Dec 7, 2023 · Rationalization empowers deep learning models with self-explaining capabilities through a cooperative game, where a generator selects a ...
Dec 15, 2023 · In this paper, we discover that rationalization is prone to a problem named rationale shift, which arises from the algorithmic bias of the.
In this paper, we discover that rationalization is prone to a problem named rationale shift, which arises from the algorithmic bias of the cooperative game.
A novel approach called DAR (Discriminatively Aligned Rationalization), which utilizes an auxiliary module pretrained on the full input to discriminatively ...
Jul 26, 2024 · Various approaches have been proposed for rationalization, spanning from early rationale sampling-based methods (Bao et al., 2018;Bastings et al ...
Sep 23, 2023 · Abstract:Rationalization empowers deep learning models with self-explaining capabilities through a cooperative game, where a generator ...
Aug 11, 2024 · As shown in Figure 1, it aims to select a small and human-intelligible subset (i.e., rationale) from the input to support and explain the ...
Co-authors ; Enhancing the rationale-input alignment for self-explaining rationalization. W Liu, H Wang, J Wang, Z Deng, Y Zhang, C Wang, R Li. 2024 IEEE 40th ...
Sep 27, 2024 · Self-rationalization works by having the model generate multiple judgments with rationales for the same input, curating a preference pair ...
Missing: explaining | Show results with:explaining
Enhancing the Rationale-Input Alignment for Self-explaining Rationalization. ICDE 2024: 2218-2230. [i6]. view. electronic edition via DOI (open access) ...