REVIEW 22 cited by
TSMixer: An All-MLP Architecture for Time Series Forecasting
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
TSMixer: An All-MLP Architecture for Time Series Forecasting
read the original abstract
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
Forward citations
Cited by 22 Pith papers
-
JaGuard: Position Error Correction of GNSS Jamming with Deep Temporal Graphs
JaGuard recasts GNSS jamming mitigation as dynamic graph regression and uses a Heterogeneous Graph ConvLSTM to estimate 2D positional deviation, achieving MAEs of 2.26-2.61 cm on mixed-power real datasets and remainin...
-
Deep Time Series Models: A Comprehensive Survey and Benchmark
This survey and benchmark of deep time series models using the released TSLib library finds that models with specific structures perform well only on distinct analysis tasks.
-
A decoder-only foundation model for time-series forecasting
A pretrained decoder-only patched transformer achieves near state-of-the-art zero-shot forecasting performance across diverse time series datasets and settings.
-
How Good Can Linear Models Be for Time-Series Forecasting?
Tuned Ridge regression with series-specific preprocessing beats prior linear forecasters and complex neural baselines on most dataset-horizon pairs across eight benchmarks.
-
How Good Can Linear Models Be for Time-Series Forecasting?
Optimized Ridge regression with series-specific preprocessing beats prior linear forecasters and exceeds Transformer, MLP, and CNN baselines on six of eight time-series benchmarks.
-
QuITE: Query-Based Irregular Time Series Embedding
QuITE is a query-token embedding layer that aggregates irregular observations via self-attention to produce backbone-compatible latents without generating artificial values.
-
Three-Stage Learning Unlocks Strong Performance in Simple Models for Long-Term Time Series Forecasting
STAIR's three-stage training enables simple temporal models to match or exceed complex baselines on long-term forecasting benchmarks by combining shared learning, individual adaptation, and residual cross-variable modeling.
-
Exploring the Potential of Probabilistic Transformer for Time Series Modeling: A Report on the ST-PT Framework
ST-PT turns transformers into explicit factor graphs for time series, enabling structural injection of symbolic priors, per-sample conditional generation, and principled latent autoregressive forecasting via MFVI iterations.
-
Amortized Predictability-aware Training Framework for Time Series Forecasting and Classification
APTF reweights training samples by loss-based predictability buckets and uses an amortization model to stabilize the estimates, improving accuracy across TSF and TSC baselines.
-
HieraMix: A Hierarchical MLP-Mixer for Large-Scale Traffic Forecasting
HSTMixer, a hierarchical all-MLP mixer with adaptive region mixing, reports state-of-the-art MAE/RMSE/MAPE on four LargeST datasets at 15-minute resolution.
-
Dynamic Relational Priming Improves Transformer in Multivariate Time Series
Prime attention modulates attention keys and values per channel-pair and reports improved MTS forecasting accuracy across several benchmarks.
-
GatedLinear: Adaptive Routing of Complementary Linear Bases for Time Series Forecasting
Adaptive soft routing among three complementary linear bases via a channel-horizon-phase gate yields competitive multivariate forecasting accuracy with a small, interpretable model.
-
PIER: Physics-Informed Environmental Retrieval for Time-Series Modeling
PIER augments embedding-based retrieval for lake modeling with a physics-aware stream scored by local verifiers, improving water temperature and dissolved oxygen prediction across 356 lakes.
-
CLOUDADV: Decision-Aligned Instance Sizing with Zero-Shot Foundation Models under Drift
CLOUDADV combines zero-shot forecasting with LLM-generated recommendations for cloud instance sizing, reporting 52.9% simulated monthly cost savings in a seven-VM Azure case study.
-
Self-Adaptive Scale Handling for Forecasting Time Series with Scale Heterogeneity
The self-adaptive scale-handling (AS) module with scale calibrating and scaling selection learns per-input scale factors to improve forecasting on scale-heterogeneous time series.
-
CHAM-net: A Contrastive Hierarchical Adaptive Meta-network for Robust Global Methane Flux Prediction
CHAM-net is a contrastive hierarchical adaptive meta-network that conditions predictions on historical site data to outperform baselines on methane flux tasks with nRMSE down to 0.43.
-
GenHAR: Generalizing Cross-domain Human Activity Recognition for Last-mile Delivery
GenHAR generalizes cross-domain human activity recognition by 9.97% accuracy and 6.4x lower FLOPs via tokenized sensor data, frequency channel correlations, selective masking, and efficient attention, with deployment ...
-
CaReTS: A Multi-Task Framework Unifying Classification and Regression for Time Series Forecasting
CaReTS forecasts multi-step time series by combining a trend classifier with a deviation regressor in a residual, uncertainty-weighted multi-task framework.
-
FlowMixer: A Depth-Agnostic Neural Architecture for Interpretable Spatiotemporal Forecasting
A single-layer architecture called FlowMixer uses constrained matrix operations and a semi-group property to enable depth-agnostic, interpretable spatiotemporal forecasting with direct eigenmode extraction.
-
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Affine mapping dominates LTSF benchmarks by learning similar input-to-output transition matrices, captures periodic signals well but struggles with non-periodic or cross-channel varying periods; reversible normalizati...
-
Understanding Key Features of Time Series Foundation Models from Epidemic Forecasting
Mixture-of-experts fusing multiple pretrained forecasters achieves strongest performance on influenza time series, with pretraining gains largest at longer horizons when domain-aligned and LLM methods underperforming.
-
SEMixer: Semantics Enhanced MLP-Mixer for Multiscale Mixing and Long-term Time Series Forecasting
SEMixer combines random-mask patch interactions with progressive adjacent-scale mixing and reports improved MSE/MAE on common long-term forecasting benchmarks.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.