From Retrieval to Recognition:Knowledge-Enhanced Post-training of Time Series Foundation Models for Classification

Zhen Liu; Yucheng Wang; Yingpeng Du; Tianjun Wei; Qianli Ma; Min Wu; Jie Zhang; Zhu Sun

cs.LGOct 19, 2026

From Retrieval to Recognition:Knowledge-Enhanced Post-training of Time Series Foundation Models for Classification

Authors: Zhen Liu, Yucheng Wang, Yingpeng Du, Tianjun Wei, Qianli Ma, Min Wu, Jie Zhang, Zhu Sun

Organizations: Nanyang Technological University, Singapore · Institute of Advanced Intelligence and Computing (IAIC), A*STAR, Singapore · AI Center of Excellence for Manufacturing (AIMfg), A*STAR, Singapore · South China University of Technology, China ·

Abstract

Time series foundation models (TSFMs) have achieved notable progress in classification tasks through large-scale pretraining. However, their post-training practice mainly focuses on implicit parameter fine-tuning, where task-relevant information is entangled with abundant irrelevant knowledge in vast parameter spaces. This may hinder TSFMs from precisely activating useful knowledge for downstream classification. To this end, we propose Retrieval-to-Recognition (R2R), a plug-and-play post-training paradigm that adapts pretrained TSFMs to time series classification. Rather than pretraining a new foundation model, R2R improves how existing TSFMs are post-trained: it introduces a decoupled query encoder to extract discriminative patterns and retrieve instance-level numerical knowledge from a constructed knowledge base. The retrieved explicit knowledge is then fused with the TSFM's implicit parametric embeddings, enabling joint knowledge transfer at both the parameter and feature levels. In this way, R2R allows TSFMs to recall task-relevant patterns learned during pretraining, improving the efficiency of knowledge utilization in downstream classification transfer. Experiments on 158 time series datasets show that R2R consistently improves the classification performance of state-of-the-art TSFMs. Ablation and knowledge-enhanced studies further validate the effectiveness of the proposed framework.

Figures & tables

Figure 1: Illustration of knowledge paradigms in TSFMs for classification. (a) Knowledge learning : TSFMs learn discriminative class patterns from multi-domain pretraining data, storing knowledge implicitly in model parameters. (b) Knowledge transfer : Existing methods adapt this knowledge to target tasks via parameter fine-tuning, without explicitly reusing pretraining data knowledge. (c) Knowledge enhancement : R2R enhances parameter transfer with feature-level retrieval, explicitly leveraging numerical patterns from a pretraining knowledge base to boost recognition. Stage (a) is pretraining; (b) and (c) are post-training, and this work targets the (b) $\rightarrow$ (c) transition.

Figure 2: The overall architecture of the R2R framework. The knowledge extraction and retrieval modules constitute the query encoder. R2R uses the output of the query encoder as feature-level knowledge and fuses it with parameter-level class-token embeddings from pre-trained TSFMs for knowledge enhancement.

Metric	RF	GPT4TS	MOMENT	Chronos	Moirai	UniTS	NuTime	Mantis	UniShape	TS-RAG	TSPulse	R2R
Avg. Rank	5.80	9.00	6.53	6.33	9.00	11.87	5.63	5.80	5.20	3.97	5.37	1.73
Win	2	1	1	2	1	0	0	1	1	5	1	21
P-value	8.52E-06	1.28E-05	1.43E-06	1.30E-05	9.31E-08	2.05E-08	2.11E-05	2.64E-05	8.20E-07	1.33E-04	9.31E-08	-

Table 1: Test accuracy results on 30 additional datasets under the zero-shot feature extraction setting.

Table 5

Figure 5: t-SNE visualization on the SonyAIBORobotSurface1 dataset.

Metric	Raw	Jitter	Time warp	Magnitude warp	Window slice	Time flip	Permutation	Scaling	R2R
Avg. Rank	2.39	5.11	5.06	4.89	4.33	5.06	5.50	4.83	2.00

Table 4: Test results on 18 UCR datasets compared with different data augmentation methods.

