Otkrivanje izvandistribucijskih podataka analizom izlaza dubokog modela

undergraduate thesis

• Title (english): Discovery of Out-of-Distribution Data by Analyzing Predictions of a Deep Model
• Author: Bruno Kovač
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Marin Oršić (član povjerenstva)
• Committee member: Ivan Krešo (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-13, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Konvolucijske neuronske mreže danas su često korištene za klasifikacijsku ulaznih podataka u pojedine moguće razrede za koje je model prethodno istreniran. Neovisno o testnom podatku, model će uvijek s određenom sigurnošću predvidjeti pripradnost ulaza u jedan od klasifikacijskih razreda. Ključna problematika ovog pristupa je nemogućnost otkrivanja ulaznih podataka za koje će model najvjerojatnije pogriješiti. Upravo zbog toga potrebno je otkriti izvandistribucijske podatke dovedene na ulaz... More modela, a to činimo analizom njegovog izlaza. Kod takvih podataka nikad ne možemo sigurno tvrditi jesu li dijelom distribucije na kojoj je model treniran ili su vandistribucijski, no vjerojatnost detekcije izvandistribucijskih podataka moguće je odrediti na više načina, npr. korištenjem AUROC i AUPR površina. Less

Abstract (english)

Convolutional neural networks are nowadays often used for classifying input data into specific possible classes for which the model was previously trained. Whatever test data we provide to the model, it will always predict the result corresponding class with specific certainty. The key problem with this approach is an inability of determining the data for which the model will most likely make a wrong prediction. Specifically because of that, it is a necessity to determine... More out-of-distribution data provided as input to the model, and the way to do that is by analyzing its output. Despite the fact that we cannot claim if the test data belongs to the distribution on which the model was trained or it is out-of-distribution, the possibility of detecting out-of-distribution data can be determined with various methods, including using AUROC and AUPR areas. Less

• Keywords: duboki model; konvolucijske neuronske mreže; otkrivanje izvandistribucijskih podataka; AUROC; AUPR; skup podataka; treniranje; testiranje; validacija; CIFAR-10; TensorFlow
• Keywords (english): deep model; convolutional neural networks; discovery of out-of-distribution data; AUROC; AUPR; dataset; training; testing; validation; CIFAR-10; TensorFlow
• Language: croatian
• URN:NBN: urn:nbn:hr:168:925571
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-13 21:47:02