WorksComputer Science
HieraNTP: Structured Risk Representations for Time-Series Anomaly Detection via Hierarchical Next-Token Prediction
Abstract
Time-series anomaly detection commonly converts deviations from learned temporal regularities into anomaly scores. This practice can conflate violations of temporal regularity with task-specific anomaly judgments: the same deviation may be anomalous under one task definition and benign under another. Aggregating heterogeneous deviations into a single score before task-specific adaptation can obscure the evidence needed to distinguish these cases. We introduce HieraNTP, a Hierarchical Next-Token Prediction framework that separates self-supervised predictive modeling from task-specific risk interpretation. HieraNTP pretrains a hierarchical tokenizer and a causal next-token predictor to capture temporal regularities at multiple scales. It then organizes prediction discrepancies into a structured risk representation, preserving complementary evidence across temporal scales in token, local-statistic, and observation spaces. A task-adaptive detector uses anomaly labels to select and combine this evidence for the target task. Across anomaly detection benchmarks, HieraNTP improves VUS-PR by 12.0% on TSB-AD-U and 15.5% on TSB-AD-M relative to the strongest baselines, and achieves the highest composite score on both TAB tracks, with controlled comparisons supporting the value of retaining distinct evidence. Our code is available at https://github.com/blisky-li/HieraNTP.
Introduction
Time-series anomaly detection (TSAD) identifies observations or intervals that depart from expected temporal behavior. It supports continuous monitoring of cloud services, industrial systems, and cyber-physical infrastructure, where anomalous events may signal service failures, equipment faults, or attacks [52, 46, 17] (Su et al. 2019).
Existing TSAD methods commonly learn temporal regularities and detect anomalies directly from forecasting or reconstruction errors [52, 3, 46, 40] (Su et al. 2019; Liu & Paparrizos 2024). Association- and contrastive-learning methods measure inconsistencies in temporal dependencies or representations [53, 54, 19, 14]. Temporal pretraining learns transferable forecasting and representation capabilities [13, 2, 5, 14, 17]. Larger departures generally indicate stronger anomaly evidence.
Direct discrepancy scoring can conflate violation of learned regularity with task-defined anomaly. The same deviation may be anomalous under one task definition and benign under another. Error magnitude alone may not capture the form and context relevant to these judgments. Contextual anomalies can disrupt temporal relations with small value deviations [19], while persistent abnormal dynamics may become predictable within context, reducing forecasting errors [47]. Forecasting or reconstruction gains may yield little improvement in detection, and simple local statistics can closely approximate foundation-model errors [59]. We refer to this mismatch between task-agnostic predictive evidence and task-specific anomaly interpretation as the predictive risk–decision gap.
Task-specific interpretation depends on the information retained in predictive evidence. Figure 1 shows two types of deviation within one task: missing local peaks in region A and disrupted cycles in region B. Although both violate the preceding pattern, the illustrated task labels only region B as anomalous. Distinguishing them requires evidence about how each departs from that pattern. Scalar reduction can discard these distinctions before a decision boundary is learned, motivating the retention of complementary discrepancies across spaces and temporal supports.
Figure 1. Schematic motivation using a Yahoo sequence from TSB-AD. The task treats missing peaks (A) as normal and disrupted cycles (B) as anomalous. (a) Early scalar reduction can obscure this distinction. (b) HieraNTP retains a structured risk representation for task-adaptive detection.
We introduce HieraNTP, a Hierarchical Next-Token Prediction framework that separates self-supervised predictive modeling from task-specific risk interpretation. Its multiscale tokenizer produces continuous causal tokens, and its hierarchical predictor models their evolution. Comparing predicted and realized targets yields token-space, local-statistic, and observation-space discrepancies. Temporal summaries preserve sustained and peak responses, while variable aggregation captures broad or concentrated departures. With the tokenizer and predictor frozen, target labels train the task-adaptive detector on the resulting structured risk representation.
On the official TSB-AD benchmark [24], HieraNTP outperforms the strongest baseline in VUS-PR on both the univariate and multivariate tracks. HieraNTP also achieves the highest dataset-mean VUS-PR on both TAB tracks [36]. Our contributions are:
We identify the predictive risk–decision gap and explain how scalar scoring can discard distinctions needed for task-specific anomaly decisions.
We propose HieraNTP, which converts multiscale next-token prediction discrepancies into a structured risk representation before task-specific anomaly scoring.
Under matched detector training, multiview evidence outperforms scalar time-domain prediction errors and mean, maximum, or mean–maximum aggregation on TSB-AD.
Related Work
Deviation-based time-series anomaly detection. Similarity-based detectors identify subsequences that lack close historical matches [56]. Generative detectors measure reconstruction discrepancies or likelihoods under learned normal dynamics [52, 46, 11] (Su et al. 2019; Aubert et al. 2020), whereas forecasting-based detectors compare predicted and realized observations (Liu & Paparrizos, 2024; Safraz et al., 2024). Association and contrastive approaches instead construct anomaly scores from inconsistencies in temporal dependencies or learned representations [53, 54].
Temporal pretraining and next-token prediction. Pretrained time-series models acquire transferable temporal structure through reconstruction or forecasting objectives [42]. MOMENT uses masked reconstruction, Chronos performs autoregressive prediction over discretized values, and TimesFM forecasts continuous patches [13, 2, 5]. Controlled evaluations find that their reconstruction or forecasting scores can resemble simple local statistics and that improved prediction quality can leave anomaly rankings unchanged [59]. Persistent abnormal dynamics can become predictable within context, concentrating forecasting discrepancies near event boundaries [47].
Anomaly-oriented pretraining and task adaptation. Anomaly-oriented models introduce detection-specific information at different stages of learning. DADA uses explicit normal–anomalous differentiation during pretraining [43]. TimeRCD learns relative-context discrepancy from synthetic sequences with context-dependent anomaly labels and applies the resulting detector to unseen datasets in zero-shot evaluation [19]. TSPulse combines reconstruction and prediction scores from multiple heads through fixed statistics, or selects a head using labeled tuning data. Its target-data fine-tuning uses self-supervised objectives [9]. Other adaptations retain a pretrained temporal model and construct anomaly scores from intermediate representation distances, supervised classifiers, or calibrated forecasting residuals [14, 17, 31].
HieraNTP retains multiscale discrepancies from hierarchical next-token prediction, comparing predicted and realized tokens, local statistics [49, 21], and decoded time-series segments. A task-adaptive detector learns joint decision rules over this structured risk representation from target labels, with the tokenizer and predictor frozen.
Method
Problem Formulation and Framework
Let denote a time series with observations and variables. HieraNTP assigns an anomaly score to each time index . As shown in Figure 2, a causal multiscale tokenizer and a hierarchical next-token predictor are pretrained with self-supervised objectives. Both are then frozen, and their prediction discrepancies form the structured risk representation . Labels supervise a task-adaptive detector that maps to .
Figure 2. Overview of HieraNTP. (a) Multiscale reconstruction pretrains the causal tokenizer. (b) Prediction and realized-target (GT) branches yield token-, local-statistic-, and observation-space discrepancies. (c) Labels train the task-adaptive detector with predictive modules frozen.
Multiscale Tokenizer Pretraining
Anomalies range from brief impulses to disrupted cycles, motivating a hierarchy of temporal tokens for each variable. For a normalized variable , the tokenizer encoding map , including the continuous-token projection, produces
Patch lengths are observations. Past-only local attention ensures causality at patch endpoints. Fine and coarse levels capture local variation and longer patterns.
Continuous tokens are regularized by a shared codebook, and a multiscale decoder reconstructs the original sequence and the temporal content at each native scale [28, 48]. Tokenizer pretraining minimizes
Reconstruction grounds tokens in observable content and supplies prediction targets at multiple supports. Token targets are the continuous encoder projections used for codebook lookup. The encoder, codebook, and decoder are then frozen.
Structured Predictive Risk via Hierarchical Next-Token Prediction
Learning regular temporal evolution. With the tokenizer frozen, the predictor estimates next tokens and local statistics from preceding observations. At target position , collects encoder states preceding the target interval at each hierarchy level. Prediction proceeds from fine to coarse levels: fine-level hidden states are merged by temporal overlap into coarser queries, then updated by causal self-attention and cross-attention to the corresponding historical states. Level-specific heads estimate
Self-supervised training matches the predicted token, patch mean, and log variance to targets from the tokenizer and observed patches.
Complementary views of predictive risk. The decoder maps predicted and realized tokens to time-series segments. Token-space discrepancies measure disagreement in the temporal content represented by predicted and realized tokens. Local-statistic discrepancies compare predicted and realized patch means and variances to measure departures in level and dispersion. Observation-space evidence comprises decoded-observation discrepancies against observed values and decoded-target discrepancies against reconstructions from realized target tokens.
Organizing predictive evidence. The four components in Figure 2(b) organize these discrepancies into the structured risk representation:
Multiscale Evidence. HieraNTP retains temporal content over six patch lengths, with token-grid offsets of 0 and 4 observations providing alternative boundaries. Native-scale and full-resolution comparisons yield 64 base discrepancy channels per variable.
Reference Calibration. Robust statistics from each sequence’s designated training prefix express responses relative to their baseline level and variability, accounting for differing error scales across sources and variables [6].
Temporal Summaries. Each evidence sequence retains its current value and centered means and maxima over six window lengths, yielding 13 temporal summaries. Means capture sustained responses, while maxima preserve local peaks that averaging can attenuate. These windows complement the patch lengths defining prediction targets by summarizing duration-dependent discrepancy responses [59].
Variable Aggregation. Variable means summarize broad responses, while maxima retain departures in a few variables. Pool-then-calibrate (PoolCal.) calibrates the aggregate, while calibrate-then-pool (Cal.Pool) calibrates individual variables before aggregation. Univariate aggregation is the identity.
These components form three parallel processing paths: Direct, PoolCal., and Cal.Pool. The Direct path projects reference-standardized patch responses over their target intervals before variable aggregation and within-sequence ranking. The PoolCal. and Cal.Pool paths hold each response until the next valid update in the processing direction. Full-resolution discrepancies remain pointwise. All three paths and their temporal summaries are computed in both directions. Let denote the base channels for variable , direction , and emission time . The index identifies the source, patch length or full resolution, and grid alignment. Combining the three processing variants with the 13 temporal summaries gives
where indexes observations and applies each variant’s time alignment and evidence transformations, followed by temporal summaries. The representation contains features for univariate sequences and 4992 for multivariate sequences with both variable means and maxima.
Task-Adaptive Anomaly Detection
HieraNTP addresses the predictive risk–decision gap by learning evidence combinations for target anomalies. With the tokenizer, decoder, and predictor frozen, an XGBoost detector {(citep}(Chen & Guestrin, 2016)} learns nonlinear decision rules from structured risk representations and anomaly labels. Each benchmark track fits a shared base detector across its training data.
Dataset-specific extensions continue boosting with the dataset’s labels, adding corrective trees while keeping the base fixed. All trees share the same risk representation, and their contributions are summed before the sigmoid. Inference uses the adapted detector when available and the shared base otherwise, averaging its outputs across both temporal directions at each timestamp. Complete-sequence ranks and centered temporal summaries define this offline configuration.
Pretraining Procedure
Multiscale tokenizer pretraining. We jointly train the causal encoder, shared codebook, and decoder on the QUALS-organized GIFT-Eval Pretrain corpus {(citep}(Li et al., 2027)} with Eq. (2). After lower-learning-rate refinement, the frozen tokenizer supplies encoder states as predictor context and projected continuous tokens as fixed targets.
Hierarchical next-token pretraining. Observed history conditions token and patch-statistic prediction at every scale. Teacher forcing uses encoded observations, whereas autoregressive rollout uses preceding predictions. Normalization statistics use only the observed history.
To encourage learning of normal temporal regularities, we continue predictor training with synthetic sequences and gradually increase real data from QUALS until the real-to-synthetic ratio reaches 4:1. Appendices B–C detail evidence construction, synthetic data generation, and model training.
Experiments
Experiments on TSB-AD and TAB test whether retaining predictive evidence improves task-specific anomaly detection. Controlled ablations examine evidence preservation and adaptation, while dataset-level analyses and visualizations investigate how detectors use complementary responses.
Experimental Setup
Benchmarks and supervision. We evaluate HieraNTP on TSB-AD {(citep}(Liu & Paparrizos, 2024)} and TAB {(citep}(Qiu et al., 2025a)}, covering univariate detection (TSB-AD-U and TAB-U) and multivariate detection (TSB-AD-M and TAB-M). TSB-AD provides 48 Tuning and 350 Eval series for TSB-AD-U, and 20 Tuning and 180 Eval series for TSB-AD-M. TAB provides official training/test splits for 1635 univariate series from 15 datasets and 220 multivariate series from 29 datasets.
Normal and anomaly labels from the TSB-AD Tuning sets and TAB training data supervise the shared base detector and dataset-specific extensions. TSB-AD uses sequence-level cross-validation within Tuning to select hyperparameters and extensions, with temporal holdouts for eligible singleton datasets. TAB follows its full-shot evaluation strategy with fixed hyperparameters shared across datasets. HieraNTP pretrains on QUALS, derived from GIFT-Eval Pretrain {(citep}(Li et al., 2027; Aksu et al., 2024)}, with no data leakage from TSB-AD or TAB. The tokenizer and predictor remain frozen.
Comparisons. TSB-AD comparisons follow the official Tuning/Eval split. Figure 3 compares leading leaderboard methods, and Table 1 reports results on the 16 datasets selected by TimeRCD ([24, 19]). Table 2 compares TAB results on the 14 univariate and 28 multivariate datasets in OpenTS, excluding OPP. These baselines combine leading leaderboard methods with deep-learning detectors. Classical baselines include KMeans, KNN, EIF, OCSVM, and DWT-MLEAD ([30, 38, 15, 41, 45]). Neural comparisons cover DLinear, iTransformer, TimesNet, ModernTCN, DUET, and CALF [57, 26, 50, 29, 37, 23], alongside TranAD, DCdetector, DualTF, ContraAD, and CATCH [46, 54, 32, 60, 51].