Dataset A	Dataset B	Only parameter transfer			R2R transfer
Dataset A	Dataset B	Acc ${}_{A_{1}}$	Acc ${}_{A_{2}}$	Forgetting	Acc ${}_{A_{1}}$	Acc ${}_{A_{2}}$	Forgetting
AsphaltRegularityUni_eq	FloodModeling1_disc	0.9654	0.4887	0.4767	0.9734	0.7909	0.1825
GesturePebbleZ1_eq	Colposcopy	0.9477	0.5233	0.4244	0.9535	0.7151	0.2384
AllGestureWiimoteX_eq	AllGestureWiimoteY_eq	0.7671	0.2429	0.5242	0.7786	0.2771	0.5015
GestureMidAirD1_eq	GestureMidAirD2_eq	0.7538	0.1231	0.6307	0.7692	0.1538	0.6154

Table 5: Impact of continual fine-tuning for TSC. Acc

{}_{A_{1}}

: Initial accuracy on dataset

A

; Acc

{}_{A_{2}}

: Accuracy on dataset

A

after continual training on dataset

B

; Forgetting: Acc

{}_{A_{1}}

- Acc

{}_{A_{2}}

Appendix figures & tables16 assets

Supplementary material from the paper’s appendix.

Appendix

ID	Dataset name	Train size	Test size	No. classes	Length	ID	Dataset name	Train size	Test size	No. classes	Length
1	ACSF1	100	100	10	1460	65	ItalyPowerDemand	67	1029	2	24
2	Adiac	390	391	37	176	66	LargeKitchenAppliances	375	375	3	720
3	AllGestureWiimoteX	300	700	10	Vary	67	Lightning2	60	61	2	637
4	AllGestureWiimoteY	300	700	10	Vary	68	Lightning7	70	73	7	319
5	AllGestureWiimoteZ	300	700	10	Vary	69	Mallat	55	2345	8	1024
6	ArrowHead	36	175	3	251	70	Meat	60	60	3	448

Appendix

Table 6: Summary of the 128 univariate datasets from the UCR archive used in the fully supervised classification setting. “Train size” and “Test size” indicate the number of instances in the corresponding splits. “No. class” represents the number of classes, and “Length” denotes the time series length. The label “Vary” in the length column signifies datasets with variable-length sequences due to missing values.

ID	Dataset name	Train size	Test size	Length	No. classes	Category
1	AconityMINIPrinterLarge_eq	2403	1184	300	2	Sensor
2	AconityMINIPrinterSmall_eq	589	292	300	2	Sensor
3	AllGestureWiimoteX_eq	300	700	500	10	Motion
4	AllGestureWiimoteY_eq	300	700	500	10	Motion
5	AllGestureWiimoteZ_eq	300	700	500	10	Motion
6	AsphaltObstaclesUni_eq	390	391	736	4	Sensor

Appendix

Table 7: Summary of the 30 additional univariate time series datasets used for zero-shot feature extraction evaluation. Dataset name suffixes indicate preprocessing protocols: _eq denotes unequal-length sequences normalized via mean-padding with Gaussian noise; _nmv refers to datasets originally containing missing values, where only complete instances are retained; and _disc indicates datasets derived from the TSER archive. “Train size” and “Test size” correspond to the number of samples in each split. “No. class” specifies the number of classes, “Length” denotes the sequence length, and “Category” indicates the application domain.

ID	Dataset	Rocket	MiniRocket	RDST	MR-H	HC2	IncTime	TS2Vec	PatchTST	TimesNet	SoftShape	MambaSL	GPT4TS	MOMENT	UniTS	NuTime	TS-RAG	Mantis	UniShape	R2R +NuTime	R2R +Mantis	R2R +UniShape
1	ACSF1	0.8600	0.9100	0.9300	0.8600	0.9100	0.9400	0.8800	0.7300	0.3000	0.8800	0.8600	0.6100	0.5000	0.7700	0.6700	0.7500	0.7000	0.8000	0.8200	0.8600	0.9300
2	Adiac	0.7903	0.8235	0.7136	0.8344	0.8184	0.8107	0.7545	0.6675	0.6292	0.8159	0.7877	0.7826	0.1458	0.5115	0.8031	0.8235	0.8159	0.8289	0.8184	0.8440	0.8414
3	AllX.	0.7629	0.7429	0.7471	0.7600	0.7671	0.7229	0.6729	0.2643	0.4529	0.7871	0.6114	0.5029	0.5157	0.6329	0.7086	0.7600	0.7914	0.7743	0.6929	0.7829	0.7743
4	AllY.	0.7486	0.7571	0.7386	0.7886	0.7586	0.7943	0.6771	0.2529	0.4557	0.7643	0.7043	0.5129	0.5986	0.6443	0.7557	0.8000	0.8243	0.7929	0.7429	0.8253	0.8086
5	AllZ.	0.7886	0.7314	0.7100	0.7529	0.7886	0.7771	0.7371	0.1757	0.4071	0.7443	0.5900	0.4071	0.4886	0.5700	0.6614	0.7300	0.7586	0.7429	0.6743	0.7300	0.7657
6	Arrd.	0.8057	0.8400	0.8686	0.8629	0.9086	0.8400	0.8400	0.6971	0.7486	0.8457	0.7429	0.8057	0.5657	0.7543	0.8343	0.8971	0.8686	0.8857	0.9257	0.9200	0.9257

