TSMixer: An all-MLP Architecture for Time Series Forecasting

By Si Chen, Chun-Liang Li, Nate Yoder, Sercan Ö. Arik, Tomas Pfister. Trans. Mach. Learn. Res., 2023

439 citations recorded by the source corpus.

Real-world time-series datasets are often multivariate with complex dynamics. To capture this complexity, high capacity architectures like recurrent- or attention-based sequential deep learning models have become popular. However, recent work demonstrates that simple univariate linear models can outperform such deep learning models on several commonly used academic benchmarks. Extending them, in this paper, we investigate the capabilities of linear models for time-series forecasting and present Time-Series Mixer (TSMixer), a novel architecture designed by stacking multi-layer perceptrons (MLPs). TSMixer is based on mixing operations along both the time and feature dimensions to extract information efficiently. On popular academic benchmarks, the simple-to-implement TSMixer is comparable to specialized state-of-the-art models that leverage the inductive biases of specific benchmarks. On the challenging and large scale M5 benchmark, a real-world retail dataset, TSMixer demonstrates superior performance compared to the state-of-the-art alternatives. Our results underline the importance of efficiently utilizing cross-variate and auxiliary information for improving the performance of time series forecasting. We present various analyses to shed light into the capabilities of TSMixer. The design paradigms utilized in TSMixer are expected to open new horizons for deep learning-based time series forecasting. The implementation is available at https://github.com/google-research/google-research/tree/master/tsmixer

Subfields: Management Science and Operations Research, Signal Processing

View Paper: https://www.semanticscholar.org/paper/59694c8dce4f13db2f486eb8102459a3f7c23da6

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Management Science and Operations Research

In Signal Processing

Citations in the corpus

TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting (2024, methodology): “Due to the limitation of the main text, we also include three advanced baselines here: the general multi-scale framework Scaleformer (Shabani et al., 2023), two concurrent MLP-based model MTSMixer (Li et al., 2023) and TSMixer (Chen et al., 2023).”
TSMixer: Lightweight MLP-Mixer Model for Multivariate Time Series Forecasting (2023, background): “4Comparison with [9] is not included, as it was released in arXiv one month after the start of the KDD 2023 peer-review process [1], where our current paper has been submitted and got accepted.”
Spatio-Temporal Graph Neural Networks for Predictive Learning in Urban Computing: A Survey (2023, neutral): “Inaddition, someadvancementsintimeseriespredictionresearchhave alsochallengedthenecessityofemployingoverlycomplex temporallearningmodels[230],[231].”
Is Mamba Effective for Time Series Forecasting? (2024, positive): “As a new architecture, Mamba [22] swiftly attracted the attention of a large number of researchers in Natural Language Processing (NLP), Computer Vision (CV), and other Artificial Intelligence communities.”
TRACE: Grounding Time Series in Context for Multimodal Embedding and Retrieval (2025, unknown): “Linear-based models assume time-series signals can be effectively decomposed and modeled with simple linear projections [29, 30].”