Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts

• URL: http://arxiv.org/abs/2410.10469v1
• Date: Mon, 14 Oct 2024 13:01:11 GMT
• Title: Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts
• Authors: Xu Liu, Juncheng Liu, Gerald Woo, Taha Aksu, Yuxuan Liang, Roger Zimmermann, Chenghao Liu, Silvio Savarese, Caiming Xiong, Doyen Sahoo,
• Abstract summary: This paper introduces Moirai-MoE, using a single input/output projection layer while delegating the modeling of diverse time series patterns to the sparse mixture of experts. Extensive experiments on 39 datasets demonstrate the superiority of Moirai-MoE over existing foundation models in both in-distribution and zero-shot scenarios.
• Score: 103.725112190618
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Time series foundation models have demonstrated impressive performance as zero-shot forecasters. However, achieving effectively unified training on time series remains an open challenge. Existing approaches introduce some level of model specialization to account for the highly heterogeneous nature of time series data. For instance, Moirai pursues unified training by employing multiple input/output projection layers, each tailored to handle time series at a specific frequency. Similarly, TimesFM maintains a frequency embedding dictionary for this purpose. We identify two major drawbacks to this human-imposed frequency-level model specialization: (1) Frequency is not a reliable indicator of the underlying patterns in time series. For example, time series with different frequencies can display similar patterns, while those with the same frequency may exhibit varied patterns. (2) Non-stationarity is an inherent property of real-world time series, leading to varied distributions even within a short context window of a single time series. Frequency-level specialization is too coarse-grained to capture this level of diversity. To address these limitations, this paper introduces Moirai-MoE, using a single input/output projection layer while delegating the modeling of diverse time series patterns to the sparse mixture of experts (MoE) within Transformers. With these designs, Moirai-MoE reduces reliance on human-defined heuristics and enables automatic token-level specialization. Extensive experiments on 39 datasets demonstrate the superiority of Moirai-MoE over existing foundation models in both in-distribution and zero-shot scenarios. Furthermore, this study conducts comprehensive model analyses to explore the inner workings of time series MoE foundation models and provides valuable insights for future research.

Related papers

• DAM: Towards A Foundation Model for Time Series Forecasting [0.8231118867997028]
  We propose a neural model that takes randomly sampled histories and outputs an adjustable basis composition as a continuous function of time. It involves three key components: (1) a flexible approach for using randomly sampled histories from a long-tail distribution; (2) a transformer backbone that is trained on these actively sampled histories to produce, as representational output; and (3) the basis coefficients of a continuous function of time.
  arXiv  Detail & Related papers  (2024-07-25T08:48:07Z)
• Deep Time Series Models: A Comprehensive Survey and Benchmark [74.28364194333447]
  Time series data is of great significance in real-world scenarios. Recent years have witnessed remarkable breakthroughs in the time series community. We release Time Series Library (TSLib) as a fair benchmark of deep time series models for diverse analysis tasks.
  arXiv  Detail & Related papers  (2024-07-18T08:31:55Z)
• MOMENT: A Family of Open Time-series Foundation Models [19.0845213853369]
  We introduce MOMENT, a family of open-source foundation models for general-purpose time series analysis. We compile a collection of public time series, called the Time series Pile, and systematically tackle time series-specific challenges. We build on recent work to design a benchmark to evaluate time series foundation models on diverse tasks and datasets in limited supervision settings.
  arXiv  Detail & Related papers  (2024-02-06T10:48:46Z)
• Unified Training of Universal Time Series Forecasting Transformers [104.56318980466742]
  We present a Masked-based Universal Time Series Forecasting Transformer (Moirai) Moirai is trained on our newly introduced Large-scale Open Time Series Archive (LOTSA) featuring over 27B observations across nine domains. Moirai achieves competitive or superior performance as a zero-shot forecaster when compared to full-shot models.
  arXiv  Detail & Related papers  (2024-02-04T20:00:45Z)
• Timer: Generative Pre-trained Transformers Are Large Time Series Models [83.03091523806668]
  This paper aims at the early development of large time series models (LTSM) During pre-training, we curate large-scale datasets with up to 1 billion time points. To meet diverse application needs, we convert forecasting, imputation, and anomaly detection of time series into a unified generative task.
  arXiv  Detail & Related papers  (2024-02-04T06:55:55Z)
• TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting [24.834846119163885]
  We propose a novel framework, TEMPO, that can effectively learn time series representations. TEMPO expands the capability for dynamically modeling real-world temporal phenomena from data within diverse domains.
  arXiv  Detail & Related papers  (2023-10-08T00:02:25Z)
• Multi-scale Attention Flow for Probabilistic Time Series Forecasting [68.20798558048678]
  We propose a novel non-autoregressive deep learning model, called Multi-scale Attention Normalizing Flow(MANF) Our model avoids the influence of cumulative error and does not increase the time complexity. Our model achieves state-of-the-art performance on many popular multivariate datasets.
  arXiv  Detail & Related papers  (2022-05-16T07:53:42Z)
• Model-Attentive Ensemble Learning for Sequence Modeling [86.4785354333566]
  We present Model-Attentive Ensemble learning for Sequence modeling (MAES) MAES is a mixture of time-series experts which leverages an attention-based gating mechanism to specialize the experts on different sequence dynamics and adaptively weight their predictions. We demonstrate that MAES significantly out-performs popular sequence models on datasets subject to temporal shift.
  arXiv  Detail & Related papers  (2021-02-23T05:23:35Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.