several recent forecasting foundation models. Existing studies report benefits at small data budgets and modest synthetic proportions, but each evaluates individual combinations of model and generator. Studying these benefits systematically requires generators that support controlled configuration, perturbation, reproducibility and mechanism-level ablation, and the 11 existing generators we review show gaps in this combined support. We present SynForecast, an open-source Python library with 32 generators, 19 of which implement named mechanisms with a published source in econometrics, quantitative finance, statistical physics and point-process theory. Perturbations and mechanism families can be varied while everything else is held fixed, output comes in the long format that forecasting libraries accept, and output does not depend on the number of parallel workers. We evaluate SynForecast on M4 at five frequencies with a neural network and a gradient-boosted ensemble. A classical baseline attains the lowest error in nine of ten frequency-history combinations. A synthetic prior outperforms the same network trained from scratch at a fixed budget by up to 34% with a few dozen observations, but loses to it at full history at every frequency, and none of three pre-specified rescalings of history predicts where the sign changes. At the hourly frequency, where the prior fails at every history, the corpus lacks daily cycles; adding them reduces zero-shot error by 43 to 63% for both models. Fine-tuning outperforms training from scratch by up to about 15%, but at short histories a fixed 300-step fine-tune is worse than the pretrained model used as is. Removing the classical mechanism family costs about 2% for both models; every other family has a smaller effect outside the hourly frequency.
Nov 2, 2026

Investigating the effects of the patch size in Time Series Foundation Models

Patch size is a central design choice in patch-based time-series foundation models, controlling temporal resolution, sequence length, and computational cost. We study how this choice interacts with transformer architecture through a controlled comparison of encoder-only and decoder-only forecasting models trained with the same data mixture and optimization budget. Across 97 GIFT-Eval tasks, encoder-only models consistently benefit from smaller patches, whereas decoder-only performance follows a U-shaped trend, with an intermediate patch size performing best in aggregate. Dataset-level results further show that the preferred decoder patch size increases with the forecasting horizon. For encoder-only models, increasing context length improves performance across patch sizes, although the gains diminish beyond a context length of 2048. Finally, training-compute and CPU-latency Pareto analyses reveal complementary regimes: decoder-only models provide efficient low-cost alternatives, while encoder-only models attain the highest overall accuracy. These results show that the choice of patch size and architecture should be made depending on the forecasting horizon and deployment constraints.