This looks More Like that: Enhancing Self-Explaining Models by Prototypical Relevance Propagation
نویسندگان
چکیده
Current machine learning models have shown high efficiency in solving a wide variety of real-world problems. However, their black box character poses major challenge for the comprehensibility and traceability underlying decision-making strategies. As remedy, numerous post-hoc self-explanation methods been developed to interpret models’ behavior. Those methods, addition, enable identification artifacts that, inherent training data, can be erroneously learned by model as class-relevant features. In this work, we provide detailed case study representative state-of-the-art self-explaining network, ProtoPNet, presence spectrum artifacts. Accordingly, identify main drawbacks especially its coarse spatially imprecise explanations. We address these limitations introducing Prototypical Relevance Propagation (PRP), novel method generating more precise model-aware Furthermore, order obtain clean, artifact-free dataset, propose use multi-view clustering strategies segregating artifact images using PRP explanations, thereby suppressing potential models.
منابع مشابه
Enhancing Relevance Models with Adaptive Passage Retrieval
Passage retrieval and pseudo relevance feedback/query expansion have been reported as two effective means for improving document retrieval in literature. Relevance models, while improving retrieval in most cases, hurts performance on some heterogeneous collections. Previous research has shown that combining passage-level evidence with pseudo relevance feedback brings added benefits. In this pap...
متن کاملLooks like multiple sclero
Positive ANAs are found in ordinary people A fairly standard definition of an abnormal ANA titre is 1 : 40, which has a prevalence in normal people of between 5 and 35%, usually higher in women. Of course, the prevalence of antibody positivity changes as the cursor shifts: 32% of normal adults were positive at 1 : 40 in one study, 13% at 1 : 80, 5% at 1 : 160 and only 3% at 1 : 320 (Craig et al...
متن کاملDevelopment: Facial Makeup Enhancing Our Looks
A recent study in mice deciphers the complex genetic regulatory network underlying the morphogenesis of the face. The enhancer landscape underlying craniofacial development provides multiple entry points to understand what makes up the face, in natural variation or pathological conditions.
متن کاملExplaining Flow-Based Propagation
Lazy clause generation is a powerful approach to reducing search in constraint programming. For use in a lazy clause generation solver, global constraints must be extended to explain themselves. In this paper we present two new generic flow-based propagators (for hard and soft flow-based constraints) with several novel features, and most importantly, the addition of explanation capability. We d...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Pattern Recognition
سال: 2023
ISSN: ['1873-5142', '0031-3203']
DOI: https://doi.org/10.1016/j.patcog.2022.109172