首页 > AI前沿 > Learning Temporal Dynamics with Derivative Representations for Time Series Forecasting

Learning Temporal Dynamics with Derivative Representations for Time Series Forecasting

arXiv机器学习 2026-11-12 17:12 26 阅读 查看原文
用户希望将一段纯文本的新闻正文排版成语义化的HTML片段。 角色:专业的新闻排版工程师。 规则: 1. 只输出HTML片段本身,不要Markdown围栏,不要解释。 2. 每个自然段落用`

`包裹,长文用`

`/`

`划分小节。 3. 列表内容用`
  • `,引用用`
    `,代码用`
    `,重要句子可用``。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出``/``/``等文档级标签。
    
    正文内容是一段关于时间序列预测(DeltaTSF)的学术摘要/新闻。
    原文:
    Time series forecasting aims to predict the future evolution of complex systems by learning their underlying dynamics from past observations. Despite recent advances in deep learning, Transformers in particular still struggle to capture the temporal evolution of system dynamics. Existing approaches have been largely driven by architectural innovation, with comparatively little attention paid to training strategies that explicitly exploit the temporal structure governing system dynamics. In this work, we introduce DeltaTSF, a dynamics-aware forecasting framework based on discrete derivative representations of time series. Specifically, DeltaTSF jointly learns the series and their discrete temporal derivatives across both input and forecast horizons within a unified, model-agnostic framework. By coupling these representations over time, the model is encouraged to maintain temporal consistency, a property we show to induce a form of discrete Taylor consistency during training. This formulation enables the model to better reflect the underlying system dynamics, resulting in more stable and physically coherent forecasts. Notably, DeltaTSF does not require explicit derivative computation and introduces no additional computational overhead at inference time. Extensive experiments across diverse benchmarks and backbone architectures demonstrate that DeltaTSF consistently improves forecasting performance and yields more dynamically coherent predictions than strong SOTA methods. These results highlight the effectiveness of enforcing dynamic coherence through derivative representations as a model-agnostic paradigm that operates alongside and complements architectural design. Source code will be made publicly available upon acceptance.
    
    分析原文结构:
    这是一段连续的文本,没有明显的列表、引用或代码。可以将其划分为几个自然段,或者加一个小标题。
    因为是新闻/摘要,可以加一个 `

    ` 作为小节标题,但规则说“不得增删改内容”,所以不能自己加标题文字。如果加 `

    ` 必须用原文的词,但原文没有标题。所以不使用 `

    ` 或 `

    `,或者把第一句作为 `

    `?不,规则说“长文用

    /

    划分小节”,如果不加标题就不加,直接用 `

    `。 可以将原文分为几个自然段: 段落1:Time series forecasting aims to predict... governing system dynamics. 段落2:In this work, we introduce DeltaTSF... model-agnostic framework. 段落3:By coupling these representations... physically coherent forecasts. 段落4:Notably, DeltaTSF does not require... inference time. 段落5:Extensive experiments across diverse... architectural design. 段落6:Source code will be made publicly available upon acceptance. 或者保持为一个段落?“每个自然段落用

    包裹”,原文是一整段,我可以按逻辑拆分成几个 `

    `,或者就一个 `

    `。为了排版美观,拆成几个逻辑段落比较好。 拆分:

    Time series forecasting aims to predict the future evolution of complex systems by learning their underlying dynamics from past observations. Despite recent advances in deep learning, Transformers in particular still struggle to capture the temporal evolution of system dynamics. Existing approaches have been largely driven by architectural innovation, with comparatively little attention paid to training strategies that explicitly exploit the temporal structure governing system dynamics.

    In this work, we introduce DeltaTSF, a dynamics-aware forecasting framework based on discrete derivative representations of time series. Specifically, DeltaTSF jointly learns the series and their discrete temporal derivatives across both input and forecast horizons within a unified, model-agnostic framework.

