Machine learning and essentialism (CROSBI ID 713904)
Prilog sa skupa u zborniku | sažetak izlaganja sa skupa | međunarodna recenzija
Podaci o odgovornosti
Šekrst, Kristina ; Skansi, Sandro
engleski
Machine learning and essentialism
The goal of this paper is to establish the metaphysical essentialist position both for machine learning and deep learning, a stance not often researched in philosophy and computing (cf. rare examples like Pelillo and Scantamburlo 2013, Scantamburlo 2014, Duin 2015). In machine learning, properties used for supervised learning and dataset tagging will be compared to minimal and maximal essentialism in metaphysics, but also with componential and semantic analysis in linguistics and psychological and linguistic theory of prototypes. To better understand different kinds of essentialism in computing, we will differentiate between supervised learning (using labeled data with pre-defined features of interest) and unsupervised learning or finding relevant groups or clusters without predefined properties (cf. Henning 2015). In various applications of machine learning, different features can be seen as essential properties. For example, in handwriting recognition, we can talk about various pixels and recorded movements, while in computer vision and pattern recognition we can talk about higher-level objects such as blobs, where regions in a digital image that differ in properties are compared to the surrounding regions. Feature engineering in computer science refers to the process of extracting features (i.e. essential properties) from raw data. In this case, we can see that either a computer can serve as a certain oracle regarding what properties are considered essential for machine learning to take place, or a human being can choose a predefined set of such features or Humean bundles. We will connect different feature engineering strategies to different metaphysical essentialist theories, in order to see how metaphysics can be reflected in machine-learning applications. Such strategies will serve the purpose of establishing that a philosophical approach in machine learning is not a matter of rejecting or supporting essentialism but choosing a specific essentialist stance for a specific application. Namely, in the background, we are dealing with ontologically different phenomena that require different metaphysical and computing analyses.
machine learning, essentialism, pattern recognition, similarity recognition
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
Podaci o prilogu
39-40.
2021.
objavljeno
Podaci o matičnoj publikaciji
Podaci o skupu
Philosophy in Informatics VI: Frontiers of philosophy of computing and information
predavanje
16.12.2021-17.12.2021
Kraków, Poljska; online
Povezanost rada
Filozofija, Interdisciplinarne humanističke znanosti, Interdisciplinarne tehničke znanosti, Računarstvo