Figure 3. TSB-AD Eval VUS-PR. HieraNTP is red, target-adapted methods are blue, and zero-shot or training-free methods are green.
| Metric | Model | IOPS | MGAB | NAB | NEK | Power | SED | Stock | TODS | UCR | WSD | YAHOO | MSL | PSM | SMAP | SMD | SWaT | 1st | 2nd |
| Affiliation-F | HieraNTP | 93.68 | 67.67 | 96.53 | 95.25 | 68.39 | 96.20 | 97.86 | 87.57 | 86.13 | 97.96 | 95.19 | 83.37 | 77.87 | 89.91 | 84.47 | 73.41 | 07 | 02 |
| Affiliation-F | TimeRCD ⋄ | 86.98 | 74.54 | 93.81 | 83.95 | 85.02 | 88.99 | 41.31 | 84.86 | 84.34 | 91.00 | 95.94 | 84.66 | 82.31 | 86.01 | 91.75 | 69.43 | 03 | 06 |
| Affiliation-F | MOMENT ⋄ | 87.54 | 66.76 | 90.45 | 92.26 | 75.97 | 59.13 | 45.26 | 59.76 | 75.77 | 95.39 | 79.99 | 74.55 | 65.79 | 77.42 | 74.00 | 70.17 | 00 | 01 |
| Affiliation-F | TimesFM ⋄ | 81.88 | 66.95 | 79.73 | 90.49 | 69.88 | 67.14 | 97.53 | 89.08 | 70.03 | 78.97 | 91.28 | 20.35 | 71.24 | 45.44 | 62.85 | 44.37 | 00 | 02 |
| Affiliation-F | Chronos ⋄ | 90.12 | 67.89 | 86.66 | 93.63 | 69.72 | 67.89 | 96.85 | 91.96 | 74.35 | 90.98 | 96.34 | 75.52 | 70.88 | 72.22 | 75.31 | 70.43 | 02 | 02 |
| Affiliation-F | Moving Var. ⋄ | 83.90 | 67.56 | 91.50 | 77.81 | 74.77 | 67.14 | 67.96 | 71.57 | 78.44 | 89.13 | 78.00 | 89.43 | 73.91 | 88.14 | 81.33 | 75.55 | 01 | 01 |
| Affiliation-F | TranAD | 83.19 | 67.28 | 90.28 | 85.02 | 71.56 | 61.03 | 57.94 | 52.76 | 73.31 | 84.34 | 76.08 | 79.91 | 73.83 | 87.39 | 92.20 | 75.37 | 01 | 00 |
| Affiliation-F | USAD | 71.08 | 67.81 | 91.54 | 71.13 | 76.48 | 55.60 | 35.92 | 47.90 | 76.00 | 65.10 | 53.05 | 81.86 | 57.86 | 87.25 | 85.09 | 75.06 | 00 | 00 |
| Affiliation-F | OmniAnomaly | 80.32 | 67.35 | 92.35 | 86.30 | 78.16 | 61.26 | 75.24 | 50.73 | 73.53 | 78.02 | 71.31 | 83.15 | 58.17 | 91.38 | 85.82 | 73.39 | 01 | 01 |
| Affiliation-F | DCdetector | 71.83 | 67.91 | 72.21 | 62.31 | 69.75 | 72.20 | 55.79 | 57.81 | 70.18 | 72.79 | 67.77 | 67.74 | 67.32 | 67.10 | 69.55 | 71.07 | 00 | 01 |
| Affiliation-F | TFMAE | 78.25 | 67.50 | 75.99 | 76.91 | 70.30 | 68.17 | 56.39 | 62.83 | 70.60 | 80.25 | 76.87 | 75.70 | 70.07 | 75.36 | 70.85 | 75.72 | 01 | 00 |
| F1-T | HieraNTP | 53.84 | 2.30 | 72.34 | 75.51 | 19.43 | 65.33 | 95.62 | 64.24 | 37.77 | 64.88 | 78.04 | 49.87 | 27.19 | 54.21 | 44.66 | 46.04 | 07 | 02 |
| F1-T | TimeRCD ⋄ | 45.39 | 9.05 | 50.49 | 76.91 | 23.05 | 60.59 | 18.38 | 68.46 | 41.18 | 44.13 | 79.14 | 43.67 | 43.86 | 35.82 | 55.96 | 35.47 | 05 | 04 |
Table 1. Four-metric results on the 16 datasets selected by TimeRCD (%). ⋄: zero-shot. Red bold and blue underlined italics mark the best and second-best results per dataset–metric column.
| Metric | HieraNTP | KMeans | DWT-MLEAD | CALF | iTransformer | EIF | ContraAD | DLinear | TranAD | DC detector | DualTF | Modern TCN |
| F1 | 0.322 | 0.293 | 0.242 | 0.182 | 0.194 | 0.166 | 0.041 | 0.165 | 0.158 | 0.024 | 0.078 | 0.163 |
| Aff-F1 | 0.849 | 0.792 | 0.716 | 0.820 | 0.810 | 0.782 | 0.684 | 0.808 | 0.774 | 0.688 | 0.728 | 0.807 |
| AP | 0.419 | 0.356 | 0.291 | 0.231 | 0.243 | 0.213 | 0.033 | 0.213 | 0.201 | 0.034 | 0.083 | 0.210 |
| AUC-ROC | 0.817 | 0.790 | 0.739 | 0.687 | 0.687 | 0.633 | 0.499 | 0.659 | 0.545 | 0.473 | 0.539 | 0.663 |
| VUS-PR | 0.439 | 0.362 | 0.312 | 0.251 | 0.244 | 0.202 | 0.059 | 0.202 | 0.179 | 0.057 | 0.108 | 0.210 |
| VUS-ROC | 0.837 | 0.784 | 0.699 | 0.758 | 0.753 | 0.671 | 0.579 | 0.721 | 0.594 | 0.491 | 0.604 | 0.725 |
| G/S/B | 62/18/29 | 28/62/27 | 29/22/22 | 8/14/7 | 3/2/24 | 4/4/8 | 1/8/3 | 1/3/5 | 0/2/6 | 1/2/0 | 0/1/3 | 1/0/0 |
| Score | 251 | 235 | 153 | 59 | 37 | 28 | 22 | 14 | 10 | 7 | 5 | 3 |
| Metric | HieraNTP | KMeans | KNN | EIF | OCSVM | TimesNet | CATCH | iTransformer | DLinear | DUET | DualTF | DC detector |
| F1 | 0.331 | 0.296 | 0.179 | 0.210 | 0.161 | 0.199 | 0.191 | 0.183 | 0.176 | 0.200 | 0.109 | 0.046 |
| Aff-F1 | 0.793 | 0.698 | 0.699 | 0.773 | 0.677 | 0.789 | 0.778 | 0.770 | 0.775 | 0.780 | 0.593 | 0.637 |
| AP | 0.376 | 0.340 | 0.336 | 0.254 | 0.176 | 0.259 | 0.240 | 0.235 | 0.227 | 0.239 | 0.147 | 0.085 |
| AUC-ROC | 0.782 | 0.744 | 0.714 | 0.732 | 0.645 | 0.688 | 0.668 | 0.667 | 0.660 | 0.663 | 0.611 | 0.496 |
| VUS-PR | 0.388 | 0.302 | 0.301 | 0.258 | 0.309 | 0.271 | 0.261 | 0.251 | 0.240 | 0.258 | 0.174 | 0.118 |
| VUS-ROC | 0.794 | 0.685 | 0.707 | 0.741 | 0.656 | 0.722 | 0.705 | 0.705 | 0.695 | 0.699 | 0.621 | 0.503 |
| G/S/B | 108/46/25 | 56/64/25 | 60/21/35 | 27/40/25 | 41/19/15 | 26/17/40 | 10/15/27 | 7/16/24 | 9/14/10 | 4/6/22 | 7/4/10 | 2/4/7 |
| Score | 441 | 321 | 257 | 186 | 176 | 152 | 87 | 77 | 65 | 46 | 39 | 21 |
Table 2. TAB full-shot results. Red bold: best; blue underlined italics: second best.
Metrics and implementation. VUS-PR is the primary metric [35], supplemented by AUC-PR (average precision, AP), VUS-ROC, AUC-ROC, and F1 metrics including Affiliation-F [16]. For TAB, the composite score is : , , and count first-, second-, and third-place finishes across dataset–metric pairs for all 16 metrics, with ties sharing gold, silver, or bronze medals. Appendices C–E detail training, evaluation, and results.
Main Results
Figure 3 shows that HieraNTP achieves the highest VUS-PR on both official TSB-AD Eval sets, improving on the strongest baselines by 12.0% on U and 15.5% on M in relative terms. Across the 16 datasets in Table 1, it ranks first in VUS-PR on eight and second on four, and obtains the most first-place results for each F1 metric. Gains vary by dataset: on SMAP, VUS-PR exceeds the strongest baseline by 24.30 percentage points, while TimeRCD leads on TODS and SMD. Joint gains in ranking and F1 metrics indicate better anomaly ordering and threshold-based detection.
On TAB, HieraNTP leads on all six metrics in both tracks in Table 2. Dataset-mean VUS-PR exceeds the strongest baseline by 7.7 percentage points on univariate detection and 7.9 on multivariate detection. HieraNTP also ranks first in composite score, with 251 on U and 441 on M. Together, the two benchmarks support learning anomaly decisions from structured predictive evidence across different datasets and evaluation protocols, with the pretrained temporal model frozen.
Ablation Study
Table 3 examines evidence content, temporal organization, and adaptation with pretrained modules frozen. Supervised comparisons fix samples, labels, weights, and detector hyperparameters. Time-domain error is scored directly or supplied to the detector. Fixed evidence averaging takes each Pool→Cal. channel’s positive part before temporal summaries, then averages channels and directions without fitting a detector.
| Evidence construction | Temporal structure | Scoring and adaptation | ||||||
| Configuration | U | M | Configuration | U | M | Configuration | U | M |
| HieraNTP | 65.56 | 50.28 | HieraNTP | 65.56 | 50.28 | HieraNTP | 65.56 | 50.28 |
| Prediction-to-observation only | 59.78 | 44.60 | w/o patch 256 | 65.53 | 49.98 | Mean + detector | 38.20 | 29.12 |
| w/o token-space | 65.35 | 49.69 | w/o patch 128 | 65.52 | 49.29 | Max + detector | 32.11 | 27.47 |
| w/o patch mean | 64.31 | 50.33 | w/o patch 64 | 65.83 | 49.92 | Mean–max + detector | 44.51 | 33.83 |
| w/o patch variance | 62.47 | 47.25 | w/o patch 32 | 66.36 | 50.08 | Time-domain error | 33.85 | 30.81 |
| w/o observation-space | 62.54 | 47.77 | w/o patch 16 | 66.12 | 49.70 | Time-domain error + detector | 29.89 | 28.34 |
| w/o reference calibration | 60.28 | 51.71 | w/o patch 8 | 63.46 | 49.46 | Instantaneous multiview | 57.80 | 44.08 |
| Variable mean only | – | 47.55 | Single token-grid offset | 64.22 | 49.62 | Fixed evidence averaging | 43.65 | 35.27 |
| Variable maximum only | – | 47.41 | w/o temporal summaries | 59.25 | 45.63 | Shared base detector only | 61.48 | 49.91 |
| Forward only | 60.32 | 47.52 |
Table 3. Evidence and adaptation analysis on TSB-AD-U/M (VUS-PR, %).
Evidence content and compression. Under matched supervised training, the complete representation outperforms mean, maximum, and mean–max inputs, improving VUS-PR over mean–max by 21.05 percentage points on U and 16.45 on M. Instantaneous multiview evidence retains 64 current-value features on U and 128 on M after variable mean/max aggregation. It also outperforms the time-domain error input. These results support retaining distinct predictive responses for task-specific decision learning to address the predictive risk–decision gap.
Prediction-to-observation only retains decoded-observation evidence at native scales and full resolution with unchanged processing. Restoring the remaining sources improves VUS-PR by 5.78 percentage points on U and 5.68 on M. Token-space removal has smaller effects and retains local-statistic and observation-space predictions. These comparisons support complementary views of the same predictive process. On M, combining variable means and maxima outperforms either alone, supporting both broad and concentrated responses across variables.
Temporal organization and calibration. Retaining only current values from all three processing paths reduces VUS-PR by 6.31 percentage points on U and 4.64 on M. Temporal means and maxima preserve sustained and peak responses. Their benefit supports evaluating how predictive responses persist and concentrate over time when learning which deviations satisfy the target anomaly criterion. Paired sequence-resampling intervals in Table 20 remain above zero for this removal and for patch-variance and observation-space removal. Removing reference calibration lowers U by 5.28 points but raises M by 1.43 points.
| Removed component | TSB-AD-U | TSB-AD-U 95% interval | TSB-AD-M | TSB-AD-M 95% interval |
| Patch variance | +3.09 | [+1.49, +4.72] | +3.03 | [+1.49, +4.70] |
| Observation-space discrepancy | +3.02 | [+1.24, +4.95] | +2.51 | [+0.90, +4.36] |
| Temporal summaries | +6.31 | [+4.54, +8.11] | +4.64 | [+2.87, +6.56] |
| Reference calibration | +5.28 | [+3.32, +7.28] | −1.43 | [−4.33, +1.37] |
Table 20. Paired VUS-PR differences on TSB-AD-U/M (percentage points), with 95% sequence-resampling intervals. Positive values favor HieraNTP.
Patch-evidence removals retain the pretrained hierarchy. Removing patch 8 lowers U by 2.10 percentage points, while removing patch 32 raises it by 0.80 points. Patch-128 removal causes the largest patch-wise decline on M. A single token-grid offset or forward-direction evidence alone also lowers both track averages. Temporal views provide unequal benefits across tracks.
Dataset-specific adaptation. With the complete representation fixed, dataset-specific extensions improve VUS-PR over the shared base detector by 4.08 percentage points on U and 0.37 on M. Since the predictive responses are unchanged, the improvement comes from refining their combination for the target dataset. This supports retaining shared evidence while learning which combinations correspond to each dataset’s anomaly criteria.
Evidence Usage and Visualization
Figure 5 summarizes extension split gain for 17 univariate and five multivariate datasets. Patch mean dominates Exathlon, patch variance SMAP, and decoded-observation evidence Stock. Figure 4 shows pronounced losses from patch-variance removal in SMAP and observation-space removal in Stock. Together, these results support dataset-dependent use of the shared evidence representation.
Figure 4. Dataset-wise VUS-PR differences: HieraNTP minus component removal (pp). Numbers mark magnitudes ; parentheses give sequence counts. Rows are clustered within each track.
Figure 5. Evidence usage in dataset-specific extensions. Split-gain shares (%) are normalized by column group; numbers mark row-wise maxima per group. Full denotes full resolution; parentheses give evaluation sequence counts. Rows are clustered by source profiles within each track.
In Figure 6, patch-mean evidence tracks the onset and persistence of two labeled level shifts in Exathlon. In TAO, token-space responses remain elevated early, while decoded-observation evidence varies rapidly and temporal averaging smooths decoded-target responses. These patterns illustrate distinct information retained for task-adaptive detection, with further cases in Appendix F.2.
Figure 6. Top-five features in Exathlon (U) and TAO (M). Colors are scaled per feature; percentages show dataset-specific split gain. VUS-PR is computed per window.
