Rezanje nevronskih mrež z matrično faktorizacijo

diplomsko delo

Teja Roštan (Author), Tomaž Curk (Mentor)

Abstract

Matrično faktorizacijo, ki se povezuje s postopkom zlivanja podatkov, uporabljamo za odkrivanje vzorcev oziroma skupin v podatkih. Faktorizirani model preslika podatke v nižje-dimenzionalen prostor, jih tako skrči in odpravi del šuma. Tovrstni modeli so zato navadno bolj robustni in imajo višjo napovedno točnost. Pri nevronskih mrežah bi tako znali reševati problem prevelike prilagojenosti podatkom (angl. overfitting) in pridobili pri generalizaciji. V nalogi smo preučili, ali s hkratno faktorizacijo parametrov nevronske mreže, ki jih je možno predstaviti z več matrikami, odstranimo (porežemo) nepomembne povezave in tako izboljšamo napovedno točnost mreže. Predlagani postopek rezanja smo preizkusili na navadnih in globokih nevronskih mrežah. Po uspešnosti je primerljiv z ostalimi najuspešnejšimi standardnimi pristopi rezanja nevronskih mrež.

Keywords

nevronske mreže;matrična faktorizacija;rezanje;računalništvo;računalništvo in informatika;univerzitetni študij;diplomske naloge;

Data

Language:	Slovenian
Year of publishing:	2015
Typology:	2.11 - Undergraduate Thesis
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[T. Roštan]
UDC:	004.85(043.2)
COBISS:	1536482243
Views:	1010
Downloads:	162
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Pruning neural network using matrix factorization
Secondary abstract:	Matrix factorization and the procedure of data fusion are used to detect patterns in data. The factorized model maps the data to a low-dimensional space, therefore shrinking it and partially eliminating noise. Factorized models are thus more robust and have a higher predictive accuracy. With this procedure we could solve the problem of overfitting in neural networks and improve their ability to generalize. Here, we report on how to simultaneously factorize the parameters of a neural network, which can be represented with multiple matrices, to prune not important connections and therefore improve predictive accuracy. We report on empirical results of pruning normal and deep neural networks. The proposed method performs similarly to the best standard approaches to pruning neural networks.
Secondary keywords:	neural networks;matrix factorization;pruning;computer science;computer and information science;diploma;
File type:	application/pdf
Type (COBISS):	Bachelor thesis/paper
Study programme:	1000468
Embargo end date (OpenAIRE):	1970-01-01
Thesis comment:	Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages:	52 str.
ID:	8900524