KNOWLEDGE SPACE DRIVEN TRAINING OF DEEP NEURAL NETWORKS
DOI:
https://doi.org/10.24867/24BE18HlozanKeywords:
Knowledge Space Theory, Curriculum Learning, Machine Learning, Deep Learning, Artificial Neural NetworksAbstract
This paper describes the classification problem in Deep learning, how it is solved today, and gives an idea how to improve the solution to this kind of problem. One of the most popular algorithms for the optimization of Artificial neural networks, Stochastic gradient descent, is described. An idea was described and implemented so that this optimization algorithm could be improved using the Knowledge Space Theory and the Curriculum learning based on the defined knowledge space.
References
[1] Jean-Claude Falmagne, Jean-Paul Doignon: “Learning Spaces”, 2011.
[2] Christina Stahl, Cord Hockemeyer: “Knowledge Space Theory”, 2022.
[3] Yoshua Bengio, Ian Goodfellow, Aaron Courville: “Deep learning”, 2015.
[4] Yoshua Bengio, Jerome Louradour, Ronan Collobert, Jason Weston: “Curriculum learning”, 2009.
[5] Milan Segedinac: https://github.com/milansegedinac/kst , 2023.
[2] Christina Stahl, Cord Hockemeyer: “Knowledge Space Theory”, 2022.
[3] Yoshua Bengio, Ian Goodfellow, Aaron Courville: “Deep learning”, 2015.
[4] Yoshua Bengio, Jerome Louradour, Ronan Collobert, Jason Weston: “Curriculum learning”, 2009.
[5] Milan Segedinac: https://github.com/milansegedinac/kst , 2023.
Downloads
Published
2023-09-06
Issue
Section
Electrotechnical and Computer Engineering