Conclusion
HieraNTP addresses the predictive risk–decision gap by separating the modeling of temporal regularities from the interpretation of their violations. Its pretrained tokenizer and hierarchical next-token predictor provide multiscale evidence in token, local-statistic, and observation spaces. Preserving these discrepancies in a structured risk representation allows a task-adaptive detector to learn task-specific evidence combinations from anomaly labels while the pretrained modules remain frozen.
HieraNTP achieves the highest aggregate VUS-PR on both tracks of TSB-AD and TAB. Under matched detector training, preserving distinct evidence improves detection over time-domain errors and early aggregation. Ablations and evidence-use analyses reveal complementary contributions whose value varies across datasets and temporal configurations. These findings support preserving predictive evidence until downstream supervision determines its relevance to the anomaly criterion.
Notation
The notation follows Figure 2: hats mark predictions, GT marks realized targets, and denotes anomaly labels. Uppercase and denote sequences or hierarchical collections, and lowercase symbols denote their elements, with token vectors in bold. Hierarchy levels are indexed from fine to coarse, consistent with the ordering in the figure, while a superscript identifies the processing direction when the same construction is applied forward and backward. Symbols used only in a local derivation are defined where they first appear.
| Symbol | Definition |
| Time index, target-token position, variable, and hierarchy level. | |
| Base discrepancy channel, temporal alignment, processing direction, dataset, and appended-tree index. | |
| Input time series with time steps and variables; and denote the observed history and realized target segment. | |
| Target-task anomaly label available on training sequences and the resulting task-specific anomaly score. | |
| Each sequence’s designated training prefix, used to estimate discrepancy location and scale. | |
| Number of hierarchy levels and temporal support of level ; the evaluated configuration uses and . | |
| Continuous token sequence containing tokens at hierarchy level ; is the complete token pyramid. | |
| Tokenizer encoding map including continuous-token projection, level- next-token predictor, and frozen codebook lookup. | |
| Level- native-resolution decoder and full-resolution decoder. | |
| Predicted and realized target token pyramids at target position . | |
| Predicted and realized patch means at level . | |
| Predicted and realized patch variances. Their logarithms are and . | |
| Decoded prediction and realized target in observation space. | |
| Level- risk evidence in Figure 2 and its time- and direction-indexed form. | |
| Complete structured risk representation for . | |
| Set of 64 base discrepancy channels indexed by emission time and its channel- component for variable . | |
| Held, log-transformed value of the base discrepancy. | |
| Robust calibration with reference median and interquartile range . | |
| Track-specific base ensemble and selected dataset ensemble. | |
| Number of appended trees and the th appended tree for dataset . | |
| Direction-specific and final bidirectional anomaly scores. |
Table 4. Notation used in Figure 2, the method, and the appendix.
Method Details
Exact Predictive-Evidence Definitions
For a fixed direction , target position , and level , abbreviate the predicted and realized target tokens and as and . We suppress the direction superscript throughout this subsection. With and , HierNTP removes the radial component of their difference and evaluates the tangent residual under the level-specific precision matrix :
This score measures the tangent component relative to the realized token direction. With identity precision, it equals one minus the squared cosine similarity between the two tokens. Patch-mean and patch-variance discrepancies use and , where and . The two temporal alignments offset the token grid by 0 and 4 observations before predictive inference. Each variable has token and local-statistic channels across the six levels.
Before decoding, the frozen codebook lookup maps a continuous token to its corresponding quantized decoder state. The level-specific decoder then supplies two additional comparisons,
where is the realized observation target at level . The two offsets and six levels yield channels. The full-resolution decoder operates on the complete predicted and realized token pyramids and , with applied to each token, and provides
The full-resolution comparisons complete the channels per variable.
Direct Evidence Construction
Each channel uses robust center and scale statistics from finite forward-direction emissions in the designated reference prefix. The center is the median, and the initial scale is 1.4826 times the median absolute deviation. Degenerate scales fall back successively to the interquartile range divided by 1.349, the standard deviation, and one. Standardized emissions are projected over their target spans and averaged where spans overlap. The four full-resolution channels remain pointwise. The patch-mean discrepancy uses excess threshold 2, while the remaining channels use threshold 1. Positive excess is clipped to and transformed by .
Variable means and maxima over valid variables yield 128 multivariate direct channels, while univariate sequences retain 64. Every pooled channel then undergoes an average-tie rank transform over the complete sequence. The precision matrices in Eq. (5) are estimated separately for TSB-AD-U and TSB-AD-M from forward, offset-0 residuals whose target spans lie within the designated Tuning prefixes. Each sequence supplies at most 512 evenly spaced residuals per level. Ledoit–Wolf shrinkage [20] estimates the covariance around the sample mean, and its inverse weights the uncentered tangent residual during scoring. Prefix boundaries follow file metadata (tr), and fitting uses all eligible residuals irrespective of labels. Each track’s matrices are shared across alignments and directions and reused for the corresponding TAB setting.
Reference Calibration
Sparse discrepancy emissions are held until the next valid update and transformed as
where and the locally defined map returns the latest valid emission time for variable and channel at or before in direction . Given reference median and interquartile range , the robust calibration is
The pool-then-calibrate branch first aggregates variables by their mean and maximum and estimates from the final 40% of the reference prefix. The calibrate-then-pool branch estimates the same statistics for each variable before applying variable mean and maximum pooling.
Support rules. Variable-relative calibration requires both the current held value and the value 32 steps earlier to be valid. Each channel becomes active only if the first 60% of the reference prefix contains at least two fresh emissions and 64 support-eligible held values, and the remaining 40% contains at least one valid value. Its median and interquartile range are then estimated from that remaining 40%. Forward-direction reference statistics are applied unchanged to both directions.
Temporal Summaries
For each instantaneous channel , the temporal summaries are
Centered windows use nearest-value extension at sequence boundaries. The current value, six means, and six maxima give 13 values per evidence channel.
Risk-Evidence Construction
The comparisons in Appendix § 7.§ 1 yield 64 base channels per variable and processing direction, as enumerated in Table 5. Each channel retains its source, temporal support, and token-grid offset throughout subsequent processing. The reference prefix is the same sequence’s training segment, delimited by tr in TSB-AD and train-lens in TAB.
| Source | Comparison at each token-grid offset | Channels |
| Token-space | Predicted and realized tokens at six levels | |
| Patch mean | Predicted and realized patch means at six levels | |
| Patch variance | Predicted and realized patch variances at six levels | |
| Decoded observation | Decoded prediction and observed target at six native scales and full resolution | |
| Decoded target | Decoded prediction and realized-token reconstruction at six native scales and full resolution | |
| Total | Two token-grid offsets: 0 and 4 observations | 64 |
Table 5. Composition of the 64 base discrepancy channels.
Time alignment. For Direct evidence, standardized patch responses are assigned to every observation in the predicted patch and averaged where target intervals overlap. The four full-resolution channels retain their pointwise responses. Patch emissions occur at the target’s right boundary in forward processing and left boundary in reverse processing. The calibrated variants carry each valid emission until the next update, with held responses initialized to zero. Both constructions preserve separate channels for each support and offset on the original observation grid.
Processing order. Table 6 summarizes the three paths using the transformations defined in Appendices § 7.§ 2–§ 7.§ 4. Reference statistics are estimated from forward-direction training responses and reused in both directions. In univariate sequences, variable aggregation is the identity. Calibrate-then-pool retains its support filtering, whereas Pool-then-calibrate uses all held reference responses, so the paths can still differ.
| Variant | Ordered operations before temporal summaries | U | M |
| Direct | Robust standardization, target-interval projection, positive-excess log transform, variable aggregation, sequence ranking | 832 | 1664 |
| Pool-then-calibrate | Hold and log-transform responses, aggregate variables, calibrate the aggregate | 832 | 1664 |
| Calibrate-then-pool | Hold and log-transform responses, calibrate eligible variable responses, aggregate valid variables | 832 | 1664 |
| Complete representation | Concatenation of the three paths after temporal summaries | 2496 | 4992 |
Table 6. Processing paths and resulting representation dimensions.
Applying the 13 temporal summaries to each path yields univariate features. Multivariate sequences retain both variable means and maxima, doubling this count. Concatenation yields 2496 and 4992 features, respectively, for each direction’s input to the detector.
Task-Adaptive Detector Definition
Let denote the additive function of the track-specific base ensemble. For a dataset , the selected ensemble is
where and when no extension is used. TAB uses for adapted datasets, while TSB-AD uses the dataset-specific settings in Appendix § 8.§ 2. For each processing direction , this ensemble produces
The base trees remain fixed while the new trees are appended, so each dataset model is a single complete ensemble. After aligning the two processing directions, the final score is
Training Data and Model Configuration
Tokenizer and Next-Token Predictor Pretraining
Multiscale tokenizer pretraining. The causal tokenizer uses a patch size and stride of 8, six Swin stages with depths , attention-head counts , and 384-dimensional encoder states. It uses rotary position embeddings, and its local attention window contains the current position and the preceding 15 positions. In the paper notation, levels are ordered from fine to coarse and have temporal supports . Every encoder state is projected to a 48-dimensional continuous token, and all levels share an vector-quantization codebook. The decoder uses the coarsest quantized sequence as its query and the complete token pyramid as multiscale key-value memory. Full-resolution and native-scale reconstruction losses use a coverage-weighted average of absolute and squared error. The multiscale reconstruction weight is 0.3, and the vector-quantization commitment weight is 0.5. The tokenizer is trained for 200,000 steps on the QUALS-organized GIFT-Eval Pretrain corpus [22], followed by 20,000 additional steps on QUALS. The two stages use initial learning rates and , respectively, with cosine decay and warmup. Training uses AdamW with , weight decay 0.01, gradient clipping at 1, and an effective batch size of 336 across seven GPUs. The evaluated checkpoint is Swin_20000.pt, with 47,512,382 encoder–decoder parameters.
Hierarchical next-token pretraining. The hierarchical next-token predictor has hidden width 256, four attention heads, and feed-forward width 1024. The finest level contains four causal self-attention blocks. Each coarser level aggregates temporally overlapping child states and applies same-scale self-attention followed by cross-attention to the corresponding frozen tokenizer memory. Attention permits a key only when its raw endpoint does not exceed the target start. Dropout is 0.1. Rotary position embeddings use base 10,000 and position scale 8. The feedback bridge adds a residual multilayer perceptron to a linear map to convert predicted tokens into encoder states. Input projection and feedback parameters are shared across levels, whereas attention blocks and prediction heads are level-specific.
The staged predictor curriculum starts with entirely synthetic regular series and increases the QUALS training-data proportion in 20-percentage-point increments at 20,000-step intervals. The evaluated checkpoint, msi_bltstep_013000.pt, is saved after 13,000 steps with an 80% QUALS and 20% synthetic mixture, initialized from the preceding stage’s step-80,000 checkpoint. The synthetic series used in this stage contain no injected anomalous events or additive noise. Training samples contain 512–1024 observed points followed by at least 1024 target points, with a maximum total length of 4096. Normalization statistics are computed only from the observed history. Teacher forcing and detached rollout are used in equal proportions. The final stage uses AdamW with constant learning rates for the predictor and for the feedback bridge, weight decay 0.01, gradient clipping at 1, and mixed precision. The effective batch size is 320 across four GPUs, with no learning-rate warmup. The loss is
where token and feedback consistency use cosine loss and the two local-statistic terms use Smooth-L1 loss. Level-balanced weighting uses exponent 0.25 and caps individual level weights at 2.0. The predictor learns continuous token targets with codebook-classification loss disabled. All pretrained components remain frozen during anomaly detection.
Synthetic sequences. The final-stage generator combines periodic patterns and trends:
allows gain, envelope, phase, and within-cycle shape variations of a periodic pattern, and is a trend. Table 7 summarizes the forms and sampling strata. Outside the strong-trend stratum, $w_P \sim and a trend is included with probability 0.45; when included, , and otherwise . Periods are sampled approximately log-uniformly within each interval, with an off-grid preference of 0.8 relative to the 8-point token grid. The table reports proportions within the synthetic stream; eight of ten equal-sized batches use QUALS, and two are synthetic.
| Name | Generating form or sampling rule | Share |
| Sinusoidal | — | |
| Multiharmonic | — | |
| Square | — | |
| Triangular | — | |
| Sawtooth | — | |
| Repeated motif | ; smooth piecewise patterns or pulse trains | — |
| Smooth trend | — | |
| General | ; optional trend | 20% |
| Strong trend | ; , trend contrast | 30% |
| Long period | or , equally weighted; optional trend | 25% |
| Short period | ; optional trend | 25% |
Table 7. Synthetic patterns and sampling strata for next-token pretraining. Component formulas are representative forms before amplitude scaling and normal shape variations; sampling proportions apply to the four strata.
For a length- parent sequence, , , , and ; is a repeating within-cycle shape. The general and strong-trend strata allocate 15/20/25/20/20% to period intervals 7–31/32–127/128–511/512–1023/1024–1536. Trend contrast is the absolute difference between the last- and first-quarter means divided by the sequence standard deviation.
Predictive inference. Detection uses 4096-observation contexts and 1024-observation chunks with 672-observation overlap, giving a stride of 352. Input normalization fixes the mean and standard deviation from the designated sequence prefix. Extraction runs in both temporal directions with token-grid offsets of 0 and 4 observations. Repeated records are identified by view, level, and global target span. Forward and reverse emissions are timestamped at the last and first target observations in the original sequence, respectively. Temporal summaries extend endpoint values.
Task-Adaptive Detector Parameters
All four configurations use GPU XGBoost 2.1.4 with binary logistic loss, 256-bin histogram tree construction, seed 20260930, and zero initial-stage regularization.
For TSB-AD-M, dataset-specific adaptation selects among no extension and 32 or 96 trees of depth 2 or 4. Appended trees use learning rate 0.03, minimum child weight 16, regularization 30, zero regularization, row subsampling 0.85, and column subsampling 0.75. TSB-AD-U uses an expanded search with 24 parameter configurations and 32, 96, 192, or 384 appended trees. The sampled parameter ranges are depth , learning rate , minimum child weight , regularization , regularization , row subsampling , and column subsampling . Two configurations use the fixed depth-2 and depth-4 recipes above; the remaining 22 are sampled with seed 20260930. The resulting detectors append trees for 17 U and five M datasets. At evaluation, the dataset identifies the corresponding complete ensemble; other datasets use the shared base.
