![[2602.12744] Adaptive Structured Pruning of Convolutional Neural Networks for Time Series Classification](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.12744] Adaptive Structured Pruning of Convolutional Neural Networks for Time Series Classification
Summary
This article presents Dynamic Structured Pruning (DSP), an innovative method for optimizing convolutional neural networks in time series classification, achieving significant model compression while maintaining accuracy.
Why It Matters
As deep learning models become more prevalent in time series classification, their high resource demands pose challenges for deployment on limited-capacity devices. DSP addresses these issues by automating the pruning process, enhancing scalability and efficiency, which is crucial for real-world applications.
Key Takeaways
- DSP automates the pruning of convolutional neural networks, eliminating the need for manual hyperparameter tuning.
- The method achieves an average model compression of 58% for LITETime and 75% for InceptionTime without sacrificing accuracy.
- DSP enhances deployment feasibility for deep learning models on resource-constrained devices.
- The framework utilizes instance-wise sparsity loss and global activation analysis for effective pruning.
- Validation across 128 UCR datasets demonstrates DSP's robustness and efficiency in time series classification.
Computer Science > Machine Learning arXiv:2602.12744 (cs) [Submitted on 13 Feb 2026] Title:Adaptive Structured Pruning of Convolutional Neural Networks for Time Series Classification Authors:Javidan Abdullayev, Maxime Devanne, Cyril Meyer, Ali Ismail-Fawaz, Jonathan Weber, Germain Forestier View a PDF of the paper titled Adaptive Structured Pruning of Convolutional Neural Networks for Time Series Classification, by Javidan Abdullayev and 5 other authors View PDF HTML (experimental) Abstract:Deep learning models for Time Series Classification (TSC) have achieved strong predictive performance but their high computational and memory requirements often limit deployment on resource-constrained devices. While structured pruning can address these issues by removing redundant filters, existing methods typically rely on manually tuned hyperparameters such as pruning ratios which limit scalability and generalization across datasets. In this work, we propose Dynamic Structured Pruning (DSP), a fully automatic, structured pruning framework for convolution-based TSC models. DSP introduces an instance-wise sparsity loss during training to induce channel-level sparsity, followed by a global activation analysis to identify and prune redundant filters without needing any predefined pruning ratio. This work tackles computational bottlenecks of deep TSC models for deployment on resource-constrained devices. We validate DSP on 128 UCR datasets using two different deep state-of-the-art archite...
