ShapeX: Shapelet-Driven Post Hoc Explanations for Time Series Classification Models
ShapeX: Shapelet-Driven Post Hoc Explanations for Time Series Classification Models [111.3] 時系列を意味のあるシェープレット駆動セグメントに分割する革新的なフレームワークであるShapeXを紹介する。 ShapeXの中核にはShapelet Describe-and-Detectフレームワークがあり、分類に不可欠なさまざまなシェイプレットを効果的に学習する。 論文参考訳（メタデータ） (Thu, 23 Oct 2025 00:01:40 GMT)
時系列分類に関する説明手法、「we introduce SHAPEX, a novel approach that segments the time series into meaningful subsequences and computes Shapley value [13] as saliency scores. Instead of distributing importance across individual timesteps, SHAPEX aggregates timesteps into cohesive, shapelet-driven segments that serve as “players” in the Shapley value computation. By measuring each segment’s marginal contribution to the black-box model’s prediction, this method clearly identifies which subsequences significantly influence classification outcomes.」