TSB-AD base-parameter selection and appended-tree adaptation use the held-sequence macro average of (VUS-PR + AP)/2. The M protocol requires at least two eligible held-out sequences and both classes in each dataset-training fold. The U search uses two dataset-stratified sequence folds when multiple Tuning sequences are available. Singleton datasets use a temporal holdout with a 4224-point exclusion on each side of the boundary when both partitions contain at least eight points of each class and 512 timestamps. Otherwise, they use the configuration selected by pooled validation on other Tuning datasets. The detector is fitted on sampled timestamps whose indices, labels, and fixed weights are recorded for every Tuning sequence in the protocol artifacts.
TAB detector training. The univariate and multivariate tracks each train a shared base detector from scratch, using the training prefixes of 1635 and 220 sequences across 15 and 29 datasets, respectively. Both tracks use the initial-stage parameters in Table 8, producing 384 base trees. Each dataset whose sampled training rows contain both normal and anomalous observations receives an independent 32-tree extension initialized from the same base ensemble. The original 384 trees remain unchanged. These appended trees have maximum depth 2, learning rate 0.03, minimum child weight 16, regularization 30, zero regularization, row subsampling 0.85, and column subsampling 0.75. The exported models contain effective dataset extensions for eight univariate and ten multivariate datasets, yielding 416 trees per adapted ensemble. Datasets with single-class training rows or no splits in the appended trees use the shared 384-tree base. This applies to seven univariate datasets and 19 multivariate datasets; five of the latter meet the two-class criterion but produce no appended-tree splits. Routing follows the dataset identity and this training-stage rule. Both stages use fixed boosting rounds and hyperparameters shared across datasets.
| Parameter | TSB-AD-U | TSB-AD-M | TAB univariate | TAB multivariate |
| Feature dimension | 2496 | 4992 | 2496 | 4992 |
| Number of trees | 384 | 384 | 384 | 384 |
| Maximum depth | 3 | 6 | 6 | 6 |
| Minimum child weight | 64 | 4 | 4 | 4 |
| regularization | 10 | 100 | 100 | 100 |
| Column subsampling | 0.35 | 0.75 | 0.75 | 0.75 |
| Learning rate | 0.025 | 0.08 | 0.08 | 0.08 |
| Row subsampling | 0.7 | 0.7 | 0.7 | 0.7 |
Table 8. Initial-stage XGBoost parameters. TAB uses the same parameters for both tracks and appends 32 trees for each adapted dataset.
Offline evidence construction uses the complete observation sequence, including the training prefix and test interval, with labels supplied separately. Training rows and their labels are then selected exclusively from the training prefix.
Normal points are sampled uniformly without replacement from each training prefix, with caps of 64 and 512 per sequence for the univariate and multivariate tracks. Anomalous points are sampled at evenly spaced positions within each labeled training event, with respective caps of 32 and 256 per event. Forward and reverse representations contribute equally weighted training rows. The resulting matrices contain 301,888 rows with 92,608 positive labels for the univariate track, and 302,286 rows with 77,006 positive labels for the multivariate track.
Sample weights assign equal total mass to each dataset. For datasets containing training anomalies, half of this mass is assigned to normal samples and half to anomalous events. The latter is divided equally across events and then across sampled points within each event. Datasets with normal-only training prefixes assign their full mass to normal samples. Each temporal direction receives half of its point’s weight. Before fitting, weights are rescaled to have mean one. For a dataset extension, rescaling applies to that dataset’s rows and preserves relative class, event, and direction weights. The global positive-class multiplier remains one. The runs use Python 3.9.24 and CUDA-enabled XGBoost 2.1.4 on NVIDIA RTX 4090 GPUs.
Implementation artifacts. The released downstream package begins from exported frozen Hier-aNTP records, prediction evidence, and decoder outputs. It contains feature construction, trained tree models, archived scores, and evaluation artifacts. Upstream tokenizer and predictor training code and checkpoints are maintained separately.
Benchmarks, Baselines, and Detailed Results
Datasets and Supervised Evaluation
HieraNTP is evaluated with target-task supervision: the downstream detector uses TSB-AD Tuning labels or TAB training-prefix labels, while its tokenizer and next-token predictor remain frozen. TSB-AD-U provides 48 Tuning and 350 Eval sequences, and TSB-AD-M provides 20 Tuning and 180 Eval sequences ([24]). The Eval collections cover 23 univariate and 17 multivariate source datasets. They include service monitoring (IOPS, NAB, and YAHOO), spacecraft telemetry (MSL and SMAP), industrial monitoring (SWaT), and physiological signals (LTDB, MITDB, and SVDB). Table 11 and Table 12 list every dataset and its Eval sequence count.
| Dataset | Sequences | AP | AUC-ROC | VUS-PR | VUS-ROC |
| CATSv2 | 1 | 0.503 | 0.725 | 0.375 | 0.741 |
| Daphnet | 1 | 0.176 | 0.871 | 0.188 | 0.876 |
| Exathlon | 30 | 0.950 | 0.996 | 0.950 | 0.996 |
| IOPS | 15 | 0.477 | 0.958 | 0.515 | 0.977 |
| LTDB | 8 | 0.329 | 0.619 | 0.408 | 0.711 |
| MGAB | 8 | 0.004 | 0.612 | 0.010 | 0.686 |
| MITDB | 7 | 0.289 | 0.742 | 0.316 | 0.789 |
| MSL | 7 | 0.666 | 0.957 | 0.777 | 0.975 |
| NAB | 23 | 0.735 | 0.932 | 0.752 | 0.937 |
| NEK | 8 | 0.833 | 0.960 | 0.853 | 0.975 |
| OPPORTUNITY | 27 | 0.811 | 0.962 | 0.808 | 0.963 |
| Power | 1 | 0.109 | 0.600 | 0.110 | 0.608 |
| SED | 2 | 0.645 | 0.976 | 0.849 | 0.992 |
| SMAP | 17 | 0.865 | 0.991 | 0.899 | 0.995 |
| SMD | 33 | 0.819 | 0.979 | 0.826 | 0.980 |
| Stock | 8 | 0.980 | 0.999 | 0.982 | 0.999 |
| SVDB | 18 | 0.539 | 0.901 | 0.518 | 0.927 |
| SWaT | 1 | 0.714 | 0.835 | 0.626 | 0.798 |
| TAO | 2 | 0.999 | 1.000 | 1.000 | 1.000 |
| TODS | 13 | 0.629 | 0.901 | 0.815 | 0.944 |
| UCR | 70 | 0.337 | 0.873 | 0.349 | 0.889 |
| WSD | 20 | 0.690 | 0.991 | 0.643 | 0.993 |
| YAHOO | 30 | 0.826 | 0.991 | 0.790 | 0.991 |
| Overall | 350 | 0.643 | 0.925 | 0.656 | 0.939 |
Table 11. Per-dataset results on the 350 TSB-AD-U evaluation sequences.
| Dataset | Sequences | AP | AUC-ROC | VUS-PR | VUS-ROC |
| CATSv2 | 5 | 0.256 | 0.796 | 0.221 | 0.799 |
| CreditCard | 1 | 0.041 | 0.758 | 0.062 | 0.753 |
| Daphnet | 1 | 0.387 | 0.938 | 0.400 | 0.941 |
| Exathlon | 25 | 0.999 | 1.000 | 0.998 | 1.000 |
| GECCO | 1 | 0.346 | 0.789 | 0.099 | 0.675 |
| Genesis | 1 | 0.005 | 0.724 | 0.015 | 0.837 |
| GHL | 23 | 0.112 | 0.692 | 0.098 | 0.700 |
| LTDB | 4 | 0.564 | 0.858 | 0.583 | 0.877 |
| MITDB | 11 | 0.341 | 0.818 | 0.267 | 0.839 |
| MSL | 14 | 0.301 | 0.739 | 0.340 | 0.764 |
| OPPORTUNITY | 7 | 0.183 | 0.671 | 0.172 | 0.666 |
| PSM | 1 | 0.148 | 0.639 | 0.174 | 0.612 |
| SMAP | 25 | 0.484 | 0.913 | 0.524 | 0.922 |
| SMD | 20 | 0.495 | 0.889 | 0.392 | 0.861 |
| SVDB | 28 | 0.616 | 0.936 | 0.623 | 0.955 |
| SWaT | 2 | 0.544 | 0.832 | 0.432 | 0.777 |
| TAO | 11 | 0.991 | 0.999 | 0.995 | 0.999 |
| Overall | 180 | 0.514 | 0.865 | 0.503 | 0.870 |
Table 12. Per-dataset results on the 180 TSB-AD-M evaluation sequences.
TAB uses the official full-shot training/test splits for 15 univariate and 29 multivariate datasets [36]. Complete results, including OPP, appear in Table 15 and Table 16. The benchmarks share some source datasets but use different splits and aggregation rules.
| Dataset | F1 | Aff-F1 | AP | AUC-ROC | VUS-PR | VUS-ROC |
| Daphnet | 0.051 | 0.712 | 0.063 | 0.642 | 0.112 | 0.715 |
| ECG | 0.616 | 0.903 | 0.728 | 0.969 | 0.842 | 0.984 |
| GAIA | 0.262 | 0.877 | 0.444 | 0.906 | 0.382 | 0.909 |
| Genesis | 0.523 | 0.974 | 0.630 | 0.983 | 0.740 | 0.994 |
| GHL | 0.000 | 0.714 | 0.003 | 0.188 | 0.005 | 0.305 |
| IOPS | 0.336 | 0.898 | 0.462 | 0.923 | 0.478 | 0.934 |
| KDD21 | 0.152 | 0.833 | 0.226 | 0.854 | 0.244 | 0.871 |
| MGAB | 0.009 | 0.673 | 0.004 | 0.585 | 0.006 | 0.590 |
| MSL | 0.359 | 0.868 | 0.446 | 0.854 | 0.518 | 0.881 |
| NAB | 0.357 | 0.915 | 0.546 | 0.874 | 0.611 | 0.885 |
| OPPORTUNITY | 0.527 | 0.952 | 0.706 | 0.965 | 0.624 | 0.950 |
| SMAP | 0.366 | 0.905 | 0.574 | 0.931 | 0.551 | 0.923 |
| SMD | 0.318 | 0.813 | 0.385 | 0.851 | 0.360 | 0.829 |
| SVDB | 0.471 | 0.862 | 0.509 | 0.905 | 0.554 | 0.933 |
| YAHOO | 0.683 | 0.942 | 0.852 | 0.977 | 0.749 | 0.967 |
| Comparison mean | 0.322 | 0.849 | 0.419 | 0.817 | 0.439 | 0.837 |
| All-dataset mean | 0.335 | 0.856 | 0.439 | 0.827 | 0.452 | 0.845 |
Table 15. HieraNTP results on all 15 TAB univariate datasets.
| Dataset | F1 | Aff-F1 | AP | AUC-ROC | VUS-PR | VUS-ROC |
| ASD | 0.384 | 0.832 | 0.377 | 0.837 | 0.389 | 0.862 |
| CalIt2 | 0.149 | 0.780 | 0.076 | 0.811 | 0.102 | 0.807 |
| CATSv2 | 0.134 | 0.689 | 0.117 | 0.663 | 0.147 | 0.709 |
| CICIDS | 0.040 | 0.854 | 0.125 | 0.958 | 0.153 | 0.931 |
| Creditcard | 0.078 | 0.754 | 0.128 | 0.955 | 0.098 | 0.914 |
| DAPHNET | 0.277 | 0.828 | 0.262 | 0.787 | 0.372 | 0.843 |
| DLR | 0.808 | 0.999 | 0.922 | 0.999 | 0.964 | 0.999 |
| ECG | 0.482 | 0.769 | 0.509 | 0.839 | 0.458 | 0.801 |
| Exathlon | 0.850 | 0.999 | 0.998 | 1.000 | 0.958 | 0.986 |
| GECCO | 0.269 | 0.917 | 0.530 | 0.976 | 0.455 | 0.983 |
| GHL | 0.013 | 0.596 | 0.018 | 0.396 | 0.027 | 0.471 |
| GutenTAG | 0.534 | 0.951 | 0.712 | 0.935 | 0.814 | 0.944 |
| KDD-99 | 0.198 | 0.758 | 0.121 | 0.973 | 0.102 | 0.964 |
| LTDB | 0.657 | 0.910 | 0.733 | 0.946 | 0.835 | 0.970 |
| M_genesis | 0.003 | 0.675 | 0.003 | 0.442 | 0.008 | 0.609 |
| MITDB | 0.213 | 0.865 | 0.305 | 0.857 | 0.304 | 0.890 |
| MSL | 0.088 | 0.657 | 0.079 | 0.381 | 0.129 | 0.496 |
| NYC | 0.075 | 0.541 | 0.035 | 0.701 | 0.049 | 0.711 |
| OPPORTUNITY | 0.132 | 0.777 | 0.119 | 0.929 | 0.141 | 0.934 |
| PSM | 0.344 | 0.783 | 0.424 | 0.650 | 0.418 | 0.636 |
| PUMP | 0.334 | 0.704 | 0.161 | 0.733 | 0.198 | 0.733 |
| SKAB | 0.428 | 0.827 | 0.480 | 0.563 | 0.595 | 0.633 |
| SMAP | 0.288 | 0.674 | 0.221 | 0.633 | 0.239 | 0.653 |
| SMD | 0.129 | 0.669 | 0.079 | 0.661 | 0.079 | 0.638 |
| SVDB | 0.541 | 0.939 | 0.675 | 0.972 | 0.687 | 0.978 |
| SWAN | 0.363 | 0.746 | 0.460 | 0.622 | 0.452 | 0.576 |
| SWAT | 0.403 | 0.758 | 0.731 | 0.858 | 0.624 | 0.833 |
| TAO | 0.977 | 0.993 | 0.999 | 1.000 | 0.951 | 0.985 |
| TODS | 0.215 | 0.742 | 0.258 | 0.738 | 0.249 | 0.682 |
| Comparison mean | 0.331 | 0.793 | 0.376 | 0.782 | 0.388 | 0.794 |
| All-dataset mean | 0.324 | 0.793 | 0.367 | 0.787 | 0.379 | 0.799 |
Table 16. HieraNTP results on all 29 TAB multivariate datasets.
Comparison Methods
The TSB-AD comparison includes leading public leaderboard submissions and established reference methods. These cover statistical and distance-based detectors, reconstruction and forecasting networks, and pretrained time-series models [24]. Representative neural detectors include Donut, OmniAnomaly, USAD, and TranAD [52, 3, 46] (Su et al. 2019). Pretrained baselines include MOMENT, Chronos, TimesFM, Lag-Llama, TimeRCD, and TSPulse [13, 2, 5, 39, 19, 9]. Recent submissions include xLSTMAD, MMPAD, StreamVAE, and CHARM [10, 55, 34, 7]. Their detection configurations follow the TSB-AD implementations and leaderboard releases.2 CHARM uses pretrained embeddings with cosine-neighbor scoring in TSB-AD. TimeRCD+MAFT is reported under its leaderboard submission name. Table 1 reports the TimeRCD selection of 11 univariate and five multivariate datasets using its four-metric implementation.
