数据扰动下模型级联的准确性与鲁棒性
Accuracy and Robustness of Model Cascades Under Data Perturbations
浏览论文内容
中文总结 AI 辅助
本文针对图像分类的基于置信度的模型级联，发现数据扰动会引发三种失效模式，需评估其在分布偏移下的路由可靠性，以优化节能模型的鲁棒性。
中文摘要 AI 辅助
预测级联在保持高预测性能的同时，可显著降低人工智能（AI）模型的能耗。其核心思路是将简单输入路由至轻量小型模型处理，而将困难的不确定案例交由更大的模型处理。尽管该设计在干净数据上可提升计算效率，但其有效性取决于基于置信度的路由的可靠性。输入退化（如静态损坏和序列扰动）会改变模型置信度与路由决策。本文针对图像分类研究基于置信度的级联框架，探究此类退化如何影响其基于置信度的延迟行为。我们选取在准确性、路由质量与能耗的帕累托最优处的模型级联，该级联实现了具有竞争力的预测性能，CO₂排放量最多可降低10倍。我们研究该模型级联在输入损坏下的行为，分析当输入分布偏移时，级联的路由决策如何变化。我们的分析识别出三种失效模式：静态损坏要么（1）在大型模型仍可用时破坏路由信号，要么（2）使两个模型均退化，导致延迟不再恢复准确性；序列扰动揭示了第三种模式：预测稳定但延迟被抑制，产生稳定但不可靠的预测。这些发现表明，节能型模型级联需要超越干净准确性的评估，明确关注分布偏移下的路由可靠性。
英文摘要
Prediction cascades significantly reduce energy consumption of Artificial Intelligence (AI) models while maintaining high predictive performance. The idea is that easy inputs are routed through a lightweight small model, and difficult uncertain cases are deferred to a larger model. While this design can improve computational efficiency on clean data, its effectiveness depends on the reliability of confidence-based routing. Input degradations, such as static corruptions and sequential perturbations, can shift model confidence and routing decisions. In this paper, we study confidence-based cascade frameworks for image classification and investigate how such degradations affect their confidence-based deferral behavior. We select a model cascade at the pareto-optimum of accuracy, routing quality, and energy consumption that achieves competitive predictive performance with an up to 10-fold decrease in CO$_2$ emissions. We study the behavior of that model cascade under input corruptions and analyze how the cascade's routing decisions change when the input distribution shifts. Our analysis identifies three failure modes. Static corruptions either (1) break the routing signal while the large model remains useful, or (2) degrade both models so deferral no longer recovers accuracy. Sequential perturbations reveal a third mode: predictions stabilize but deferral suppresses, yielding stable but unreliable predictions. These findings demonstrate that energy efficient model cascades require evaluation beyond clean accuracy, with explicit attention to routing reliability under distribution shift.
发表机构
- AUMOVIO AI Lab（奥莫维奥人工智能实验室）
- Berliner Hochschule für Technik（柏林工程应用科学大学）
机构由 AI 辅助整理，请以论文原文为准。