Appendix

Table 8: Detailed test classification accuracy results on 128 UCR datasets under the fully supervised classification setting. Due to space constraints, dataset names are abbreviated; full names can be referenced via their IDs in Table 6 . “IncTime” denotes InceptionTime. The best results are in bold .

ID	Dataset name	RF	GPT4TS	Moment	Chronos	Moirai	UniTS	NuTime	Mantis	UniShape	TS-RAG	TSPulse	R2R
1	AconityMINIPrinterLarge_eq	0.9096	0.6596	0.8660	0.9265	0.8049	0.5819	0.8438	0.8710	0.8691	0.9206	0.9215	0.9417
2	AconityMINIPrinterSmall_eq	0.9486	0.7158	0.9589	0.9315	0.8562	0.6575	0.8562	0.9315	0.9486	0.9555	0.9692	0.9863
3	AllGestureWiimoteX_eq	0.5286	0.4186	0.5343	0.4100	0.2929	0.0800	0.5414	0.4943	0.5343	0.5786	0.5341	0.6571
4	AllGestureWiimoteY_eq	0.5571	0.4157	0.5857	0.3829	0.3014	0.0971	0.5129	0.5086	0.5771	0.6200	0.5729	0.7000
5	AllGestureWiimoteZ_eq	0.4643	0.3871	0.5286	0.4343	0.2614	0.1029	0.4543	0.4757	0.5557	0.5700	0.4886	0.6943
6	AsphaltObstaclesUni_eq	0.7366	0.7417	0.7647	0.8159	0.4731	0.2379	0.7954	0.7980	0.8235	0.8568	0.8210	0.8261

Appendix

Table 9: Detailed test classification accuracy results on 30 additional time series datasets under the zero-shot feature extraction setting. The best results are in bold .

ID	Dataset name	R2R+ UniShape	w/o Feature transfer	w/o Parameter transfer
				MiniRocket	MiniRocket	MiniRocket	Rocket	MultiRocket	UniShape
				with Retrieval	w/o Retrieval	w/o ReMGP	w/o Retrieval	w/o Retrieval	w/o Retrieval
1	ACSF1	0.9300	0.8000	0.9100	0.9000	0.8900	0.6800	0.8700	0.7500
2	Adiac	0.8414	0.8289	0.7612	0.7570	0.7428	0.6880	0.8338	0.8312
3	AllGestureWiimoteX	0.7743	0.7743	0.6714	0.6657	0.6186	0.5129	0.6600	0.7357
4	AllGestureWiimoteY	0.8086	0.7929	0.7271	0.7257	0.6400	0.6600	0.7300	0.7729

Appendix

Table 10: Detailed test classification accuracy results of the ablation study on 128 UCR time series datasets. The best results are in bold .

ID	Dataset name	Balanced						Random	Unbalanced
ID	Dataset name	50K	100K	200K	500K	1M	1.89M	100K	100K
1	ArrowHead	0.8628	0.8857	0.8857	0.8800	0.8571	0.8685	0.8857	0.8628
2	CBF	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
3	CricketX	0.8077	0.8282	0.8205	0.8205	0.8205	0.8154	0.8203	0.7997
4	DistalPhalanxOutlineAgeGroup	0.7698	0.7698	0.7626	0.7842	0.7914	0.8058	0.7698	0.7483
5	DistalPhalanxOutlineCorrect	0.7826	0.7899	0.7754	0.8152	0.7826	0.7826	0.7863	0.7790

Appendix

Table 11: Detailed test classification accuracy results on 18 UCR time series datasets under varying knowledge base sizes and sampling strategies. The best results are in bold .

