An Analysis on Disentanglement in Machine Learning Makine Öǧrenmesinde Ayrişiklik Üzerine Bir Analiz

Date

2022-01-01

Author

Moğultay, Hazal
Kalkan, Sinan
Yarman Vural, Fatoş Tunay

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

75
views

0
downloads

Learnt representations by Deep autoencoders is not capable of decomposing the complex information into simple notion. In other words, attributes of samples are entangled in the basis vectors spanning the learned space. This leads to significant errors in deep learning algorithms. In order to avoid these errors, it is necessary to separate the feature space according to the common features shared between classes and to define a simple subspace for each feature. This approach has led to the birth of a new paradigm in Machine Learning, called disentanglement.Roughly, disentangled models can be defined as models that can independently learn the different components of the probability density function that produces the dataset in the feature space. Unfortunately, it is not always possible to learn these models. For this reason, there is still no easily applicable mathematical definition of disentanglement in the literature. In this study, a mathematical definition of the concept of disentanglement will be made and methods and metrics related to this approach will be discussed.

Subject Keywords

Disentanglement, representation learning, variational autoencoder

URI

https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85138708729&origin=inward
https://hdl.handle.net/11511/101663

DOI

https://doi.org/10.1109/siu55565.2022.9864743

Conference Name

30th Signal Processing and Communications Applications Conference, SIU 2022

Collections

Department of Computer Engineering, Conference / Seminar

Suggestions

OpenMETU
Core

Mathematical ontology the question of the mathematical source of objectivity Dorrikhteh, Omid; Grünberg, David; Department of Philosophy (2019) This thesis traces the source of mathematical objectivity, as an approach to justify mathematical properties to be real, through how our mind and language were evolved. In the mirror of the indispensability argument, and the unreasonable effectiveness of mathematics, it will be argued that the reason why the world and the mind exhibit ontologically similar structures (and properties) is because they have the same ontological origin. Accordingly, it will be shown that (1) why/how that “the world and the mind...
Edmund Burke’de Yüce ve Jouissance, Lacancı bir okuma Birlik, Nurten (Akdeniz University; 2018-05-18) Burke’çü anlamda yüce, şiir kişisinin, doğada gördüğü tekil nesnelere odaklanıp bu nesnelerin arkasında yatan bütüncül metafizik yapının izlerini görebilme ya da doğada görülen nesneler aracılığıyla aşkınlığı deneyimleme çabası olarak tanımlanabilir. Bu deneyim aynı zamanda modernitede, ampirizm ve pozitivizm ile birlikte ortaya çıkan insan ve doğa kategorilerini epistemolojik bağlamda sorunsallaştırır çünkü ikisi arasında oluşmuş kopukluğu gidermeye hizmet eder. Bu bildiri odak noktası ...
Husserl's conception of intentionality phenomenological analysis of noesis and noema Gözetlik, Servet; İnam, Ahmet; Department of Philosophy (2003) Husserl̕s phenomenology can be analyzed simply by relying on the conception of intentionality. What I want to do is to put forward the logical grounds on which I can construct an acceptable account of Husserl̕s theory of intentionality. For this aim, firstly, I need to put some light on the nature of intentional acts or experiences.This suggests us that there is a close connection between the acts and what they are directed towards. Actually many have specified the relation between the act and the object, b...
A structural equation modeling study : the metacognition-knowledge model for geometry Aydın, Utkun; Ubuz, Behiye; Department of Secondary Science and Mathematics Education (2007) The purpose of this study is twofold: (1) to examine the effects of knowledge on cognition and regulation of cognition on declarative knowledge, conditional knowledge, and procedural knowledge in geometry and (2) to examine the interrelationships among declarative knowledge, conditional knowledge, and procedural knowledge in geometry. The reciprocal relationships between metacognitive and knowledge factors were modeled by using data from tenth grade secondary school students. Structural equation modeling wa...
Demonstration of quantum contextuality via Hardy Paradox Yolsever, Yankı; Turgut, Sadi; Department of Physics (2022-9) Quantum contextuality is one of the most fundamental foundations of Quantum Mechanics. In this thesis, we have examined the Hardy-type proof of quantum contextuality which is first developed by Lucien Hardy for quantum non-locality and later adapted to Quantum contextuality by Adan Cabello and his colleagues to 3-dimensional systems with a minimum of 5 observables. We have reformulated their method in terms of logical implications and adapted it to 4-dimensional and 5-dimensional systems with using fewer ob...

Citation Formats

H. Moğultay, S. Kalkan, and F. T. Yarman Vural, “An Analysis on Disentanglement in Machine Learning Makine Öǧrenmesinde Ayrişiklik Üzerine Bir Analiz,” presented at the 30th Signal Processing and Communications Applications Conference, SIU 2022, Safranbolu, Türkiye, 2022, Accessed: 00, 2023. [Online]. Available: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85138708729&origin=inward.