Time series classification (TSC) has numerous applications across various domains. This research introduces a federated hybrid TSC method that combines image-based time series representation techniques with Convolutional Neural Networks (CNNs) in a decentralized framework. Specifically, three image representation techniques—Motif representation of numerical time series (NMotif), Motif representation of fuzzy time series (FTSMotif), and Polar—are utilized to convert raw data into image representations, which are then used as inputs for CNNs. Federated Learning (FL) enables collaborative training across distributed clients while ensuring data privacy. The method’s effectiveness is evaluated using 40 datasets from the UCR Archive. The results show that the FL model surpasses traditional baselines by up to 1.9% while enabling collaborative learning and preserving privacy. Although centralized state-of-the-art (SOTA) models such as Rocket and ResNet achieve higher overall accuracy, the proposed approach offers a competitive and practical alternative for real-world scenarios where centralized training is infeasible due to privacy constraints. These findings highlight the potential of our proposed federated image-based TSC as a scalable and privacy-aware solution for time series analysis.
Federated learning (FL) is a cutting-edge technology in artificial intelligence that preserves data privacy and security while reducing the cost of computation and communication. It transforms traditional centralized machine learning and deep learning approaches to enable decentralized model training without the need for data exchange. This work presents Flautim, the first implementation of an FL platform in Brazil and all around Latin America based on Kubernetes (K8S) and Flower framework. Flautim is designed for academic use, enabling researchers without a technical background to conduct FL experiments on this platform easily. Also, this platform allows for the development of applications involving data gathered from connected vehicles. Thus, this study aims to introduce this new FL platform, providing comprehensive details of its architecture.
In recent years, the application of Large Language Models (LLMs) to time series forecasting (TSF) has garnered significant attention among researchers. This study presents a new frame of LLMs named CGF-LLM using GPT-2 combined with fuzzy time series (FTS) and causal graph to predict multivariate time series, marking the first such architecture in the literature. The key objective is to convert numerical time series into interpretable forms through the parallel application of fuzzification and causal analysis, enabling both semantic understanding and structural insight as input for the pretrained GPT-2 model. The resulting textual representation offers a more interpretable view of the complex dynamics underlying the original time series. The reported results confirm the effectiveness of our proposed LLM-based time series forecasting model, as demonstrated across four different multivariate time series datasets. This initiative paves promising future directions in the domain of TSF using LLMs based on FTS.