AI Insight
ConceptTime is a new machine learning method for time-series classification that achieves accuracy, interpretability, and the ability to abstain from predictions when evidence is uncertain. The approach uses "predictive surprise"—the mismatch between a probabilistic forecaster's predictions and actual observations—to automatically identify meaningful segment boundaries and characterize temporal patterns without requiring fixed windows or expert annotations. Testing on 17 public benchmarks shows ConceptTime matches state-of-the-art black-box model accuracy while outperforming all interpretable baselines, and maintains 98% accuracy on epilepsy detection using only 1% of training labels.
Why it matters
This method addresses a critical challenge in deploying AI systems for medical diagnosis, financial monitoring, and other risk-sensitive applications where both accuracy and explainability are essential. By eliminating the need for expert annotations or predefined pattern vocabularies, it could enable interpretable AI deployment in domains where expert time is scarce or expensive.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Time-series classifiers deployed in risk-sensitive domains must be (1) accurate, (2) interpretable, and (3) able to abstain when evidence is weak. Existing methods do not deliver all 3 at once, and the obstacle is structural, rather than logistical. Meaningful temporal patterns cannot be defined without knowing segment boundaries, yet meaningful boundaries cannot be placed without knowing which patterns to expect. Existing approaches break this circularity by using fixed windows or expert-annotated vocabularies, thereby sacrificing at least one of the 3 properties. We introduce ConceptTime, which closes the loop through a single self-supervised signal. A frozen probabilistic forecaster predicts a distribution over the next observation at every timestep of the sequence. We call the mismatch between its prediction and the realized signal the predictive surprise. This signal both locates segment boundaries and characterizes each resulting segment through the forecaster. Segments are thus grounded by how they behaved relative to expectation, not by what they look like. These summaries cluster into a vocabulary of dynamical regimes that we call concepts. A lightweight head classifies from the concept sequence, and the distance to the nearest concept yields reliability by design. Across 17 public benchmarks, ConceptTime is competitive with state-of-the-art black-box accuracy, beats every interpretable baseline, and retains near-98% accuracy on Epilepsy with 1% of labels. It further remains competitive with softmax, prototype-distance, and SHAP/TimeX/TimeX++ on OOD detection and deletion faithfulness, without a single annotation, language model call, or domain expert. Code is available at: https://github.com/Sachithx/ConceptTime.git.
Source: Predictive Surprise as Self-Grounding Concept Bottleneck for Interpretable Time Series