CAST: Context- and Anomaly Structure-Conditioned Time Series Anomaly Generation

Haochen Zhang; Jie Peng; Songyuan Sui; Yu-Chao Huang; Xiangqi Zhu; Tianlong Chen

cs.LGAug 19, 2026

CAST: Context- and Anomaly Structure-Conditioned Time Series Anomaly Generation

Authors: Haochen Zhang, Jie Peng, Songyuan Sui, Yu-Chao Huang, Xiangqi Zhu, Tianlong Chen

Organizations: UNITES Lab, University of North Carolina at Chapel Hill · Rice University · Oregon State University

Abstract

Anomalous time series play a critical role in safety-critical domains, yet they are inherently scarce, heterogeneous, and costly to obtain. Existing time series generation methods predominantly focus on synthesizing normal data, providing limited value when anomalous samples are needed. We identify two fundamental challenges in anomaly generation: (i) the scarcity of anomaly data, and (ii) the heterogeneous morphological characteristics of anomalies. To address these challenges, we propose CAST, a Context- and Anomaly Structure-conditioned Time series anomaly generation framework with principled two-stage pretraining and finetuning strategy. In pretraining stage, we leverage abundant normal time series data to learn underlying system dynamics and substantially mitigate the limited availability of anomaly data. During finetuning, CAST explicitly conditions the generator on learned anomaly structure representations, enabling it to capture heterogeneous anomaly morphologies under similar contextual conditions. Extensive experiments on multiple real-world univariate and multivariate datasets demonstrate that CAST consistently outperforms state-of-the-art anomaly generation methods in terms of both generation fidelity and downstream task utility, highlighting the effectiveness of the proposed approach.

Figures & tables

Figure 1: Overview of CAST. The representation encoder first infers a anomaly structure latent from a known anomaly segment. The time series anomaly generator takes the anomaly structure latent and normal context as conditions to generate anomalous time series consistent with normal context and well aligned with the given anomaly structure.

Figure 2: Two-stage training pipeline of CAST.

		CAST	FlowTS	Diffusion -TS	TimeVAE	CGATS	GenIAS
MIT-DB	MSE ( $\times 10^{-3}$ )	2.9	8.1	5.4	6.6	4.5	50.0
	MAE ( $\times 10^{-2}$ )	3.6	6.3	4.9	5.9	4.0	17.0
	FID ( $\times 10^{-1}$ )	0.74	1.41	9.78	1.24	4.44	32.0
	KID ( $\times 10^{-3}$ )	0.67	2.32	21.14	1.83	2.74	60.0
	MMD ( $\times 10^{-2}$ )	1.15	3.73	9.18	1.84	8.33	13.07
SVDB	MSE ( $\times 10^{-3}$ )	4.3	11.0	9.1	12.0	9.8	35.0

Table 1: Comparison on five real-world datasets. Metrics are scaled for readability (see row labels). MSE and MAE are computed on anomalous segments only. Lower is better. Negative values in MMD (and KID) may occur due to finite-sample estimation error and should be interpreted as values close to zero.

	Random -Forest	Cat -Boost	Robust -TAD	TCN	Moment	OneFits All	LTSM	Average
MIT-DB
CAST	0.905	0.830	0.828	0.653	0.741	0.945	0.919	0.832
FlowTS	0.655	0.507	0.701	0.607	0.464	0.910	0.802	0.664
Diffusion-TS	0.823	0.594	0.732	0.683	0.499	0.910	0.867	0.730
TimeVAE	0.575	0.480	0.687	0.682	0.509	0.832	0.740	0.644
C-GATS	0.600	0.368	0.740	0.707	0.510	0.905	0.831	0.666

Table 2: Downstream anomaly detection performance (F1-score). Anomaly detectors are trained using synthetic anomalies and evaluated on real anomalous time series in the test sets. The best and the second best results are highlighted in bold and underline, respectively

	Components			Metrics
	Main	ASC	CUP	MSE $\downarrow$	MAE $\downarrow$	F1 $\uparrow$
(a)	$\checkmark$			0.023	0.080	0.595
(b)	$\checkmark$	$\checkmark$		0.040	0.080	0.599
(c)	$\checkmark$		$\checkmark$	0.021	0.072	0.686
(d)	$\checkmark$	$\checkmark$	$\checkmark$	0.011	0.053	0.711

Table 3: Ablation study of model components. ASC denotes anomaly structure conditioning, and CUP denotes context-unaware pretraining. The best metrics are highlighted in bold.

Panel of Figure 3 — Figure 3: Controllability of anomaly generation. Left ( 3(a) ): varying anomaly structure while fixing the normal context produces diverse anomaly patterns. The upper panel shows two anomaly segments. The lower panel shows generated time series conditioned on the same normal context but on these two different anomaly structures. Right ( 3(b) ): varying the normal context while fixing the anomaly structure preserves the anomaly pattern but adapts it to different contexts.

