20.500.12556/RUL-128965 Predznanje za izboljšavo razlag modelov strojnega učenja Background knowledge for better explanation of machine learning models S porastom uporabe kompleksnih modelov strojnega učenja so postale priljubljene post hoc razlagalne metode, kot sta SHAP in LIME. Te metode ustvarijo razlage v obliki prispevkov značilk za določeno napoved. Razlage na nivoju značilk zaradi nejasne povezave s predznanjem niso nujno razumljive človeškim strokovnjakom. V diplomskem delu predstavimo metodo ReEx (Sklepanje iz razlag), ki je združljiva z razlagami metod, ki delujejo na nivoju značilk, in z uporabo predznanja v obliki ontologij posploši razlage po principu najmanj splošne posplošitve. Rezultat so semantične razlage, ki so specifične za posamezne razrede in so potencialno bolj informativne, saj predstavijo ključne značilke v kontekstu predznanja. ReEx smo implementirali v obliki python knjižnice, kompatibilne z metodami, kot sta SHAP in LIME. Za ovrednotenje delovanja metode definiramo mere posploševanja in medsebojne povezanosti razlag. Metodo empirično ovrednotimo na treh tekstovnih podatkovnih množicah. With the wide adoption of complex black-box models, instance-based post hoc explanation tools, such as SHAP and LIME became popular. These tools produce explanations as contributions of features to a given prediction. The obtained explanations at the feature level are not necessarily understandable by human experts because of unclear connections with the background knowledge. We propose ReEx (Reasoning from Explanations), a method applicable to explanations generated by instance-level explainers. By using background knowledge in the form of ontologies, ReEx generalizes instance explanations with the least general generalization principle. The resulting symbolic descriptions are specific for individual classes and offer generalizations based on the explainer's output. The derived semantic explanations are potentially more informative, as they describe the key attributes in the context of background knowledge. ReEx is available as a python library and is compatible with explanation approaches such as SHAP and LIME. For the evaluation of ReEx's performance, we define measures of generalization and overlap of explanations. We conduct experiments on three textual datasets. razložljiva umetna inteligenca razlage modelov posploševanje razlag razlage SHAP strojno učenje explainable AI model explanations explanation generalization SHAP explanations machine learning true false false Slovenski jezik Angleški jezik Diplomsko delo/naloga 2021-08-19 17:40:00 2021-08-19 17:40:06 2022-09-07 04:17:53 0000-00-00 00:00:00 2021 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 30188 78873603 Stepisnik_perdih_Timen_-_Predznanje_za_izboljsavo_razlag_modelov_strojnega_ucenja.pdf Stepisnik_perdih_Timen_-_Predznanje_za_izboljsavo_razlag_modelov_strojnega_ucenja.pdf 1 FD71434C5B1D8805E81BE607413F1636 6535627fa62635f6e91d088a4689c6c00decd66126a3a1b38318423585c90113 8478b21c-0103-11ec-a523-00155dcfd717 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=145877 Fakulteta za računalništvo in informatiko 0 0 0