Syväoppimisen laskennallinen vaativuus
Syväoppiminen on maailmanlaajuisesti käytössä oleva teknologia, jota hyödynnetään yhä etenevässä määrin eri aloilla. Tässä kandidaatintutkielmassa selvitetään mikä on syväoppimisen laskennallinen vaativuus. Tutkielmassa avataan syväoppimisen käsitteistöä sekä laskennallisen vaativuuden teoriaa. Syväoppiminen on koneoppimisen alalaji, jossa jäljitellään ihmisaivojen neuronien toimintaperiaatteita. Tutkielma antaa pohjaa neuroverk- kojen optimoinnin tutkimuksiin. Lähdekirjallisuus on kerätty pääosin tuoreista alan kunnioitetuista julkaisuista ja tutkielma on toteutettu kirjallisuuskatsauksena. Tutkielmassa on esitetty yksi mahdollinen esitys konvoluutionaalisen neuroverkon laskennalliselle vaativuudelle. Muisti- sekä aikavaativuus konvoluutionaaliselle neuroverkolle on esitetty käyttäen "iso O-notaatiota". Aikavaativuudelle löydettiin yksi notaatio, mutta muistivaativuus on kahdelle eri kerrokselle eli konvoluutio- sekä lajittelukerrokselle. Deep learning is a technology which is increasingly being used in different sectors worldwide. In this bachelor’s thesis the subject is to find out what the computational complexity for deep learning is. The thesis discusses the concepts of deep learning and so- me theory of computational complexity. Deep learning is a subset of machine learning that exploits the operating principles of neurons in the human brain. This thesis provides a basis for research into the optimization of neural networks. The references for this thesis has been collected mainly form recent reputable publications of the field and the thesis has been conducted as a literature review. The thesis presents one possible representation of the computational complexity of a convolutional neural network (CNN). The time and space complexity for a CNN is represented using "Big O notation". Complexities for CNN were presented using multiple notations. Time complexity was presented using only one notation, but space complexity has two diffrent notations, one for convolutional layer and the other for fully connected layer.
...
Metadata
Show full item recordCollections
- Kandidaatintutkielmat [5362]
License
Related items
Showing items with similar title or keywords.
-
Mathematical Fuzzy Logic in the Emerging Fields of Engineering, Finance, and Computer Sciences
Shukla, Amit K. (MDPI, 2022) -
LightSleepNet : A Lightweight Deep Model for Rapid Sleep Stage Classification with Spectrograms
Zhou, Dongdong; Xu, Qi; Wang, Jian; Zhang, Jiacheng; Hu, Guoqiang; Kettunen, Lauri; Chang, Zheng; Cong, Fengyu (IEEE, 2021)Deep learning has achieved unprecedented success in sleep stage classification tasks, which starts to pave the way for potential real-world applications. However, due to its enormous size, deployment of deep neural networks ... -
Reducing redundancy in the bottleneck representation of autoencoders
Laakom, Firas; Raitoharju, Jenni; Iosifidis, Alexandros; Gabbouj, Moncef (Elsevier, 2024)Autoencoders (AEs) are a type of unsupervised neural networks, which can be used to solve various tasks, e.g., dimensionality reduction, image compression, and image denoising. An AE has two goals: (i) compress the original ... -
Robustness, Stability, and Fidelity of Explanations for a Deep Skin Cancer Classification Model
Saarela, Mirka; Geogieva, Lilia (MDPI AG, 2022)Skin cancer is one of the most prevalent of all cancers. Because of its being widespread and externally observable, there is a potential that machine learning models integrated into artificial intelligence systems will ... -
Causality-Aware Convolutional Neural Networks for Advanced Image Classification and Generation
Terziyan, Vagan; Vitko, Oleksandra (Elsevier, 2023)Smart manufacturing uses emerging deep learning models, and particularly Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), for different industrial diagnostics tasks, e.g., classification, ...