The TAB main table combines leading OpenTS methods with deep-learning baselines. KMeans, KNN, EIF, and OCSVM provide clustering, distance, isolation, and one-class references. DLinear, iTransformer, and TimesNet represent forecasting architectures used for detection [57, 26, 50]. TranAD, DCdetector, and DualTF provide anomaly-detection comparisons [46, 54, 32]. The extended tables also include UniTS and PatchTST [12, 33], language-model adaptations GPT4TS, UniTime, and LLMMixer [58, 25, 18], and pretrained Timer and TTM [27, 8]. The detection adaptations and scores follow TAB’s OpenTS implementations [36].3 The extended tables cover the OpenTS baselines, with missing metrics left blank.
Evaluation Metrics
AP summarizes precision–recall performance across score thresholds and corresponds to the AUC-PR field computed by both benchmark implementations. AUC-ROC summarizes the true-positive/false-positive trade-off. VUS-PR and VUS-ROC additionally integrate over temporal tolerance [35]. Standard-F1 evaluates pointwise predictions, range-based F1 measures interval detection, and Affiliation-F evaluates local event correspondence [16]. The TimeRCD comparison retains its F1-T implementation and metric names [19]. TSB-AD results average sequence-level metrics, while the TAB leaderboard comparison averages dataset-level metrics. The TAB threshold rule and medal aggregation are specified in Appendix § 9.6.
TSB-AD score evaluation. VUS uses 250 thresholds. The tolerance window is estimated from the first channel by autocorrelation, using at most the first 20,000 observations and 400 lags; values above 300 and failed estimates fall back to 125. Multivariate channels are not flattened. Scores are evaluated at their archived scale without complete-sequence min–max normalization.
Extended TSB-AD Leaderboard Results
Tables 9 and 10 consolidate public TSB-AD results. Recent submissions use three decimals, and the original leaderboard retains its published two-decimal precision. Entries absent from the public reports are left blank. For HieraNTP, the benchmark’s AUC-PR field uses average-precision_score, reported as AP. Uncomputed F1 metrics remain blank.
TSPulse, CHARM, and StreamVAE publish only VUS-PR for the corresponding submissions.
| Method | AUC-PR | AUC-ROC | VUS-PR | VUS-ROC | Std.-F1 | PA-F1 | Event-F1 | R-F1 | Aff.-F |
| HieraNTP | 0.643 | 0.925 | 0.656 | 0.939 | – | – | – | – | – |
| Time-RCD+MAFT (FT) | 0.590 | 0.901 | 0.586 | 0.920 | 0.555 | 0.701 | – | – | 0.915 |
| TSPulse (FT) | – | – | 0.552 | – | – | – | – | – | – |
| TSPulse (ZS) | – | – | 0.480 | – | – | – | – | – | – |
| CHARM | – | – | 0.501 | – | – | – | – | – | |
| Time-RCD | 0.495 | 0.843 | 0.522 | 0.871 | 0.527 | 0.672 | – | – | 0.886 |
| StreamVAE | – | – | 0.451 | – | – | – | – | – | – |
| MMPAD | 0.403 | 0.783 | 0.440 | 0.787 | 0.447 | 0.592 | 0.503 | 0.435 | 0.850 |
| xLSTMAD | 0.343 | 0.761 | 0.398 | 0.829 | 0.405 | 0.604 | 0.507 | 0.353 | 0.856 |
| Sub-PCA | 0.37 | 0.71 | 0.42 | 0.76 | 0.42 | 0.56 | 0.49 | 0.41 | 0.85 |
| KShapeAD | 0.35 | 0.74 | 0.40 | 0.76 | 0.39 | 0.58 | 0.46 | 0.40 | 0.83 |
| POLY | 0.31 | 0.73 | 0.39 | 0.76 | 0.37 | 0.53 | 0.45 | 0.35 | 0.85 |
| Series2Graph | 0.33 | 0.76 | 0.39 | 0.80 | 0.38 | 0.65 | 0.50 | 0.35 | 0.85 |
| MOMENT (FT) | 0.30 | 0.69 | 0.39 | 0.76 | 0.35 | 0.65 | 0.49 | 0.35 | 0.86 |
| MOMENT (ZS) | 0.30 | 0.68 | 0.38 | 0.75 | 0.35 | 0.61 | 0.49 | 0.36 | 0.86 |
| KMeansAD | 0.32 | 0.74 | 0.37 | 0.76 | 0.37 | 0.56 | 0.44 | 0.38 | 0.82 |
| USAD | 0.32 | 0.66 | 0.36 | 0.71 | 0.37 | 0.50 | 0.43 | 0.40 | 0.84 |
| Sub-KNN | 0.27 | 0.76 | 0.35 | 0.79 | 0.34 | 0.61 | 0.43 | 0.32 | 0.84 |
| MatrixProfile | 0.26 | 0.73 | 0.35 | 0.76 | 0.33 | 0.63 | 0.44 | 0.32 | 0.84 |
| SAND | 0.29 | 0.73 | 0.34 | 0.76 | 0.35 | 0.56 | 0.42 | 0.36 | 0.81 |
| CNN | 0.33 | 0.71 | 0.34 | 0.79 | 0.38 | 0.78 | 0.66 | 0.35 | 0.88 |
| LSTMAD | 0.31 | 0.68 | 0.33 | 0.76 | 0.37 | 0.71 | 0.59 | 0.34 | 0.86 |
| SR | 0.32 | 0.74 | 0.32 | 0.81 | 0.38 | 0.87 | 0.67 | 0.35 | 0.89 |
| TimesFM | 0.28 | 0.67 | 0.30 | 0.74 | 0.34 | 0.84 | 0.63 | 0.34 | 0.89 |
| IForest | 0.29 | 0.71 | 0.30 | 0.78 | 0.35 | 0.73 | 0.56 | 0.30 | 0.84 |
| OmniAnomaly | 0.27 | 0.65 | 0.29 | 0.72 | 0.31 | 0.59 | 0.46 | 0.29 | 0.83 |
| Lag-Llama | 0.25 | 0.65 | 0.27 | 0.72 | 0.30 | 0.77 | 0.59 | 0.31 | 0.88 |
| Chronos | 0.26 | 0.66 | 0.27 | 0.73 | 0.32 | 0.83 | 0.61 | 0.33 | 0.88 |
| TimesNet | 0.18 | 0.61 | 0.26 | 0.72 | 0.24 | 0.67 | 0.47 | 0.21 | 0.86 |
| AutoEncoder | 0.19 | 0.63 | 0.26 | 0.69 | 0.25 | 0.54 | 0.36 | 0.28 | 0.82 |
| TranAD | 0.20 | 0.57 | 0.26 | 0.68 | 0.25 | 0.58 | 0.43 | 0.25 | 0.83 |
| FITS | 0.17 | 0.61 | 0.26 | 0.73 | 0.23 | 0.65 | 0.42 | 0.20 | 0.86 |
| Sub-LOF | 0.16 | 0.68 | 0.25 | 0.73 | 0.24 | 0.57 | 0.35 | 0.25 | 0.82 |
| OFA | 0.16 | 0.59 | 0.24 | 0.71 | 0.22 | 0.67 | 0.45 | 0.20 | 0.86 |
| Sub-MCD | 0.15 | 0.67 | 0.24 | 0.72 | 0.23 | 0.54 | 0.32 | 0.24 | 0.81 |
| Sub-HBOS | 0.18 | 0.61 | 0.23 | 0.67 | 0.23 | 0.60 | 0.35 | 0.27 | 0.79 |
| Sub-OCSVM | 0.16 | 0.65 | 0.23 | 0.73 | 0.22 | 0.55 | 0.32 | 0.23 | 0.79 |
| Sub-IForest | 0.16 | 0.63 | 0.22 | 0.72 | 0.22 | 0.63 | 0.34 | 0.23 | 0.80 |
| Donut | 0.14 | 0.56 | 0.20 | 0.68 | 0.20 | 0.57 | 0.38 | 0.20 | 0.82 |
| LOF | 0.14 | 0.58 | 0.17 | 0.68 | 0.21 | 0.63 | 0.40 | 0.22 | 0.79 |
| AnomalyTransformer | 0.08 | 0.50 | 0.12 | 0.56 | 0.12 | 0.53 | 0.34 | 0.14 | 0.77 |
Table 9. Results on TSB-AD-U (350 evaluation sequences).
| Method | AUC-PR | AUC-ROC | VUS-PR | VUS-ROC | Std.-F1 | PA-F1 | Event-F1 | R-F1 | Aff.-F |
| HieraNTP | 0.514 | 0.865 | 0.503 | 0.870 | – | – | – | – | – |
| TSPulse (FT) | – | – | 0.387 | – | – | – | – | – | |
| TSPulse (ZS) | – | – | 0.361 | – | – | – | – | – | |
| xLSTMAD | 0.344 | 0.739 | 0.392 | 0.778 | 0.425 | 0.621 | 0.532 | 0.390 | 0.833 |
| MMPAD | 0.309 | 0.744 | 0.354 | 0.759 | 0.359 | 0.480 | 0.391 | 0.300 | 0.814 |
| CHARM | – | – | 0.391 | – | – | – | – | – | – |
| Time-RCD | 0.340 | 0.722 | 0.317 | 0.707 | 0.384 | 0.681 | – | – | 0.823 |
| StreamVAE | – | – | 0.435 | – | – | – | – | – | – |
| CNN | 0.32 | 0.73 | 0.31 | 0.76 | 0.37 | 0.78 | 0.65 | 0.37 | 0.87 |
| OmniAnomaly | 0.27 | 0.65 | 0.31 | 0.69 | 0.32 | 0.55 | 0.41 | 0.37 | 0.81 |
| PCA | 0.31 | 0.70 | 0.31 | 0.74 | 0.37 | 0.79 | 0.59 | 0.29 | 0.85 |
| LSTMAD | 0.31 | 0.70 | 0.31 | 0.74 | 0.36 | 0.79 | 0.64 | 0.38 | 0.87 |
| USAD | 0.26 | 0.64 | 0.30 | 0.68 | 0.31 | 0.53 | 0.40 | 0.37 | 0.80 |
| AutoEncoder | 0.30 | 0.67 | 0.30 | 0.69 | 0.34 | 0.60 | 0.44 | 0.28 | 0.80 |
| KMeansAD | 0.25 | 0.69 | 0.29 | 0.73 | 0.31 | 0.68 | 0.49 | 0.33 | 0.82 |
| CBLOF | 0.28 | 0.67 | 0.27 | 0.70 | 0.32 | 0.65 | 0.45 | 0.31 | 0.81 |
| MCD | 0.27 | 0.65 | 0.27 | 0.69 | 0.33 | 0.46 | 0.33 | 0.20 | 0.76 |
| OCSVM | 0.23 | 0.61 | 0.26 | 0.67 | 0.28 | 0.48 | 0.41 | 0.30 | 0.80 |
| Donut | 0.20 | 0.64 | 0.26 | 0.71 | 0.28 | 0.52 | 0.36 | 0.21 | 0.81 |
| RobustPCA | 0.24 | 0.58 | 0.24 | 0.61 | 0.29 | 0.60 | 0.42 | 0.33 | 0.81 |
| FITS | 0.15 | 0.58 | 0.21 | 0.66 | 0.22 | 0.72 | 0.32 | 0.16 | 0.81 |
| OFA | 0.15 | 0.55 | 0.21 | 0.63 | 0.21 | 0.72 | 0.41 | 0.17 | 0.83 |
| EIF | 0.19 | 0.67 | 0.21 | 0.71 | 0.26 | 0.74 | 0.44 | 0.26 | 0.81 |
| COPOD | 0.20 | 0.65 | 0.20 | 0.69 | 0.27 | 0.72 | 0.41 | 0.24 | 0.80 |
| IForest | 0.19 | 0.66 | 0.20 | 0.69 | 0.26 | 0.68 | 0.41 | 0.24 | 0.80 |
| HBOS | 0.16 | 0.63 | 0.19 | 0.67 | 0.24 | 0.67 | 0.40 | 0.24 | 0.80 |
| TimesNet | 0.13 | 0.56 | 0.19 | 0.64 | 0.20 | 0.68 | 0.32 | 0.17 | 0.82 |
| KNN | 0.14 | 0.51 | 0.18 | 0.59 | 0.19 | 0.69 | 0.45 | 0.21 | 0.79 |
| TranAD | 0.14 | 0.59 | 0.18 | 0.65 | 0.21 | 0.68 | 0.40 | 0.21 | 0.79 |
| LOF | 0.10 | 0.53 | 0.14 | 0.60 | 0.15 | 0.57 | 0.32 | 0.14 | 0.76 |
| AnomalyTransformer | 0.07 | 0.52 | 0.12 | 0.57 | 0.12 | 0.53 | 0.33 | 0.14 | 0.74 |
Table 10. Results on TSB-AD-M (180 evaluation sequences).
Per-Dataset TSB-AD Results
Tables 11 and 12 report HieraNTP results for all datasets. Each dataset row is the mean over its Eval sequences; the overall row averages all Eval sequences. AP denotes average precision.
TAB Full-Shot Evaluation and OpenTS Comparison
We use TAB’s full-shot evaluation strategy for univariate and multivariate anomaly detection (Qiu et al., [36]). HieraNTP fits its task-adaptive detector with labels from the official training prefixes and retains the frozen pretrained tokenizer and predictor. Training pools all 15 univariate or 29 multivariate datasets into a shared base detector per track, followed by dataset-specific extensions. The split follows the training and test lengths in TAB’s metadata. Detector fitting uses sampled training-prefix points, and evaluation uses the test suffixes. Sampling, weights, and boosting parameters are given in Appendix § 8.§ 2. The OpenTS leaderboard contains baseline results for 14 of TAB’s 15 univariate datasets and 28 of its 29 multivariate datasets. OPP, named OPPORTUNITY in the local data files, has no baseline entry and is excluded from the comparison in Table 2. API names CreditCard, M-Genesis, and KDD-99 map to Credit, Genesis, and KDDcup99.
TAB reports ten label-based and six score-based metrics. The OpenTS fields A–P, A–R, and V–PR denote AUC-PR, AUC-ROC, and VUS-PR, respectively. TAB computes AUC-PR with average precision. The metric columns in Table 2 are arithmetic means of the dataset-level public values on the common 14/28-dataset sets. Public API values have three decimal places, and HieraNTP results are rounded to the same precision before computing those means. TSB-AD averages its Eval sequences, while this TAB comparison averages datasets.
