magistrsko delo
Jernej Janež (Author), Matjaž Kukar (Mentor)

Abstract

Za uspešno gradnjo modelov metode strojnega učenje potrebujejo čim več označenih primerov, s katerimi želimo modele naučiti v takšni meri, da bi nove neoznačene primere s čim večjo zanesljivostjo razvrščali v pravilne razrede. Te običajno pridobimo od strokovnjaka področja, ki najprej podatke zbere, jih ustrezno označi in jih nato posreduje nam. V magistrski nalogi z omenjenimi metodami ter na podlagi označenih podatkov naslavljamo problem pridobivanja in upoštevanja verodostojne in kvalitetne povratne informacije o naših napovedih. Za varno prenašanje podatkov opišemo nov način uporabe tehnologije veriženja blokov kot vmesni sloj – nespremenljivo referenco med uporabniki, s katero so uporabniki med seboj sinhronizirani ter lahko kadarkoli preverijo, ali so informacije, shranjene na verigi, še zmeraj enake in pravilne. Klasičnim metodam strojnega učenja dodamo koraka pridobivanja povratne informacije in posodabljanja modelov ter predstavimo izzive in mogoče rešitve motiviranja uporabnikov za posredovanje iskrene in kvalitetne povratne informacije. Na podlagi zanesljivosti naših napovedi predlagamo način ocenjevanja uporabnikov in njihovih odzivov. Na medicinskih podatkih nato zgradimo klasifikator ter na podlagi izračunane ocene zdravnikov, predstavimo tri načine bogatenja učne množice in posodabljanja klasifikatorja, od katerih se dva izkažeta kot uspešna.

Keywords

tehnologija veriženja blokov;strojno učenje;povratna informacija;motivacija;računalništvo;računalništvo in informatika;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [J. Janež]
UDC: 004.85(043.2)
COBISS: 45349891 Link will open in a new window
Views: 738
Downloads: 146
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Collecting credible feedback with machine learning and blockchain technologies
Secondary abstract: In order to successfully build machine learning models, we need as many annotated examples as possible such that new unlabeled examples can be classified into correct classes with the greatest possible reliability. These can be obtained from an expert in the field, who first collects the data, marks it accordingly and then sends it to us. On the basis of the annotated data and machine learning methods, we predict classes for new examples and address the problem of obtaining credible and quality feedback for our predictions. For secure data transfer, we describe a new way of using blockchain technologies as middleware for immutable reference between users to ensure that all users are synchronized and can check at any time that the information stored on the blockchain is still correct. We add steps of obtaining feedback and updating models to classical machine learning methods, and present challenges and possible solutions to motivate users to provide honest and quality feedback. Based on the reliability of our predictions, we suggest a way of assessing users and their responses. We build a classifier on medical data and, based on the calculated assessment of doctors, present three ways of enriching the learning examples and updating the classifier, two of which prove to be successful.
Secondary keywords: blockchain technologies;machine learning;user feedback;motivation;computer science;computer and information science;master's degree;
Type (COBISS): Master's thesis/paper
Study programme: 1000471
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 84 str.
ID: 12414364