TexTSC: Class-Texture Preserving Data Condensation for Time Series Classification

Pouya Hosseinzadeh, Peiyu Li, Omar Bahri, Soukaina Filali Boubrahimi, Shah Muhammad Hamdi
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:4222-4230, 2026.

Abstract

Dataset condensation seeks to generate a small set of synthetic examples that can replace large real datasets for training, but existing methods for time series often rely on unstable training-trajectory matching or capture only limited signal structure. We present TexTSC, a condensation framework that preserves class structure using spectro-temporal second-order statistics instead of trajectory replay. TexTSC models each class’s “texture” as the co-activation pattern among intermediate teacher features, aligning Gram matrices of activations in time to capture temporal correlations and in frequency to capture spectral envelopes and harmonics. A short-lag autocorrelation term stabilizes local rhythm, while a lightweight gradient anchor at the final layer ensures discriminative power. TexTSC optimizes synthetic sequences directly, remains model-agnostic, and requires only closed-form statistics, making it simple and stable. Experiments on standard benchmarks show that TexTSC produces compact datasets that retain class-conditional structure and achieve higher classification accuracy than first-order or single-domain baselines.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-hosseinzadeh26a, title = { TexTSC: Class-Texture Preserving Data Condensation for Time Series Classification }, author = {Hosseinzadeh, Pouya and Li, Peiyu and Bahri, Omar and Boubrahimi, Soukaina Filali and Hamdi, Shah Muhammad}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {4222--4230}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/hosseinzadeh26a/hosseinzadeh26a.pdf}, url = {https://proceedings.mlr.press/v300/hosseinzadeh26a.html}, abstract = { Dataset condensation seeks to generate a small set of synthetic examples that can replace large real datasets for training, but existing methods for time series often rely on unstable training-trajectory matching or capture only limited signal structure. We present TexTSC, a condensation framework that preserves class structure using spectro-temporal second-order statistics instead of trajectory replay. TexTSC models each class’s “texture” as the co-activation pattern among intermediate teacher features, aligning Gram matrices of activations in time to capture temporal correlations and in frequency to capture spectral envelopes and harmonics. A short-lag autocorrelation term stabilizes local rhythm, while a lightweight gradient anchor at the final layer ensures discriminative power. TexTSC optimizes synthetic sequences directly, remains model-agnostic, and requires only closed-form statistics, making it simple and stable. Experiments on standard benchmarks show that TexTSC produces compact datasets that retain class-conditional structure and achieve higher classification accuracy than first-order or single-domain baselines. } }
Endnote
%0 Conference Paper %T TexTSC: Class-Texture Preserving Data Condensation for Time Series Classification %A Pouya Hosseinzadeh %A Peiyu Li %A Omar Bahri %A Soukaina Filali Boubrahimi %A Shah Muhammad Hamdi %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-hosseinzadeh26a %I PMLR %P 4222--4230 %U https://proceedings.mlr.press/v300/hosseinzadeh26a.html %V 300 %X Dataset condensation seeks to generate a small set of synthetic examples that can replace large real datasets for training, but existing methods for time series often rely on unstable training-trajectory matching or capture only limited signal structure. We present TexTSC, a condensation framework that preserves class structure using spectro-temporal second-order statistics instead of trajectory replay. TexTSC models each class’s “texture” as the co-activation pattern among intermediate teacher features, aligning Gram matrices of activations in time to capture temporal correlations and in frequency to capture spectral envelopes and harmonics. A short-lag autocorrelation term stabilizes local rhythm, while a lightweight gradient anchor at the final layer ensures discriminative power. TexTSC optimizes synthetic sequences directly, remains model-agnostic, and requires only closed-form statistics, making it simple and stable. Experiments on standard benchmarks show that TexTSC produces compact datasets that retain class-conditional structure and achieve higher classification accuracy than first-order or single-domain baselines.
APA
Hosseinzadeh, P., Li, P., Bahri, O., Boubrahimi, S.F. & Hamdi, S.M.. (2026). TexTSC: Class-Texture Preserving Data Condensation for Time Series Classification . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:4222-4230 Available from https://proceedings.mlr.press/v300/hosseinzadeh26a.html.

Related Material