Wyniki wyszukiwania - BazTech

1

Interpretability versus explainability: classification for understanding deep learning systems and models

Namatēvs Ivars, Sudars Kaspars, Dobrājs Artis

Computer Assisted Methods in Engineering and Science

|

2022

|

Vol. 29, no. 4

297--356

EN

The techniques of explainability and interpretability are not alternatives for many realworld problems, as recent studies often suggest. Interpretable machine learning is nota subset of explainable artificial intelligence or vice versa. While the former aims to build glass-box predictive models, the latter seeks to understand a black box using an explanatory model, a surrogate model, an attribution approach, relevance importance, or other statistics. There is concern that definitions, approaches, and methods do not match, leading to the inconsistent classification of deep learning systems and models for interpretation and explanation. In this paper, we attempt to systematically evaluate and classify the various basic methods of interpretability and explainability used in the field of deep learning.One goal of this paper is to provide specific definitions for interpretability and explainability in Deep Learning. Another goal is to spell out the various research methods for interpretability and explainability through the lens of the literature to create a systematic classifier for interpretability and explainability in deep learning. We present a classifier that summarizes the basic techniques and methods of explainability and interpretability models. The evaluation of the classifier provides insights into the challenges of developinga complete and unified deep learning framework for interpretability and explainability concepts, approaches, and techniques.

2

On ordered minimal structures

Jagiella G., Newelski L.

Reports on Mathematical Logic

|

2014

|

Vol. 49

35--46

EN

We investigate minimal first-order structures and consider interpretability and definability of orderings on them. We also prove the minimality of their canonical substructures.

3

The maintenance costs estimation of electrical lines with the use of interpretability-oriented genetic fuzzy rule-based systems

Rudziński F., Piekoszowski J.

Przegląd Elektrotechniczny

|

2013

|

R. 89, nr 8

43--47

EN

The paper demonstrates an evolutionary technique to design an interpretability-oriented fuzzy rule-based system and its application to estimate the maintenance costs of medium voltage electrical lines. The main goal of the proposed technique is to design the system with not only a relatively high accuracy for estimating the costs, but also with a clear and transparent structure which is easy to interpret by humans. The structure includes easily readable and understandable fuzzy logic rules that represent the knowledge about the considered problem.

PL

Praca demonstruje ewolucyjną technikę projektowania przejrzystych systemów regułowo-rozmytych i jej zastosowanie do szacowania kosztów utrzymania linii energetycznej średniego napięcia. Zaprojektowany system posiada względnie wysoką dokładność i przejrzystą strukturę w formie czytelnych, zrozumiałych i łatwych do interpretacji przez człowieka reguł logicznych, będących kwintesencją wiedzy o rozważanym problemie.

4

New frontiers of analysis, interpretation and classification of biomedical signals: a computational intelligence framework

Gacek A.

Journal of Medical Informatics & Technologies

|

2011

|

Vol. 17

23--36

EN

The methods of Computational Intelligence (CI) including a framework of Granular Computing, open promising research avenues in the realm of processing, analysis and interpretation of biomedical signals. Similarly, they augment the existing plethora of "classic" techniques of signal processing. CI comes as a highly synergistic environment in which learning abilities, knowledge representation, and global optimization mechanisms and this essential feature is of paramount interest when processing biomedical signals. We discuss the main technologies of Computational Intelligence (namely, neural networks, fuzzy sets, and evolutionary optimization), identify their focal points and elaborate on possible limitations, and stress an overall synergistic character, which ultimately gives rise to the highly symbiotic CI environment. The direct impact of the CI technology on ECG signal processing and classification is studied with a discussion on the main directions present in the literature. The design of information granules is elaborated on; their design realized on a basis of numeric data as well as pieces of domain knowledge is considered. Examples of the CI-based ECG signal processing problems are presented. We show how the concepts and algorithms of CI augment the existing classification methods used so far in the domain of ECG signal processing. A detailed construction of granular prototypes of ECG signals being more in rapport with the diversity of signals analyzed is discussed as well. ECG signals, Computational Intelligence, neurocomputing, fuzzy sets, information granules, Granular Computing, interpretation, classification, interpretability.

5

New Methods for Designing and Reduction of Neuro-Fuzzy Systems

Cpałka K.

Journal of Applied Computer Science Methods

|

2010

|

Vol. 2 No. 2

113--126

EN

In the paper, we propose novel methods for designing and reduction of neuro-fuzzy systems without the deterioration of their accuracy. The reduction and merging algorithms gradually eliminate inputs, rules, antecedents, and the number of discretization points of integrals in the center of area defuzzification method. Our algorithms have been tested using well known classification benchmark.

6

Ocena i wstępna analiza 8-kanałowego obrazu bardzo wysokiej rozdzielczości z satelity WorldView-2

Wyczałek I., Królewicz S., Wyczałek E.

Archiwum Fotogrametrii, Kartografii i Teledetekcji

|

2010

|

Vol. 21

471--480

PL

Od początku roku 2010 osiągalne są wysokorozdzielcze obrazy satelitarne z nowego satelity – WorldView-2. Charakteryzują się one rejestracją powierzchni Ziemi z rozdzielczością geometryczną 2 × 2 m w 8-kanałach spektralnych oraz 0,5 × 0,5 m w zakresie panchromatycznym. Nowe kanały wprowadzono przede wszystkim w celu znacznego zwiększenia zakresu i zdolności interpretacyjnej obrazu. Dzięki połączeniu dużej rozdzielczości przestrzennej z wysoką precyzją pozycjonowania powstał nowy standard danych teledetekcyjnych o zdecydowanie większym potencjale aplikacyjnym niż to było dotychczas możliwe. Możliwość codziennego pozyskiwania obrazów o 1-metrowej rozdzielczości stwarza korzystne warunki do śledzenia dynamiki procesów i zjawisk zachodzących na powierzchni Ziemi lub w wodach przybrzeżnych. Wysokorozdzielcze obrazy satelitarne tej klasy są szczególnie przydatne w aktualizacji baz danych topograficznych oraz innych opracowaniach odpowiadających mapom w skali co najmniej 1:5000, co czyni je przydatnymi do wielu analiz dla potrzeb zarządzania, monitorowania i planowania przestrzennego. Dociekliwość potencjalnego użytkownika obrazu wymaga sprawdzenia wiarygodności tych informacji oraz potencjalnych innych cech tej wyjątkowej klasy danych teledetekcyjnych. Wyniki takiej wstępnej analizy stanowią przedmiot niniejszego artykułu.

EN

Since the beginning of 2010 a high resolution satellite imagery from the new satellite – WorldView-2 are attainable. It is characterized by recording of the Earth's surface with geometric resolution 2 × 2 m in 8 spectral channels and 0.5 × 0.5 m in panchromatic channel. The new channels were introduced primarily to significantly increase the scope and capacity of interpretation of the image. Thanks to the combination of high spatial resolution with high precision positioning, results a new standard for remote sensing data with a much greater potential than was previously possible. The possibility of daily obtaining images of 1-meter resolution creates favorable conditions for tracking the dynamics of the processes and phenomena occurring on the land surface or in coastal waters. High resolution satellite images of this class are particularly useful for updating databases and other studies of topographic maps on a scale equivalent to at least 1:5k, what makes them useful for many studies for management, monitoring and planning. Inquisitiveness of potential user requires verification of this information and potentially of other features of this unique class of remote sensing data. The results of this preliminary analysis are the subject of this paper.