    By coupling these representations over time, the model is encouraged to maintain temporal consistency, a property we show to induce a form of discrete Taylor consistency during training. This formulation enables the model to better reflect the underlying system dynamics, resulting in more stable and physically coherent forecasts.

    Notably, DeltaTSF does not require explicit derivative computation and introduces no additional computational overhead at inference time.

    Extensive experiments across diverse benchmarks and backbone architectures demonstrate that DeltaTSF consistently improves forecasting performance and yields more dynamically coherent predictions than strong SOTA methods. These results highlight the effectiveness of enforcing dynamic coherence through derivative representations as a model-agnostic paradigm that operates alongside and complements architectural design.

    Source code will be made publicly available upon acceptance.

    检查规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 -> 好的。 2. 每个自然段落用

    包裹,长文用

    /

    划分小节。 -> 没有加额外的标题文字,所以不用h2/h3。 3. 列表内容用
    • ,引用用
      ,代码用
      ,重要句子可用。 -> 使用了强调DeltaTSF。
      4. 保留原文所有信息,不得增删改内容。 -> 检查文本是否完全一致。
      原文:Time series forecasting aims to predict the future evolution of complex systems by learning their underlying dynamics from past observations. Despite recent advances in deep learning, Transformers in particular still struggle to capture the temporal evolution of system dynamics. Existing approaches have been largely driven by architectural innovation, with comparatively little attention paid to training strategies that explicitly exploit the temporal structure governing system dynamics. In this work, we introduce DeltaTSF, a dynamics-aware forecasting framework based on discrete derivative representations of time series. Specifically, DeltaTSF jointly learns the series and their discrete temporal derivatives across both input and forecast horizons within a unified, model-agnostic framework. By coupling these representations over time, the model is encouraged to maintain temporal consistency, a property we show to induce a form of discrete Taylor consistency during training. This formulation enables the model to better reflect the underlying system dynamics, resulting in more stable and physically coherent forecasts. Notably, DeltaTSF does not require explicit derivative computation and introduces no additional computational overhead at inference time. Extensive experiments across diverse benchmarks and backbone architectures demonstrate that DeltaTSF consistently improves forecasting performance and yields more dynamically coherent predictions than strong SOTA methods. These results highlight the effectiveness of enforcing dynamic coherence through derivative representations as a model-agnostic paradigm that operates alongside and complements architectural design. Source code will be made publicly available upon acceptance.
      
      对比HTML中的文本:
      Time series forecasting aims to predict the future evolution of complex systems by learning their underlying dynamics from past observations. Despite recent advances in deep learning, Transformers in particular still struggle to capture the temporal evolution of system dynamics. Existing approaches have been largely driven by architectural innovation, with comparatively little attention paid to training strategies that explicitly exploit the temporal structure governing system dynamics.
      In this work, we introduce DeltaTSF, a dynamics-aware forecasting framework based on discrete derivative representations of time series. Specifically, DeltaTSF jointly learns the series and their discrete temporal derivatives across both input and forecast horizons within a unified, model-agnostic framework.
      By coupling these representations over time, the model is encouraged to maintain temporal consistency, a property we show to induce a form of discrete Taylor consistency during training. This formulation enables the model to better reflect the underlying system dynamics, resulting in more stable and physically coherent forecasts.
      Notably, DeltaTSF does not require explicit derivative computation and introduces no additional computational overhead at inference time.
      Extensive experiments across diverse benchmarks and backbone architectures demonstrate that DeltaTSF consistently improves forecasting performance and yields more dynamically coherent predictions than strong SOTA methods. These results highlight the effectiveness of enforcing dynamic coherence through derivative representations as a model-agnostic paradigm that operates alongside and complements architectural design.
      Source code will be made publicly available upon acceptance.
      
      完全一致,没有增删改。
      
      格式:
      

      Time series forecasting aims to predict the future evolution of complex systems by learning their underlying dynamics from past observations. Despite recent advances in deep learning, Transformers in particular still struggle to capture the temporal evolution of system dynamics. Existing approaches have been largely driven by architectural innovation, with