Recurrent Domain Incremental Adaptation of Time-Series Foundation Models

Hugging Face Daily Papers Papers

Summary

The paper proposes PACT (Piecewise Adaptation via Chart Transport), a method that adapts pretrained time-series foundation models to new operating conditions by partitioning the frozen embedding space into local charts and learning gated residual transports, achieving macro-F1 above 0.95 with only 5% labeled calibration data and no condition identity at test time.

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

Cached at: 11/19/26, 03:39 AM

Paper page - Recurrent Domain Incremental Adaptation of Time-Series Foundation Models

Source: https://huggingface.co/papers/2611.24075

Abstract

PACT adapts a pretrained time-series foundation model to new operating conditions by partitioning its frozen embedding space into local charts and learning gated residual transports that preserve previously learned conditions, achieving high macro-F1 with limited labeled calibration data and no condition identity at test time.

Pretrained Time-Series Foundation Models (TSFMs)provide strong representations fordownstream classificationtasks. Yet adapting a TSFM under changing operating conditions remains challenging. A model must improve on each newly observed condition without sacrificing conditions it has already learned, often from only a small labeled calibration set and without knowing the operating condition at inference time. We find that this difficulty has an important geometric structure. Under anoperating-condition shift, theembedding spaceof a TSFM does not move uniformly: some local regions change substantially, while others remain nearly unchanged. Consequently, a global adaptation can unnecessarily move embeddings that were already correctly classified. To address this, we propose the novelPACT (Piecewise Adaptation via Chart Transport). PACT partitions thefrozen source embedding spaceintolocal chartsand learnslightweight residual transportsfor the regions affected by each new condition. An explicitidentity pathis always retained, and alearned gatesoftly selects among the identity and accumulated transports for each input. Thepretrained TSFM encoder,classifier head, and previously learned transports remain fixed. With only 5%labeled calibration dataper arriving condition and nocondition identity at test time, PACT maintains an averagemacro-F1above 0.95 across a four-conditionbearing-fault stream. Underhomogeneous tasksandlabel spaces, similarretention-adaptationbehavior is observed across public benchmarks. Code is available at https://github.com/Sachithx/PACT.

View arXiv pageView PDFProject pageGitHub0Add to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2611.24075 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2611.24075 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2611.24075 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Adaptive Patching Is Harder Than It Looks For Time-Series Forecasting

arXiv cs.LG

This paper theoretically and empirically examines adaptive patching for time-series Transformers, deriving conditions under which content-adaptive tokenization should outperform tuned uniform patching. Controlled experiments on standard benchmarks show that a well-tuned uniform baseline is competitive with dynamic patching methods, challenging the assumed benefit of adaptive approaches.

Time-Series Foundation Models That Understand Data Revisions

arXiv cs.LG

The paper proposes VINTAGE-TS, a revision-aware adaptation of a time-series foundation model that distinguishes observation time from information-availability time, and provides a framework for evaluating forecasts with data revisions.