Figure 6: t-SNE visualization on the Chinatown dataset.

Figure 7: t-SNE visualization on the ArrowHead dataset.

Figure 8: t-SNE visualization on the UWaveGestureLibraryAll dataset.

Figure 9: t-SNE visualization on the MelbournePedestrian dataset.

ID	Dataset name	Raw	Jitter	Time warp	Magnitude warp	Window slice	Time flip	Permutation	Scaling	R2R
1	ArrowHead	0.8857	0.8629	0.8457	0.8686	0.8629	0.8857	0.8400	0.8743	0.9257
2	CBF	1.0000	1.0000	1.0000	1.0000	1.0000	0.8389	1.0000	1.0000	1.0000
3	CricketX	0.8282	0.8154	0.8205	0.8128	0.8179	0.8308	0.8308	0.8179	0.8308
4	DistalPhalanxOutlineAgeGroup	0.7698	0.7698	0.7698	0.7554	0.7842	0.7842	0.7482	0.7338	0.7914
5	DistalPhalanxOutlineCorrect	0.7899	0.7935	0.7899	0.7971	0.7609	0.7609	0.7971	0.7935	0.7862
6	ECG5000	0.9498	0.9464	0.9387	0.9458	0.9444	0.9442	0.9464	0.9451	0.9462

Appendix

Table 12: Detailed test classification accuracy on 18 UCR datasets compared with different time series data augmentation methods. The best results are in bold .

Dataset A	Dataset B	Only Parameter transfer			R2R transfer
Dataset A	Dataset B	Acc ${}_{A_{1}}$	Acc ${}_{A_{2}}$	Forgetting	Acc ${}_{A_{1}}$	Acc ${}_{A_{2}}$	Forgetting
AconityMINIPrinterSmall_eq	AsphaltRegularityUni_eq	0.9795	0.8048	0.1747	0.9829	0.8253	0.1576
Colposcopy	GesturePebbleZ1_eq	0.3564	0.2574	0.0990	0.3762	0.2970	0.0792
AllGestureWiimoteY_eq	AllGestureWiimoteZ_eq	0.7971	0.2829	0.5142	0.8071	0.3286	0.4785
GestureMidAirD2_eq	GestureMidAirD3_eq	0.6923	0.1769	0.5154	0.7000	0.2308	0.4692

Appendix

Table 13: Impact of continual fine-tuning for classification on another four group datasets selected from the additional 30 time series datasets. Acc

{}_{A_{1}}

: Initial accuracy on dataset

A

; Acc

{}_{A_{2}}

: Accuracy on dataset

A

after continual training on dataset

B

; Forgetting: Acc

{}_{A_{1}}

- Acc

{}_{A_{2}}

ID	Dataset name	IP	L2	Cosine
1	ArrowHead	0.8800	0.8743	0.9257
2	CBF	1.0000	1.0000	1.0000
3	CricketX	0.8257	0.8257	0.8308
4	DistalPhalanxOutlineAgeGroup	0.7554	0.7339	0.7914
5	DistalPhalanxOutlineCorrect	0.7754	0.8080	0.7862
6	ECG5000	0.9400	0.9420	0.9462

Appendix

Table 14: Detailed test classification accuracy results on 18 UCR time series datasets using different distance metrics within R2R. The best results are in bold .

ID	Dataset name	$n_{\text{list}}$ =128	$n_{\text{list}}$ =256	$n_{\text{list}}$ =512	$n_{\text{list}}$ =1024	$n_{\text{list}}$ =2048	$n_{\text{list}}$ =4096	$n_{\text{list}}$ =8192
1	ArrowHead	0.9200	0.9257	0.9314	0.9314	0.9314	0.9543	0.9486
2	CBF	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
3	CricketX	0.8283	0.8308	0.8026	0.8232	0.8283	0.7975	0.8283
4	DistalPhalanxOutlineAgeGroup	0.7842	0.7914	0.7842	0.7886	0.7770	0.7698	0.7555
5	DistalPhalanxOutlineCorrect	0.8007	0.7862	0.8079	0.8043	0.7971	0.8007	0.8079
6	ECG5000	0.9444	0.9462	0.9462	0.9509	0.9431	0.9480	0.9391

Appendix

Table 15: Detailed test classification accuracy results on 18 UCR time series datasets using different

n_{\text{list}}

for retrieval within R2R. The best results are in bold .