HieraNTP evaluates label-based metrics at anomaly ratios of 0.1%, 0.5%, 1%, 2%, 3%, 5%, 10%, 15%, 20%, and 25%. Each ratio defines a candidate threshold for each test series. The cutoff is the corresponding percentile of the combined training and test anomaly scores, without using labels to set the cutoff. Evaluation labels select the ratio maximizing Affiliation-F1, and the other label-based metrics are read at that threshold. TAB describes reporting the best result across candidate thresholds. Its threshold selection differs from the shared Affiliation-F1 rule used for HieraNTP. The F1-based columns describe each method under its stated evaluation rule. AP, AUC-ROC, VUS-PR, and VUS-ROC are computed from the original anomaly scores.
Medals are recomputed after adding HieraNTP to each public dataset–metric cell. The three highest distinct values receive gold, silver, and bronze (), with ties sharing medals. All 16 metrics contribute to the weighted composite score .
Extended TAB Model Comparisons
Table 13 and Table 14 compare HieraNTP with OpenTS methods reporting at least one of the six metrics. Means use the common 14/28-dataset comparison sets, following Appendix § 9.6. Bold entries mark the highest dataset-mean value in each column.
| Method | F1 | Aff-F1 | AP | AUC-ROC | VUS-PR | VUS-ROC |
| HieraNTP | 0.322 | 0.849 | 0.419 | 0.817 | 0.439 | 0.837 |
| KMeans | 0.293 | 0.792 | 0.356 | 0.790 | 0.362 | 0.784 |
| DWT-MLEAD | 0.242 | 0.716 | 0.291 | 0.739 | 0.312 | 0.699 |
| CALF | 0.182 | 0.820 | 0.231 | 0.687 | 0.251 | 0.758 |
| iTransformer | 0.194 | 0.810 | 0.243 | 0.687 | 0.244 | 0.753 |
| HBOS | 0.107 | 0.696 | 0.104 | 0.599 | 0.242 | 0.625 |
| TimesNet | 0.172 | 0.805 | 0.224 | 0.677 | 0.224 | 0.742 |
| UniTS | 0.161 | 0.798 | 0.208 | 0.651 | 0.221 | 0.718 |
| OCSVM | 0.107 | 0.700 | 0.148 | 0.619 | 0.218 | 0.658 |
| TimesFM | 0.196 | 0.808 | 0.223 | 0.622 | 0.217 | 0.677 |
| GPT4TS | 0.173 | 0.805 | 0.223 | 0.667 | 0.215 | 0.724 |
| KNN | 0.109 | 0.708 | 0.192 | 0.626 | 0.213 | 0.638 |
| ModernTCN | 0.163 | 0.807 | 0.210 | 0.663 | 0.210 | 0.725 |
| UniTime | 0.163 | 0.803 | 0.204 | 0.662 | 0.209 | 0.729 |
| PatchTST | 0.157 | 0.803 | 0.208 | 0.667 | 0.208 | 0.728 |
| DAGMM | 0.023 | 0.571 | 0.104 | 0.548 | 0.204 | 0.599 |
| DLinear | 0.165 | 0.808 | 0.213 | 0.659 | 0.202 | 0.721 |
| LLMMixer | 0.148 | 0.804 | 0.196 | 0.654 | 0.202 | 0.723 |
| EIF | 0.166 | 0.782 | 0.213 | 0.633 | 0.202 | 0.671 |
| LOF | 0.101 | 0.674 | 0.104 | 0.576 | 0.197 | 0.608 |
| Timer | 0.157 | 0.774 | 0.198 | 0.615 | 0.195 | 0.683 |
| Chronos | 0.160 | 0.796 | 0.191 | 0.629 | 0.195 | 0.682 |
| LSTMED | 0.079 | 0.704 | 0.213 | 0.619 | 0.193 | 0.669 |
| Moment | 0.160 | 0.796 | 0.189 | 0.626 | 0.191 | 0.695 |
| AE | 0.076 | 0.682 | 0.198 | 0.574 | 0.189 | 0.609 |
| ZMS | 0.073 | 0.709 | 0.169 | 0.592 | 0.184 | 0.650 |
| NLinear | 0.144 | 0.794 | 0.171 | 0.653 | 0.183 | 0.719 |
| TranAD | 0.158 | 0.774 | 0.201 | 0.545 | 0.179 | 0.594 |
| VAE | 0.071 | 0.681 | 0.200 | 0.574 | 0.177 | 0.622 |
| Torsk | 0.141 | 0.703 | 0.143 | 0.681 | 0.177 | 0.662 |
| Dada | 0.137 | 0.801 | 0.187 | 0.625 | 0.174 | 0.682 |
| CBLOF | 0.100 | 0.692 | 0.189 | 0.603 | 0.173 | 0.651 |
| PCA | 0.095 | 0.674 | 0.188 | 0.573 | 0.172 | 0.606 |
| LOF | 0.098 | 0.706 | 0.149 | 0.613 | 0.169 | 0.638 |
| IF | 0.089 | 0.718 | 0.132 | 0.604 | 0.168 | 0.662 |
| FTM | 0.120 | 0.755 | 0.170 | 0.600 | 0.167 | 0.663 |
| StatThreshold | 0.065 | 0.637 | 0.140 | 0.528 | 0.148 | 0.578 |
| Series2Graph | 0.114 | 0.781 | 0.090 | 0.633 | 0.129 | 0.675 |
| ARIMA | 0.070 | 0.652 | 0.117 | 0.508 | 0.129 | 0.564 |
| SARIMA | 0.071 | 0.682 | 0.114 | 0.525 | 0.127 | 0.586 |
| SpectralResidual | 0.065 | 0.698 | 0.114 | 0.533 | 0.123 | 0.610 |
| DualTF | 0.078 | 0.728 | 0.083 | 0.539 | 0.108 | 0.604 |
| LeftSTAMPi | 0.084 | 0.726 | 0.066 | 0.534 | 0.104 | 0.583 |
| ContraAD | 0.041 | 0.684 | 0.033 | 0.499 | 0.059 | 0.579 |
| ATrans | 0.025 | 0.700 | 0.038 | 0.473 | 0.058 | 0.485 |
| DDetector | 0.024 | 0.688 | 0.034 | 0.473 | 0.057 | 0.491 |
Table 13. Extended TAB univariate comparison on 14 datasets.
| Method | F1 | Aff-F1 | AP | AUC-ROC | VUS-PR | VUS-ROC |
| HieraNTP | 0.331 | 0.793 | 0.376 | 0.782 | 0.388 | 0.794 |
| OCSVM | 0.161 | 0.677 | 0.176 | 0.645 | 0.309 | 0.656 |
| KMeans | 0.296 | 0.698 | 0.340 | 0.744 | 0.302 | 0.685 |
| KNN | 0.179 | 0.699 | 0.336 | 0.714 | 0.301 | 0.707 |
| DAGMM | 0.083 | 0.561 | 0.171 | 0.648 | 0.294 | 0.660 |
| TimesNet | 0.199 | 0.789 | 0.259 | 0.688 | 0.271 | 0.722 |
| LSTMED | 0.109 | 0.587 | 0.281 | 0.686 | 0.269 | 0.696 |
| VAE | 0.103 | 0.554 | 0.278 | 0.689 | 0.263 | 0.694 |
| CATCH | 0.191 | 0.778 | 0.240 | 0.668 | 0.261 | 0.705 |
| TranAD | 0.238 | 0.759 | 0.280 | 0.685 | 0.259 | 0.680 |
| AE | 0.109 | 0.621 | 0.285 | 0.697 | 0.258 | 0.693 |
| CBLOF | 0.185 | 0.676 | 0.273 | 0.710 | 0.258 | 0.716 |
| EIF | 0.210 | 0.773 | 0.254 | 0.732 | 0.258 | 0.741 |
| DUET | 0.200 | 0.780 | 0.239 | 0.663 | 0.258 | 0.699 |
| PatchTST | 0.196 | 0.785 | 0.240 | 0.677 | 0.257 | 0.716 |
| iTransformer | 0.183 | 0.770 | 0.235 | 0.667 | 0.251 | 0.705 |
| HBOS | 0.170 | 0.660 | 0.250 | 0.664 | 0.245 | 0.677 |
| ModernTCN | 0.187 | 0.776 | 0.229 | 0.656 | 0.244 | 0.695 |
| DLinear | 0.176 | 0.775 | 0.227 | 0.660 | 0.240 | 0.695 |
| LOF | 0.164 | 0.693 | 0.245 | 0.635 | 0.236 | 0.631 |
| NLinear | 0.170 | 0.773 | 0.211 | 0.665 | 0.232 | 0.701 |
| PCA | 0.155 | 0.651 | 0.244 | 0.686 | 0.229 | 0.672 |
| IF | 0.078 | 0.579 | 0.199 | 0.636 | 0.212 | 0.675 |
| LODA | 0.157 | 0.656 | 0.187 | 0.578 | 0.181 | 0.604 |
| DualTF | 0.109 | 0.593 | 0.147 | 0.611 | 0.174 | 0.621 |
| Torsk | 0.139 | 0.602 | 0.137 | 0.575 | 0.171 | 0.555 |
| ContraAD | 0.103 | 0.621 | 0.118 | 0.498 | 0.137 | 0.516 |
| ATrans | 0.055 | 0.653 | 0.092 | 0.500 | 0.122 | 0.499 |
| DCdetector | 0.046 | 0.637 | 0.085 | 0.496 | 0.118 | 0.503 |
Table 14. Extended TAB multivariate comparison on 28 datasets.
Complete Per-Dataset TAB Results
Tables 15 and 16 report HieraNTP on all 15 univariate and 29 multivariate TAB datasets, including OPP (OPPORTUNITY). Dataset names follow OpenTS. The comparison mean excludes OPP, matching Table 2; the all-dataset mean includes it. Values retain three decimals.
Ablation Studies
Tables 17 and 18 report VUS-PR, average precision (AP), VUS-ROC, and AUC-ROC for the representation and adaptation comparisons. All configurations evaluate the same 350 U or 180 M sequences. All detector-based comparisons use the original training samples, labels, weights, random seed, and hyperparameters. When the detector input changes, the shared base detector and applicable dataset-specific extensions are refitted under the same evaluation protocol. Scores average both directions except in the forward-only configuration. Dataset routing is unchanged except for Shared base detector only. The fixed scorers defined below use no fitted trees. Component-removal variants modify the final 2496D/4992D representation. The patch-length rows remove the corresponding discrepancy channels and derived summaries while retaining all tokenizer and predictor layers. Full-resolution decoded discrepancies remain in those rows and still depend on the complete token hierarchy. The reference-calibration ablation retains the reference-derived precision matrices, support masks, and fixed clipping rules. The forward-only configuration retains centered temporal summaries. All results use fixed single-run configurations.
Aggregation before supervised detection. Mean + detector and Max + detector reduce each complete 2496D or 4992D feature vector to its arithmetic mean or maximum, respectively. Mean–max + detector concatenates both scalars. Aggregation operates across the final feature coordinates at each timestamp, using their existing values without additional cross-feature standardization.
Prediction-to-observation evidence. Prediction-to-observation only retains the decoded-observation source, comparing decoded predictions with observations at six native scales and full resolution. The two temporal alignments yield 14 base channels. Reference calibration, the three processing variants, and all 13 temporal summaries are unchanged, giving 546 features on U and 1092 on M with both variable means and maxima. Columns are selected directly from the complete feature matrices in their original relative order. This comparison evaluates complementary evidence sources with the tokenizer, predictor, and decoder fixed.
Stage-wise evidence construction. Time-domain error uses the scalar time-domain prediction error directly as the anomaly score. Time-domain error + detector supplies the same scalar to the supervised downstream detector. Instantaneous multiview evidence replaces this scalar with 64 in-
| Configuration | VUS-PR | AP | VUS-ROC | AUC-ROC |
| HieraNTP | 65.56 | 64.26 | 93.85 | 92.51 |
| Shared base detector only | 61.48 | 59.93 | 93.36 | 91.87 |
| Time-domain error | 33.85 | 34.42 | 80.36 | 76.18 |
| Fixed evidence averaging | 43.65 | 39.96 | 87.42 | 84.61 |
| Time-domain error + detector | 29.89 | 29.71 | 78.77 | 73.77 |
| Mean + detector | 38.20 | 32.71 | 86.72 | 82.56 |
| Max + detector | 32.11 | 25.91 | 76.33 | 74.47 |
| Mean–max + detector | 44.51 | 38.13 | 86.87 | 83.77 |
| Instantaneous multiview | 57.80 | 56.78 | 92.27 | 89.85 |
| Prediction-to-observation only | 59.78 | 58.23 | 90.96 | 89.12 |
| w/o reference calibration | 60.28 | 58.75 | 91.53 | 89.60 |
| w/o temporal summaries | 59.25 | 58.46 | 92.61 | 90.36 |
| Forward only | 60.32 | 58.93 | 92.73 | 91.17 |
| w/o token-space | 65.35 | 63.98 | 93.94 | 92.52 |
| w/o patch mean | 64.31 | 62.71 | 93.50 | 92.07 |
| w/o patch variance | 62.47 | 61.35 | 92.06 | 90.49 |
| w/o observation-space | 62.54 | 56.84 | 92.72 | 90.71 |
| w/o full-resolution decoded discrepancies | 63.23 | 57.94 | 92.91 | 91.00 |
| w/o native-scale decoded discrepancies | 65.19 | 64.05 | 93.91 | 92.48 |
| w/o decoded observation | 64.67 | 62.40 | 93.69 | 92.29 |
| w/o decoded target | 65.88 | 64.56 | 93.83 | 92.50 |
| w/o patch 256 | 65.53 | 64.59 | 93.58 | 92.12 |
| w/o patch 128 | 65.52 | 64.17 | 93.83 | 92.46 |
| w/o patch 64 | 65.83 | 64.53 | 93.75 | 92.33 |
| w/o patch 32 | 66.36 | 65.16 | 93.93 | 92.63 |
| w/o patch 16 | 66.12 | 64.61 | 93.89 | 92.53 |
| w/o patch 8 | 63.46 | 62.24 | 93.86 | 92.41 |
| Single token-grid offset | 64.22 | 62.38 | 93.50 | 92.12 |
Table 17. Complete TSB-AD-U ablations: sequence-mean metrics (%, higher is better).
