Time series classification with random convolution kernels based transforms: pooling operators and input representations matter

2024-09-02

arXiv.org

AI Summary

This work addresses the limitation in random convolutional kernel time-series classification—namely, the fixed input representations (e.g., raw, differenced, or magnitude) and pooling operations—which hinder adaptability to diverse datasets. We propose SelF-Rocket, an adaptive method built upon the MiniRocket framework that jointly learns, during training, the optimal input representation and pooling operator (e.g., max or min). Its key innovation is the first integration of a differentiable, learnable selection mechanism into the random convolutional kernel paradigm, overcoming the constraints of static preprocessing and pooling. Evaluated on all 128 UCR benchmark datasets, SelF-Rocket achieves state-of-the-art classification accuracy while maintaining inference speed comparable to MiniRocket. The method thus delivers both high efficiency and strong generalization across heterogeneous time-series domains.

Technology Category

• Machine Learning: Kernel Methods
• Reasoning under Uncertainty: Stochastic Optimization
• Search and Optimization: Non-convex Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

This article presents a new approach based on MiniRocket, called SelF-Rocket, for fast time series classification (TSC). Unlike existing approaches based on random convolution kernels, it dynamically selects the best couple of input representations and pooling operator during the training process. SelF-Rocket achieves state-of-the-art accuracy on the University of California Riverside (UCR) TSC benchmark datasets.

Problem

• Improves time series classification accuracy
• Dynamically selects input representations and pooling operators
• Achieves state-of-the-art results on UCR benchmark datasets

Innovation

• Dynamic selection of input representations
• Optimized pooling operator during training
• State-of-the-art accuracy on UCR datasets