Memory efficient time series classification adapts to new classes fast
TRACE: Memory-Efficient Time Series Class-Incremental Learning via Temporal-Spectral Reconstruction
Machine Learning
Summary
Time series data from sensors often changes over time with new patterns emerging, and it’s important to recognize these new patterns without forgetting old ones. The authors propose TRACE, a method that remembers different patterns by building small models for each one, instead of storing all old data. This approach uses less memory and doesn’t require replaying previous examples, helping devices learn continuously in a privacy-friendly way. TRACE showed good accuracy on tests while using far less storage than previous methods.
What this means in practice
- •For sensor network engineers: Maintain accurate classification of evolving sensor data categories without storing all previous data, aiding memory-limited IoT devices.
- •For wearable device developers: Enable continuous learning of new user activity patterns on wearable devices while minimizing memory use and preserving privacy.
Authors
Junhui Mao, Junru Zhang, Liming Ye, Cheng Peng, Haoran Shi, Xiaoli Tang, Baokun Wang, Bin Dou, Yabo Dong, Han Yu
Abstract
Web-connected sensing applications continuously generate multivariate time series with evolving behavioral categories, making Time Series Class-Incremental Learning (TSCIL) essential for adapting to new classes without forgetting previously acquired knowledge. However, preserving diverse and evolving class distributions under strict privacy and memory constraints remains a fundamental challenge. Existing approaches often face a trade-off between retaining sufficient historical knowledge for accurate recognition and minimizing the memory required for continual adaptation. To address these limitations, we propose TRACE, a memory-efficient, replay-free framework that preserves class-specific knowledge through generative modeling rather than sample rehearsal. TRACE employs a compact, frozen temporal-spectral representation and lightweight class-specific autoencoders to capture complex intra-class variations. By independently preserving each class model, TRACE avoids recursive generation and catastrophic forgetting, while enabling direct classification through reconstruction-based scoring without retaining historical samples or large statistical matrices. Extensive experiments across five TSCIL benchmarks demonstrate that TRACE achieves competitive classification accuracy while reducing mean retained memory by approximately 98.26% compared with TS-ACL, the strongest evaluated analytical baseline, and achieving an approximately 33.43% relative improvement in accuracy on GrabMyo.