![[2410.02081] MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with 0.1K Parameters](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2410.02081] MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with 0.1K Parameters
Summary
MixLinear introduces an ultra-lightweight model for multivariate time series forecasting, achieving high accuracy with only 0.1K parameters, suitable for resource-constrained devices.
Why It Matters
As demand for efficient forecasting models grows, MixLinear addresses the challenge of deploying complex models on devices with limited computational resources. This innovation could enhance accessibility and performance in various applications, particularly in IoT and mobile devices.
Key Takeaways
- MixLinear reduces parameter count significantly, enhancing efficiency.
- The model captures both temporal and frequency domain features effectively.
- It provides comparable or superior forecasting performance to state-of-the-art models.
- Designed for deployment on devices with limited computational capacity.
- Addresses the growing need for long-term time series forecasting solutions.
Computer Science > Machine Learning arXiv:2410.02081 (cs) [Submitted on 2 Oct 2024 (v1), last revised 16 Feb 2026 (this version, v2)] Title:MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with 0.1K Parameters Authors:Aitian Ma, Dongsheng Luo, Mo Sha View a PDF of the paper titled MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with 0.1K Parameters, by Aitian Ma and 2 other authors View PDF HTML (experimental) Abstract:Recently, there has been a growing interest in Long-term Time Series Forecasting (LTSF), which involves predicting long-term future values by analyzing a large amount of historical time-series data to identify patterns and trends. There exist significant challenges in LTSF due to its complex temporal dependencies and high computational demands. Although Transformer-based models offer high forecasting accuracy, they are often too compute-intensive to be deployed on devices with hardware constraints. On the other hand, the linear models aim to reduce the computational overhead by employing either decomposition methods in the time domain or compact representations in the frequency domain. In this paper, we propose MixLinear, an ultra-lightweight multivariate time series forecasting model specifically designed for resource-constrained devices. MixLinear effectively captures both temporal and frequency domain features by modeling intra-segment and inter-segment variations in the time domain and extracting frequency variati...