Time series models improve classification by using explicit knowledge retrieval
From Retrieval to Recognition:Knowledge-Enhanced Post-training of Time Series Foundation Models for Classification
Machine Learning
Summary
Classifying time series data like heartbeats or stock prices can be tricky because current models mix useful information with lots of irrelevant details. The authors introduce a method called Retrieval-to-Recognition (R2R) that helps models pull out and use relevant information more clearly. Instead of just tweaking all model settings, R2R uses a special part to fetch helpful data from a knowledge base, which is then combined with what the model already knows. This makes the classification more accurate and efficient, as tests on many datasets showed.
What this means in practice
- •For data science teams: Enhance time series classification models by integrating instance-level knowledge retrieval for better accuracy on diverse datasets.
- •For financial analytics teams: Improve anomaly detection and pattern recognition in financial time series by using knowledge-enhanced post-training methods.
Authors
Zhen Liu, Yucheng Wang, Yingpeng Du, Tianjun Wei, Qianli Ma, Min Wu, Jie Zhang, Zhu Sun
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.