Politehnika i dizajn, Vol. 12 No. 1, 2024.
Pregledni rad
https://doi.org/10.19279/TVZ.PD.2024-12-1-06
CONCEPTUAL DISCUSSION OF EXPLAINABILITY OF SUPERVISED FEATURE LEARNING FOR CLASSIFICATION
Dino Vlahek
; UM FERI, Koroška cesta 46, 2000 Maribor, Slovenija
*
Bojan Nožica
; Tehničko veleučilište u Zagrebu, Vrbik 8, Zagreb, Hrvatska
*
* Dopisni autor.
Sažetak
This paper presents the basic ideas of supervised feature learning for classification. Special attention is given to the explainability of these approaches. Feature learning methods are either inexplicable or limited in their prediction results due to the inability to recombine input features. Approaches that increase the dimensionality of the input feature space are slow because they require iterative non-convex optimizations and tuning of numerous configurations of hidden dimensions. In these cases, authors generally do not provide explanation of the learned model. However, explanations can be achieved in various degrees of success by learning interpretive models around a given pattern of interest or by evaluating the importance of each feature in the classification output.
Ključne riječi
explainable artificial intelligence; classification; feature learning; knowledge discovery
Hrčak ID:
326434
URI
Datum izdavanja:
15.3.2024.
Posjeta: 0 *