NeurIPS2026论文:极简可解释架构DynaBase实现动力系统零样本重构
A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems [R]
这篇论文挑战了复杂模型在特定领域的必要性,用极简结构实现了超越主流基线的效果,对研究可解释AI和高效模型架构的同学极具参考价值。
In our #NeurIPS2026 paper “A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems (DS)” (preprint: https://arxiv.org/abs/2607.14937) we reduce a DS foundation model to the ingredients minimally necessary to faithfully reproduce long-term statistical and geometrical properties of DS:
在我们发表于 #NeurIPS2026 的论文《用于动力系统(DS)零样本重建的最小可解释架构》(预印本:https://arxiv.org/abs/2607.14937)中,我们将一个 DS 基础模型简化为忠实复现 DS 长期统计和几何特性所需的最少要素:
1) A piecewise affine map with only a single (!!) parameter α that controls local con-/divergence rates, and …
1) 一个分段仿射映射,仅含一个 (!!) 参数 α,用于控制局部的收敛/发散率,以及 …
2) … a context selector that chooses from the provided context signal the data point closest to the current state of the map, thus ensuring the generated dynamics stays close to the context in its temporal and geometrical properties.
2) … 一个上下文选择器,从提供的上下文信号中选择与当前映射状态最接近的数据点,从而确保生成的动力学在时间和几何特性上保持与上下文相近。
With just these two mechanisms, this minimal form – which we coined DynaBase – can reproduce all major dynamical regimes, including fixed points (α<1), limit cycles (α=1), and chaotic attractors (α>1). Thus, unlike other simple mechanisms like context parroting, DynaBase even preserves the correct dynamical regime!
仅凭这两个机制,这种我们命名为 DynaBase 的最小形式即可复现所有主要的动力学状态,包括不动点(α<1)、极限环(α=1)和混沌吸引子(α>1)。因此,与其他简单的机制(如上下文鹦鹉学舌)不同,DynaBase 甚至能保留正确的动力学状态!
Surprisingly, it turns out that this simple context-driven 1-parameter map outperforms most major time series and DS foundation models, as well as custom-trained models, in both long-term statistics and even short-term predictions, even when run in zero-shot mode.
令人惊讶的是,事实证明,这个简单的、由上下文驱动的 1 参数映射在长期统计和短期预测方面均优于大多数主流的时间序列和 DS 基础模型,以及经过定制训练的模型,即使在零样本模式下运行也是如此。
Both inference and training are extremely cheap – training can be done either analytically in one step by linear regression on forward-predictions, or by 1-parameter grid search directly on DS reconstruction objectives → this reveals interesting performance differences induced by different training mechanisms.
推理和训练成本都极低——训练可以通过对前向预测进行线性回归,以解析方式一步完成;或者直接在 DS 重建目标上进行 1 参数网格搜索 → 这揭示了由不同训练机制引起的有趣性能差异。
Most importantly in our minds, DynaBase owing to its formal simplicity may thus provide a tractable mathematical handle on analyzing, improving & understanding the performance and training of some time series and DS foundation models.
最重要的是,我们认为,DynaBase 凭借其形式上的简洁性,可能为分析、改进及理解某些时间序列和 DS 基础模型的性能与训练提供一个易于处理的数学切入点。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力