… PIQ, a post-hoc neural network interpretation method that utilizes vector quantization to learn
class… We show that PIQ quantitatively outperforms other interpretation methods on black-and…

… In the experiment, we demonstrate the effectiveness of our method on benchmark datasets
from the perspectives of posthoc explanation and semantic explanation respectively. Our …

… Saliency maps are a widely used tool for interpreting DNNs by highlighting the input features
… Unlike post-hoc techniques, which generate explanations after the model has been trained, …

… that are more tractable for analysis and interpretation. Mapping to CCS … using bootstrap,
was approximately 0.003. Non-linearly transforming our count features using quantization …

… To model the post-hoc analysis quality, we first determine the error distribution from the
compressor quantization step. Then, we analyze the impact on post-hoc analysis quality by a …

… within the error bound against the original, even without quantization. As mentioned in the
previous theoretical analysis for post-hoc analysis, we can easily adopt our models based on …

… performs significantly better when the assumption on crosscorrelation is strong in both post-hoc
analysis and end-toend training settings. When trained end-to-end, DQ reduces the cross…

… , have proven effective for interpreting deep learning models. … , an efficient post-hoc
explanation method incorporating LM-… ) was developed via canonical quantization by treating the …

… If significant, we applied the Nemenyi post-hoc test for pairwise comparisons to pinpoint
significant performance differences, using a critical distance (CD) to define significant rank …

… survival analysis … using quantization improved the accuracy of logistic regression to nearly
match that of XGBoost on this dataset as measured by AUROC. Hence, we used quantization …