| Configuration | VUS-PR | AP | VUS-ROC | AUC-ROC |
| HieraNTP | 50.28 | 51.39 | 86.99 | 86.48 |
| Shared base detector only | 49.91 | 51.04 | 86.79 | 86.33 |
| Time-domain error | 30.81 | 30.91 | 74.86 | 72.06 |
| Fixed evidence averaging | 35.27 | 34.09 | 77.53 | 75.74 |
| Time-domain error + detector | 28.34 | 28.44 | 74.09 | 70.57 |
| Mean + detector | 29.12 | 24.77 | 78.22 | 74.25 |
| Max + detector | 27.47 | 22.32 | 67.65 | 66.18 |
| Mean–max + detector | 33.83 | 29.66 | 76.29 | 73.33 |
| Instantaneous multiview | 44.08 | 44.38 | 84.01 | 82.34 |
| Prediction-to-observation only | 44.60 | 45.59 | 84.39 | 83.47 |
| w/o reference calibration | 51.71 | 52.73 | 88.09 | 87.18 |
| w/o temporal summaries | 45.63 | 46.16 | 85.58 | 84.31 |
| Forward only | 47.52 | 48.53 | 86.07 | 85.35 |
| Variable mean only | 47.55 | 48.13 | 85.56 | 84.85 |
| Variable maximum only | 47.41 | 47.61 | 85.31 | 84.64 |
| w/o token-space | 49.69 | 50.31 | 86.52 | 85.77 |
| w/o patch mean | 50.33 | 51.23 | 87.06 | 86.47 |
| w/o patch variance | 47.25 | 48.81 | 85.83 | 85.27 |
| w/o observation-space | 47.77 | 44.67 | 86.05 | 84.50 |
| w/o full-resolution decoded discrepancies | 48.03 | 45.35 | 86.16 | 84.55 |
| w/o native-scale decoded discrepancies | 49.57 | 50.53 | 87.31 | 86.85 |
| w/o decoded observation | 49.14 | 49.61 | 87.02 | 86.32 |
| w/o decoded target | 49.51 | 50.36 | 86.71 | 86.24 |
| w/o patch 256 | 49.98 | 51.06 | 86.42 | 85.86 |
| w/o patch 128 | 49.29 | 50.56 | 87.16 | 86.48 |
| w/o patch 64 | 49.92 | 50.86 | 86.83 | 86.22 |
| w/o patch 32 | 50.08 | 50.98 | 86.79 | 86.06 |
| w/o patch 16 | 49.70 | 50.78 | 86.93 | 86.24 |
| w/o patch 8 | 49.46 | 50.59 | 86.99 | 86.36 |
| Single token-grid offset | 49.62 | 50.46 | 87.27 | 86.66 |
Table 18. Complete TSB-AD-M ablations: sequence-mean metrics (%, higher is better).
stantaneous features on U, or 128 after variable mean/max aggregation on M. The next stage incorporates all three processing variants while retaining only their current values, corresponding to w/o temporal summaries. HieraNTP adds the temporal summaries to form the complete representation. Shared base detector only evaluates this representation without dataset-specific extensions. Fixed evidence averaging is a distinct label-free reference, defined below.
Fixed evidence averaging. The fixed scorer uses the pool-then-calibrate evidence before temporal summaries. It takes the positive part of each channel, averages over the 64 univariate or 128 multivariate channels, and averages the forward and reverse scores. It uses neither anomaly labels nor fitted trees and retains the same 64 base discrepancy definitions as the complete method. Its comparison with HieraNTP changes both the representation and the scoring rule.
Target-dataset label availability. We partition the existing full-model Eval results according to whether the target dataset contributes anomaly-labeled samples to the detector’s Tuning set. Every dataset represented in that training set contains sampled anomalous points. The remaining U datasets are CATSv2, Daphnet, Power, and SWaT; the remaining M datasets are CreditCard, Daphnet, GECCO, Genesis, PSM, and SWaT. These datasets use the shared supervised detector trained on other datasets. This grouping concerns the detector’s actual training samples, independently of whether an Eval sequence has a normal training prefix. Table 19 reports sequence-mean metrics for each subset. Because the groups contain different datasets, this comparison is descriptive.
| Track | Labels | Datasets | Sequences | VUS-PR | AP | VUS-ROC | AUC-ROC |
| U | Available | 19 | 346 | 65.94 | 64.57 | 94.07 | 92.70 |
| U | Unavailable | 4 | 4 | 32.45 | 37.55 | 75.58 | 75.77 |
| M | Available | 11 | 173 | 51.38 | 52.31 | 87.40 | 86.79 |
| M | Unavailable | 6 | 7 | 23.03 | 28.78 | 76.75 | 78.75 |
Table 19. Full-model results by target-dataset anomaly-label availability (%). Datasets and sequences refer to the Eval subsets.
For each selected ablation, we match its VUS-PR with HieraNTP on the same evaluation sequence and compute as full minus ablated performance. Within each track, 10,000 bootstrap replicates resample complete sequences with replacement, using identical indices for both configurations. The 2.5th and 97.5th percentiles of the macro-average differences form the intervals in Table 20. These quantify sequence-resampling uncertainty for fixed fitted models. Within-dataset dependence and variability across training runs remain outside this estimate. The reference-calibration interval is positive on U and spans zero on M.
Token-angle diagnostic. We examine the angular response of the token-space statistic in Equation (5). With random seed 20260924, we sample one Eval sequence per dataset and one variable per multivariate sequence, retaining at most 256 token pairs per scale, direction, and alignment. The sample contains 201,044 pairs from 23 U and 17 M sequences. Target spans are grouped as normal, anomalous, or mixed according to whether their point labels are all normal, all anomalous, or contain both. Labels are used only for this diagnostic. Table 21 summarizes these groups.
| Track | Target-span labels | Token pairs | (%) | (%) | Median |
| U | All | 111,516 | 5.09 | 33.58 | 0.656 |
| U | Normal | 95,723 | 4.75 | 33.11 | 0.659 |
| U | Anomalous | 4,837 | 5.50 | 30.39 | 0.710 |
| U | Mixed | 10,956 | 7.85 | 39.10 | 0.612 |
| M | All | 89,528 | 5.96 | 31.05 | 0.687 |
| M | Normal | 76,965 | 5.83 | 30.62 | 0.699 |
| M | Anomalous | 4,593 | 8.14 | 39.32 | 0.633 |
| M | Mixed | 7,970 | 5.93 | 30.41 | 0.639 |
Table 21. Cosine similarity between predicted and realized tokens in the diagnostic sample.
Negative cosine similarity occurs in 5.09% of sampled U pairs and 5.96% of M pairs, corresponding to prediction–target angles above . The tangent statistic measures predicted mismatch, weighted by the fitted scale-specific precision matrix. With identity precision, it equals and is non-monotonic over the full angular range. Using the fitted matrices, the negative-cosine bin has lower median tangent energy than the adjacent cosine bin in eight of 48 track–scale–direction–alignment groups. No sampled pair has cosine similarity below .
Visualization Details
Main-Text Visualizations
Figure 4 averages full-minus-ablated VUS-PR over each dataset’s Eval sequences. Tracks share a symmetric percentage-point color scale. Within-track clustering uses average linkage and Euclidean distance over six unstandardized differences.
Figure 5 aggregates split gain from dataset-specific appended trees. Source, temporal support, and processing variant partition the same total, each summing to 100% before rounding. Totals include all temporal summaries, alignments, and variable aggregations. Decoded sources include native-scale and full-resolution evidence, with Full identifying the latter. Rows use average-linkage clustering of the five source shares with Euclidean distance. Parentheses give Eval sequence counts.
Figure 6 shows manually selected contiguous 75% windows from Exathlon (U) and TAO (M). Each retains the five highest-gain exported features with nonconstant window responses. Curves are binned, orange bars mark anomaly labels, and dashed lines mark reference-prefix boundaries. TAO displays the temporally smoothest variable.
Labels identify source, patch length or full resolution, temporal summary, grid offset, processing variant, and variable aggregation. Mean/Max numbers specify window lengths in observations; Current value uses no temporal aggregation. Variable mean/max aggregates across variables. A shared grid offset appears once beside the sequence. Percentages report each feature’s share of dataset-extension split gain. Pool denotes variable aggregation and Cal. reference calibration. Patch mean and Patch variance compare local statistics. Decoded observation and Decoded target compare predictions with observations and realized-token reconstructions, respectively. Heatmaps use full-sequence 2nd–98th percentile ranges per feature; window VUS-PR uses unbinned scores and labels.
Additional Full-Sequence Cases
Figure 7–Figure 9 extend § 4.4 with eight univariate and four multivariate sequences. Stock, TAO, SMD entity 18, and SVDB were selected for high full-sequence VUS-PR. Each panel displays five nonconstant exported forward-direction features, ranked by split gain in the dataset-specific extension. Gain shares describe the fitted extension across its training samples. Reported VUS-PR uses the bidirectional detector’s original-resolution scores and labels over the complete sequence.
Figure 7. Evidence responses in (a) YAHOO, (b) SED, (c) NAB, and (d) NEK. YAHOO emphasizes short-patch variance, whereas SED combines patch lengths 8 and 64 with sharp maxima and sustained temporal means. NAB’s large-patch mean evidence forms broad responses around later changes. In NEK, short-patch mean evidence concentrates around early deviations, while token-space and variance responses extend over different intervals. Throughout the gallery, bins show mean responses and any anomaly; observations are min–max normalized. Multivariate panels show the smoothest variable with appreciable variation, selected without labels. Axes give time indices, and dashed lines mark reference-prefix boundaries.
Figure 9. Multivariate evidence in (a) SMD entity 1, (b) TAO entity 7, (c) SMD entity 18, and (d) SVDB entity 14. The same SMD features respond to recurring changes in (a) and the later level drop in (c). TAO combines slowly varying token-space responses with faster decoded-observation and temporally averaged decoded-target evidence. SVDB combines decoded, variance, and mean evidence with distinct temporal response patterns.
Figure 8. Evidence responses in (a) SMAP, (b) WSD, (c) Stock, and (d) IOPS. SMAP’s highest-gain variance feature follows the downward excursion alongside broader decoded-target responses. WSD combines decoded-observation and variance evidence across recurring cycles. Stock emphasizes current full-resolution responses. IOPS’s decoded summaries extend beyond the labeled event.
Limitations
HieraNTP relies on anomaly labels for evidence selection. With normal-only target training data, the shared detector learned from other datasets may inadequately capture the target anomaly definition. The current XGBoost detector also relies on empirically designed risk evidence, motivating future work on adaptive evidence representations learned with deep models.
Multivariate detection aggregates evidence across variables without explicitly modeling their dependencies, limiting sensitivity to anomalies in inter-variable relationships. The pipeline is currently offline, and removing reverse-direction evidence reduces performance in our ablations. Adapting it to real-time detection may therefore incur an accuracy cost.
References
- [2]Abdul Fatir Ansari, Lorenzo Stella, Caner Turkmen, Xiyuan Zhang, Pedro Mercado, Huibin Shen, Oleksandr Shchur, Syama Sundar Rangapuram, Sebastian Pineda Arango, Shubham Kapoor, Jasper Zschiegner, Danielle C. Maddix, Hao Wang, Michael W. Mahoney, Kari Torkkola, Andrew Gordon Wilson, Michael Bohlke-Schneider, and Yuyang Wang. Chronos: Learning the language of time series. Transactions on Machine Learning Research, 2024. URL https://openreview.net/forum?id=gerNCVqqtR.arxiv.org/abs/2403.07815
- [3]Julien Audibert, Pietro Michiardi, Frédéric Guyard, Sébastien Marti, and Maria A. Zuluaga. USAD: Unsupervised anomaly detection on multivariate time series. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2020. doi: 10.1145/3394486.3403392.DOI
- [5]Abhimanyu Das, Weihao Kong, Rajat Sen, and Yichen Zhou. A decoder-only foundation model for time-series forecasting. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, 2024. URL https://proceedings.mlr.press/v235/das24c.html.arxiv.org/abs/2310.10688
- [6]Ailin Deng and Bryan Hooi. Graph neural network-based anomaly detection in multivariate time series. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pp. 4027–4035, 2021. doi: 10.1609/aaai.v35i5.16523. URL https://ojs.aaai.org/index.php/AAAI/article/view/16523.DOI
- [7]Utsav Dutta, Gerardo Pastrana, Sina Khoshfetrat Pakazad, and Henrik Ohlsson. Giving sensors a voice: Multimodal JEPA for semantic time-series embeddings. In Proceedings of the 43rd International Conference on Machine Learning, volume 306 of Proceedings of Machine Learning Research, 2026. URL https://arxiv.org/abs/2605.31580.
- [8]Vijay Ekambaram, Arindam Jati, Pankaj Dayama, Sumanta Mukherjee, Nam H. Nguyen, Wesley M. Gifford, Chandra Reddy, and Jayant Kalagnanam. Tiny time mixers (TTMs): Fast pre-trained models for enhanced zero/few-shot forecasting of multivariate time series. In Advances in Neural Information Processing Systems, volume 37, pp. 74147–74181, 2024. doi: 10.52202/079017-2359.DOI
- [9]Vijay Ekambaram, Subodh Kumar, Arindam Jati, Sumanta Mukherjee, Tomoya Sakai, Pankaj Dayama, Wesley M. Gifford, and Jayant Kalagnanam. TSPulse: Tiny pre-trained models with disentangled representations for rapid time-series analysis. In International Conference on Learning Representations, 2026. URL https://arxiv.org/abs/2505.13033.
- [10]Kamil Faber, Marcin Pietroń, Dominik Žurek, and Roberto Corizzo. xLSTMAD: A powerful xLSTM-based method for anomaly detection. In IEEE International Conference on Data Mining, pp. 247–256, 2025. doi: 10.1109/ICDM65498.2025.00032.DOI
- [11]Yuchen Fang, Jiandong Xie, Yan Zhao, Lu Chen, Yunjun Gao, and Kai Zheng. Temporal-frequency masked autoencoders for time series anomaly detection. In 2024 IEEE 40th International Conference on Data Engineering, pp. 1228–1241, 2024. doi: 10.1109/ICDE60146.2024.00099.DOI
- [12]Shanghua Gao, Teddy Koker, Owen Queen, Thomas Hartvigsen, Theodoros Tsiligkaridis, and Marinka Zitnik. UniTS: A unified multi-task time series model. In Advances in Neural Information Processing Systems, volume 37, pp. 140589–140631, 2024. doi: 10.52202/079017-4463.DOI
- [13]Monomito Goswami, Konrad Szafer, Arjun Choudhry, Yifu Cai, Shuo Li, and Artur Dubrawski. MOMENT: A family of open time-series foundation models. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, 2024. URL https://proceedings.mlr.press/v235/goswami24a.html.arxiv.org/abs/2402.03885
- [14]Chan Sik Han and Keon Myung Lee. Leveraging intermediate representations of time series foundation models for anomaly detection. arXiv preprint arXiv:2509.12650, 2025. URL https://arxiv.org/abs/2509.12650.