Figure 4: Generated anomalies conditioned on the same context with interpolated latent codes

u_{\alpha}=\alpha u_{1}+(1-\alpha)u_{0}

, where

\alpha\in\{0.0,0.25,0.5,0.75,1.0,1.25,1.5\}

denotes the interpolation coefficient;

u_{0}

and

u_{1}

are two inferred latent variables from seen anomalies. Anomaly morphology varies smoothly with latent interpolation

\alpha

, while the normal context remains unchanged.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Dimension	Sequence Length	Minimum Anomaly Length	Maximum Anomaly Length
MIT-DB	2	1000	180	800
QT-DB	2	600	80	450
SV-DB	2	800	30	360
PV	1	200	20	144
Metro	1	72	24	24

Appendix

Table 4: Statistics of the time series datasets used in our experiments.

Figure 5: Comparison between CAST and simple baselines. Mean and standard deviation of F1 scores are reported over 5 runs.

Method	Random Forest	OneFitsAll	Avg
CAST	0.479	0.917	0.698
FlowTS	0.393	0.776	0.5845
Diffusion-TS	0.395	0.778	0.5865
TimeVAE	0.382	0.407	0.3945
C-GATS	0.369	0.221	0.295
GenIAS	0.384	0.162	0.273

Appendix

Table 5: Comparison of FID on Incart dataset. Higher is better.

Figure 6: Latent structure diagnostic. Time series neighbors in the trained VQVAE’s latent space are shown in blue. Randomly selected time series are shown in red.

Method	Random Forest	CatBoost	RobustTAD	TCN	Moment	OneFitsAll	Avg
MIT-DB
Modified	0.609	0.410	0.782	0.604	0.590	0.837	0.6387
Original	0.480	0.425	0.733	0.424	0.531	0.754	0.5578
SVDB
Modified	0.632	0.558	0.581	0.733	0.495	0.903	0.6503
Original	0.620	0.434	0.063	0.657	0.268	0.832	0.4790

Appendix

Table 6: Comparison between modified GenIAS and original GenIAS across multiple anomaly detectors (F1 score). The average performance is reported in the last column.

Explore similar work

ASTER: Latent Pseudo-Anomaly Generation for Unsupervised Time-Series Anomaly Detection
Apr 15, 2026Romain Hermary, Samet Hicsonmez, Dan Pineau +2Time-Series Anomaly Detection Unsupervised Anomaly Detection
University of Luxembourg, Luxembourg
Fun-TSG: A Function-Driven Multivariate Time Series Generator with Variable-Level Anomaly Labeling
Apr 14, 2026Pierre Lotte, André Péninou, Olivier TesteTime Series Generation Anomaly Localization
EPE UT, IRIT · UT2J, IRIT-SIG, IRIT · IRIT-SIG, IRIT, UT2J, Comue de Toulouse
BoundAD: Boundary-Aware Negative Generation for Time Series Anomaly Detection
Mar 18, 2026Xiancheng Wang, Lin Wang, Zhibo Zhang +2Time-Series Anomaly Detection Unsupervised Anomaly Detection
School of Ocean Engineering, Harbin Institute of Technology, West Wenhua Road, Weihai, 264209, Shandong, China. · Technical Center, Bogie Development Department, CRRC Qingdao Sifang Locomotive and Rolling Stock Co., Ltd., Jinhong East Road, Qingdao, 266111, Shandong, China.

Apr 15, 2026J/K move · Enter open · S save

ASTER: Latent Pseudo-Anomaly Generation for Unsupervised Time-Series Anomaly Detection

Romain Hermary, Samet Hicsonmez, Dan Pineau, Abd El Rahman Shabayek+1

University of Luxembourg, Luxembourg

Time-Series Anomaly Detection Unsupervised Anomaly Detection Synthetic Anomaly Generation

Open paper AskPDF arXiv

Time-series anomaly detection (TSAD) is critical in domains such as industrial monitoring, healthcare, and cybersecurity, but it remains challenging due to rare and heterogeneous anomalies and the scarcity of labelled data. This scarcity makes unsupervised approaches predominant, yet existing methods often rely on reconstruction or forecasting, which struggle with complex data, or on embedding-based approaches that require domain-specific anomaly synthesis and fixed distance metrics. We propose ASTER, a framework that generates pseudo-anomalies directly in the latent space, avoiding handcrafted anomaly injections and the need for domain expertise. A latent-space decoder produces tailored pseudo-anomalies to train a Transformer-based anomaly classifier, while a pre-trained LLM enriches the temporal and contextual representations of this space. Experiments on three benchmark datasets show that ASTER achieves state-of-the-art performance and sets a new standard for LLM-based TSAD.

Abstract

Explore similar work

ASTER: Latent Pseudo-Anomaly Generation for Unsupervised Time-Series Anomaly Detection

Fun-TSG: A Function-Driven Multivariate Time Series Generator with Variable-Level Anomaly Labeling

BoundAD: Boundary-Aware Negative Generation for Time Series Anomaly Detection