ID	Dataset name	K=1	K=5	K=10	K=20	K=32	K=64
1	ArrowHead	0.9128	0.9543	0.9371	0.9257	0.9428	0.9486
2	CBF	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
3	CricketX	0.8287	0.8461	0.8385	0.8308	0.8410	0.8461
4	DistalPhalanxOutlineAgeGroup	0.7626	0.7770	0.7914	0.7914	0.7770	0.7842
5	DistalPhalanxOutlineCorrect	0.7608	0.7789	0.7681	0.7862	0.7789	0.7391
6	ECG5000	0.9409	0.9398	0.9467	0.9462	0.9495	0.9438

Appendix

Table 16: Detailed test classification accuracy results on 18 UCR time series datasets using different Top-

K

retrieval embeddings within R2R. The best results are in bold .

ID	Dataset name	Mantis			UniShape-based (ours)
ID	Dataset name	None	PCA	Linear	UniShape (Linear)	R2R ${}_{\text{UniShape}}$ (Linear)
1	CharacterTrajectories	0.976	0.977	0.988	0.995	0.995
2	DuckDuckGeese	0.540	0.580	0.600	0.600	0.627
3	EigenWorms	0.814	0.863	0.855	0.898	0.906
4	FaceDetection	0.549	0.553	0.592	0.664	0.657
5	HandMovementDirection	0.297	0.338	0.338	0.351	0.446

Appendix

Table 17: Test classification accuracy on ten UEA multivariate datasets under the linear channel-adapter setting. The Mantis columns are taken from Appendix Table 12 of [ 19 ] , and all accuracies are reported to three decimals. The best results are in bold .

Time Series Classification

Time series classification assigns labels to sequential data, using temporal patterns to support decisions in health, sensing, science, and industry.

See all 174 papers

Post-Training in Time Series Foundation Models: A Unifying Framework
Jul 22, 2026Shifeng Xie, Ambroise Odonnat, Zehao Xiao +7Model Compression Time-Series Foundation Models
Noah’s Ark Lab · Université Paris Cité · Inria +2
Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference
Jan 31, 2026Juntao Fang, Shifeng Xie, Shengbin Nie +7Time Series Classification In-Context Learning
Guangdong University of Technology, Guangzhou, China · Huawei Noah’s Ark Lab, Paris, France · Paris Descartes University, Paris, France +2
KairosHope: A Next-Generation Time-Series Foundation Model for Specialized Classification via Dual-Memory Architecture
May 18, 2026Luis Balderas, José Alberto Rodríguez, Miguel Lastra +2Memory-Augmented Neural Networks Time Series Classification
Department of Computer Science and Artificial Intelligence · DiCITS, iMUDS, DaSCI · Advanced Medical Imaging Group, Instituto de Investigación Biosanitaria de Granada (ibs.Granada) +3

Jul 22, 2026J/K move · Enter open · S save

Post-Training in Time Series Foundation Models: A Unifying Framework

Shifeng Xie, Ambroise Odonnat, Zehao Xiao, Lei Zan+6

Noah’s Ark Lab · Université Paris Cité · Inria+2

Model Compression Time-Series Foundation Models

Open paper AskPDF arXiv

Time series foundation models (TSFMs) have emerged as general-purpose models for time series analysis, but pretraining alone is often insufficient for reliable downstream deployment. Bridging this gap requires further intervention to handle domain shift, task heterogeneity, limited supervision, and computational constraints, which motivates post-training as a broad class of methods to adapt, augment, compose, calibrate, or specialize pretrained TSFMs for downstream tasks. In this work, we analyze TSFM post-training methods based on their locus of intervention in the prediction pipeline, yielding five categories: parameter adaptation, context augmentation, model composition, output processing and uncertainty control, and compression and specialization. Within each category, we study main representative methods and discuss their current limitations. We further identify future directions toward controlled adaptation, reliable context construction, uncertainty-aware model composition, calibrated output processing, and deployment-aware specialization. Overall, by providing a unifying framework for the emerging TSFM post-training landscape, this work aims to support future research to navigate the design space between a pretrained TSFM and its reliable downstream deployment.

Abstract

Time Series Classification

Explore similar work

Post-Training in Time Series Foundation Models: A Unifying Framework

Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference

KairosHope: A Next-Generation Time-Series Foundation Model for Specialized Classification via Dual-Memory Architecture