- [15]Sahand Hariri, Matias Carrasco Kind, and Robert J. Brunner. Extended isolation forest. IEEE Transactions on Knowledge and Data Engineering, 33(4):1479–1489, 2021. doi: 10.1109/TKDE.2019.2947676.
- [16]Alexis Huet, Jose Manuel Navarro, and Dario Rossi. Local evaluation of time series anomaly detection algorithms. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022. doi: 10.1145/3534678.3539339. URL https://doi.org/10.1145/3534678.3539339.
- [17]Uzair Khan, Luigi Capogrosso, Francesco Biondani, Michele Magno, Franco Fummi, Francesco Setti, and Marco Cristani. ChronosAD: Leveraging time series foundation models for accurate anomaly detection. In IEEE International Conference on Industrial Informatics, 2026. URL https://arxiv.org/abs/2606.01300.
- [18]Md Kowsher, Md. Shohanur Islam Sobuj, Nusrat Jahan Prottasha, E. Alejandro Alanis, Ozlem Ozmen Garibay, and Niloofar Yousefi. LLM-Mixer: Multiscale mixing in LLMs for time series forecasting. arXiv preprint arXiv:2410.11674, 2024. URL https://arxiv.org/abs/2410.11674.
- [19]Tian Lan, Hao Duong Le, Jinbo Li, Wenjun He, Meng Wang, Chenghao Liu, and Chen Zhang. Towards foundation models for zero-shot time series anomaly detection: Leveraging synthetic data and relative context discrepancy. In Proceedings of the 43rd International Conference on Machine Learning, volume 306 of Proceedings of Machine Learning Research, 2026. URL https://arxiv.org/abs/2509.21190.
- [20]Olivier Ledoit and Michael Wolf. A well-conditioned estimator for large-dimensional covariance matrices. Journal of Multivariate Analysis, 88(2):365–411, 2004. doi: 10.1016/S0047-259X(03)00096-4.DOI
- [21]Yujie Li, Zhezhi Shao, Chengqing Yu, Yisong Fu, Tao Sun, Yongjun Xu, and Fei Wang. APT: Affine prototype-timestamp for time series forecasting under distribution shift. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pp. 15180–15188, 2026. doi: 10.1609/aaai.v40i18.38542.DOI
- [22]Yujie Li, Zhezhi Shao, Chengqing Yu, Yisong Fu, Weijie Zhu, Yifan Du, Jilin Hu, Bin Yang, Yongjun Xu, and Fei Wang. QUALS: Corpus equilibrium for universal forecasting via pattern quantization and learnability synchronization. Proceedings of the VLDB Endowment, 20(2):108–121, 2027. URL https://arxiv.org/abs/2609.20156.
- [23]Peiyuan Liu, Hang Guo, Tao Dai, Naiqi Li, Jigang Bao, Xudong Ren, Yong Jiang, and Shu-Tao Xia. CALF: Aligning LLMs for time series forecasting via cross-modal fine-tuning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp. 18915–18923, 2025. doi: 10.1609/aaai.v39i18.34082.DOI
- [24]Qinghua Liu and John Paparrizos. The elephant in the room: Towards a reliable time-series anomaly detection benchmark. In Advances in Neural Information Processing Systems: Datasets and Benchmarks Track, 2024. URL https://proceedings.neurips.cc/paper_files/paper/2024/hash/c3f3c690b7a99fba16d0efd35c838b2c_Abstract-Datasets_and_Benchmarks_Track.html.DOI
- [25]Xu Liu, Junfeng Hu, Yuan Li, Shizhe Diao, Yuxuan Liang, Bryan Hooi, and Roger Zimmermann. UniTime: A language-empowered unified model for cross-domain time series forecasting. In Proceedings of the ACM Web Conference 2024, 2024a. URL https://arxiv.org/abs/2310.09751.
- [26]Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. iTransformer: Inverted transformers are effective for time series forecasting. In International Conference on Learning Representations, 2024b. URL https://proceedings.iclr.cc/paper_files/paper/2024/hash/2ea18fdc667e0ef2ad82b2b4d65147ad-Abstract-Conference.html.arxiv.org/abs/2310.06625
- [27]Yong Liu, Haoran Zhang, Chenyu Li, Xiangdong Huang, Jianmin Wang, and Mingsheng Long. Timer: Generative pre-trained transformers are large time series models. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, pp. 32369–32399, 2024c. URL https://proceedings.mlr.press/v235/liu24cb.html.arxiv.org/abs/2402.02368
- [28]Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021.arxiv.org/abs/2103.14030
- [29]Donghao Luo and Xue Wang. ModernTCN: A modern pure convolution structure for general time series analysis. In International Conference on Learning Representations, 2024. URL https://proceedings.iclr.cc/paper_files/paper/2024/file/86b1437c1e4c3b3c4debf98234a67e7-Paper-Conference.pdf.
- [30]J. B. MacQueen. Some methods for classification and analysis of multivariate observations. In Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, volume 1, pp. 281–297, 1967. URL https://digicoll.lib.berkeley.edu/record/113015?v=pdf.
- [31]Natalia Martinez Gil, Fearghal O’Donncha, Wesley M. Gifford, Nianjun Zhou, Dhaval C. Patel, and Roman Vaculin. Adaptive conformal anomaly detection with time series foundation models for signal monitoring. In International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=7uFbs68MSI.
- [32]Youngeun Nam, Susik Yoon, Yooju Shin, Minyoung Bae, Hwanjun Song, Jae-Gil Lee, and Byung Suk Lee. Breaking the time-frequency granularity discrepancy in time-series anomaly detection. In Proceedings of the ACM Web Conference 2024, pp. 4204–4215, 2024. doi: 10.1145/3589334.3645556.DOI
- [33]Yuqi Nie, Nam H. Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. A time series is worth 64 words: Long-term forecasting with transformers. In International Conference on Learning Representations, 2023. URL https://arxiv.org/abs/2211.14730.
- [34]Kadir-Kaan Özer, René Ebeling, and Markus Enzweiler. STREAM-VAE: Dual-path routing for slow and fast dynamics in vehicle telemetry anomaly detection. In IEEE Intelligent Vehicles Symposium, pp. 444–451, 2026. doi: 10.1109/IV66570.2026.11623993.DOI
- [35]John Paparrizos, Paul Boniol, Themis Palpanas, Ruey S. Tsay, Aaron Elmore, and Michael J. Franklin. Volume under the surface: A new accuracy evaluation measure for time-series anomaly detection. Proceedings of the VLDB Endowment, 15(11):2774–2787, 2022. doi: 10.14778/3551793.3551830. URL https://doi.org/10.14778/3551793.3551830.
- [36]Xiangfei Qiu, Zhe Li, Wanghui Qiu, Shiyan Hu, Lekui Zhou, Xingjian Wu, Zhengyu Li, Chenjuan Guo, Aoying Zhou, Zhenli Sheng, Jilin Hu, Christian S. Jensen, and Bin Yang. TAB: Unified benchmarking of time series anomaly detection methods. Proceedings of the VLDB Endowment, 18(9):2775–2789, 2025a. doi: 10.14778/3746405.3746407. URL https://www.vldb.org/pvldb/vol18/p2775-qiu.pdf.DOI
- [37]Xiangfei Qiu, Xingjian Wu, Yan Lin, Chenjuan Guo, Jilin Hu, and Bin Yang. DUET: Dual clustering enhanced multivariate time series forecasting. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 1185–1196, 2025b. doi: 10.1145/3690624.3709325.DOI
- [38]Sridhar Ramaswamy, Rajeev Rastogi, and Kyuseok Shim. Efficient algorithms for mining outliers from large data sets. In Proceedings of the 2000 ACM SIGMOD International Conference on Management of Data, pp. 427–438, 2000. doi: 10.1145/342009.335437.
- [39]Kashif Rasul, Arjun Ashok, Andrew Robert Williams, Hena Ghonia, Rishika Bhagwatkar, Arian Khorasani, Mohammad Javad Darvishi Bayazi, George Adamopoulos, Roland Riachi, Nadhir Hassen, Marin Biloš, Sahil Garg, Anderson Schneider, Nicolas Chapados, Alexandre Drouin, Valentina Zantedeschi, Yuriy Nevmyvaka, and Irina Rish. Lag-Llama: Towards foundation models for probabilistic time series forecasting. arXiv preprint arXiv:2310.08278, 2024. URL https://arxiv.org/abs/2310.08278.
- [40]M. Saquib Sarfraz, Mei-Yen Chen, Lukas Layer, Kunyu Peng, and Marios Koulakis. Position: Quo vadis, unsupervised time series anomaly detection? In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, 2024. URL https://proceedings.mlr.press/v235/sarfraz24a.html.arxiv.org/abs/2405.02678
- [41]Bernhard Schölkopf, Robert C. Williamson, Alex Smola, John Shawe-Taylor, and John Platt. Support vector method for novelty detection. In Advances in Neural Information Processing Systems, volume 12, 1999. URL https://proceedings.neurips.cc/paper_files/paper/1999/file/8725fb777f25776ffa9076e44fcfd776-Paper.pdf.
- [42]Zezhi Shao, Yujie Li, Fei Wang, Chengqing Yu, Yisong Fu, Tangwen Qian, Bin Xu, Boyu Diao, Yongjun Xu, and Xueqi Cheng. BLAST: Balanced sampling time series corpus for universal forecasting models. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2, pp. 2502–2513, 2025. doi: 10.1145/3711896.3736860.DOI
- [43]Qichao Shentu, Beibu Li, Kai Zhao, Yang Shu, Zhongwen Rao, Lujia Pan, Bin Yang, and Chen-juan Guo. Towards a general time series anomaly detector with adaptive bottlenecks and dual adversarial decoders. In International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=AKcd7ImG5e.
- [45]Markus Thill, Wolfgang Konen, and Thomas Bäck. Time series anomaly detection with discrete wavelet transforms and maximum likelihood estimation. In International Work-Conference on Time Series, 2017. URL https://blogs.gm.fh-koeln.de/ciop/files/2019/01/thillwavelet.pdf.
- [46]Shreshth Tuli, Giuliano Casale, and Nicholas R. Jennings. TranAD: Deep transformer networks for anomaly detection in multivariate time series data. Proceedings of the VLDB Endowment, 15(6): 1201–1214, 2022. doi: 10.14778/3514061.3514067.DOI
- [47]Martin Uray, Saverio Messineo, Roland Kwitt, and Stefan Huber. Exploring zero-shot foundation models for multivariate time series anomaly detection. arXiv preprint arXiv:2607.12454, 2026. URL https://arxiv.org/abs/2607.12454.
- [48]Aaron van den Oord, Oriol Vinyals, and Koray Kavukcuoglu. Neural discrete representation learning. In Advances in Neural Information Processing Systems, 2017.arxiv.org/abs/1711.00937
- [49]Fei Wang, Yujie Li, Zezhi Shao, Chengqing Yu, Yisong Fu, Zhulin An, Yongjun Xu, and Xueqi Cheng. ARIES: Relation assessment and model recommendation for deep time series forecasting. arXiv preprint arXiv:2509.06060, 2025. URL https://arxiv.org/abs/2509.06060.
- [50]Haixu Wu, Tengge Hu, Yong Liu, Hang Zhou, Jianmin Wang, and Mingsheng Long. TimesNet: Temporal 2d-variation modeling for general time series analysis. In International Conference on Learning Representations, 2023. URL https://arxiv.org/abs/2210.02186.
- [51]Xingjian Wu, Xiangfei Qiu, Zhengyu Li, Yihang Wang, Jilin Hu, Chenjuan Guo, Hui Xiong, and Bin Yang. CATCH: Channel-aware multivariate time series anomaly detection via frequency patching. In International Conference on Learning Representations, 2025. URL https://proceedings.iclr.cc/paper_files/paper/2025/file/2b25c39788e5cf11d354de433ebf4c0-Paper-Conference.pdf.arxiv.org/abs/2410.12261
- [52]Haowen Xu, Wenxiao Chen, Nengwen Zhao, Zeyan Li, Jiahao Bu, Zhihan Li, Ying Liu, Youjian Zhao, Dan Pei, Yang Feng, Jie Chen, Zhaogang Wang, and Honglin Qiao. Unsupervised anomaly detection via variational auto-encoder for seasonal KPIs in web applications. In Proceedings of The Web Conference, 2018. doi: 10.1145/3178876.3185996.DOI
- [53]Jiehui Xu, Haixu Wu, Jianmin Wang, and Mingsheng Long. Anomaly transformer: Time series anomaly detection with association discrepancy. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=LzQQ89U1qm_.arxiv.org/abs/2110.02642
- [54]Yiyuan Yang, Chaoli Zhang, Tian Zhou, Qingsong Wen, and Liang Sun. DCdetector: Dual attention contrastive representation learning for time series anomaly detection. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023. doi: 10.1145/3580305.3599295.DOI
- [55]Chin-Chia Michael Yeh. Matrix profile for time-series anomaly detection: A reproducible open-source benchmark on TSB-AD. arXiv preprint arXiv:2604.02445, 2026. URL https://arxiv.org/abs/2604.02445.
- [56]Chin-Chia Michael Yeh, Yan Zhu, Liudmila Ulanova, Nurjihan Begum, Yifei Ding, Hoang Anh Dau, Diego Furtado Silva, Abdullah Mueen, and Eamonn J. Keogh. Matrix profile I: All pairs similarity joins for time series: A unifying view that includes motifs, discords and shapelets. In 2016 IEEE 16th International Conference on Data Mining, pp. 1317–1322, 2016. doi: 10.1109/ICDM.2016.0179.DOI
- [57]Ailing Zeng, Muxi Chen, Lei Zhang, and Qiang Xu. Are transformers effective for time series forecasting? In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pp. 11121–11128, 2023. doi: 10.1609/aaai.v37i9.26317.DOI
- [58]Tian Zhou, Peisong Niu, Xue Wang, Liang Sun, and Rong Jin. One fits all: Power general time series analysis by pretrained LM. In Advances in Neural Information Processing Systems, volume 36, pp. 43322–43355, 2023. doi: 10.52202/075280-1877.DOI
- [59]Xiaokun Zhu, Louis Carpentier, and Mathias Verbeke. When foundation models are one-liners: Limitations and future directions for time series anomaly detection. In International Conference on Learning Representations, 2026. URL https://proceedings.iclr.cc/paper_files/paper/2026/hash/cf70320e93c08b39b1b29a348097a376-Abstract-Conference.html.
- [60]Zihao Zhuang, Yingying Zhang, Kai Zhao, Chenjuan Guo, Bin Yang, Qingsong Wen, and Lunting Fan. Noise matters: Cross contrastive learning for flink anomaly detection. Proceedings of the VLDB Endowment, 18(4):1159–1168, 2024. doi: 10.14778/3717755.3717773.DOI








