arXiv cs.AI· Vasilis Gkolemis, Loukas Kavouras, Dimitrios Kyriakopoulos, Konstantinos Tsopelas, Dimitrios Rontogiannis, Giuseppe Casalicchio, Theodore Dalamagas, Christos Diou·· 4 小时前AI 评分39
CALMs 条件可加局部模型：兼顾 GAM 可解释性与 GA²M 精度的设计化可解释方案
Interpretability-by-Design with Accurate Locally Additive Models and Conditional Feature Effects
AI 导读
论文提出条件可加局部模型（CALMs），缩小 GAM 与 GA²M 在可解释性与精度之间的取舍。CALMs 允许每个特征拥有多个单变量形状函数，各在由阈值逻辑条件定义的输入子区域内生效，从而保持局部可加性的同时捕捉特征交互，并配套基于蒸馏与区域感知 backfitting 的训练流程。
正文 · 原文
Abstract:Generalized additive models (GAMs) offer high interpretability through independent univariate feature effects but underfit when interactions are present in data. GA$^2$Ms add selected pairwise interactions which improves accuracy, but sacrifices interpretability. We propose Conditionally Additive Local Models (CALMs), a new model class that narrows this trade-off: it stays close to GAMs in interpretability and close to GA$^2$Ms in accuracy. CALMs allow multiple univariate shape functions per feature, each active in different regions of the input space. These regions are defined independently for each feature as simple logical conditions (thresholds) on the features it interacts with. As a result, effects remain locally additive while varying across subregions to capture interactions. We further propose a principled distillation-based training pipeline that identifies homogeneous regions with limited interactions and fits interpretable shape functions via region-aware backfitting. Experiments on diverse classification and regression tasks show that CALMs consistently outperform GAMs and achieve accuracy broadly comparable to GA$^2$Ms, while preserving the univariate auditability that GA$^2$Ms forfeit.
| Comments: | Accepted at NeurIPS 2026. 35 pages 10 figures. Code: this https URL |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.16503 [cs.LG] |
| (or arXiv:2602.16503v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2602.16503 arXiv-issued DOI via DataCite |
Submission history
From: Vasilis Gkolemis [view email]
[v1]
Wed, 18 Feb 2026 14:45:33 UTC (350 KB)
[v2]
Fri, 30 Oct 2026 17:53:59 UTC (459 KB)
来源：arXiv cs.AI · arxiv.org