Kristína Malinovská

Kristína Malinovská
Czech Technical University in Prague | ČVUT · Czech Institute of Informatics, Robotics and Cybernetics

PhD.

About

19
Publications
3,223
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
47
Citations
Additional affiliations
January 2020 - May 2020
Comenius University Bratislava
Position
  • Researcher
Education
November 2011 - November 2012
Comenius University Bratislava
Field of study
  • Informatics
September 2009 - March 2014
Comenius University Bratislava
Field of study
  • Informatics
September 2007 - June 2009
Comenius University Bratislava
Field of study
  • Cognitive Science

Network

Cited By

Projects

Project (1)
Project
A novel learning algorithm for artificial neural networks proposed as a more biologically plausible alternative to error backpropagation. A bidirectional heteroassociative network where patterns presented to both visible layers simultaneously serve as targets to each other. Along with learning from the teaching signal the network also learns from bounced-back predictions (echoes). In each direction of activation propagation there is a separate weight matrix, therefore UBAL does not suffer from the weight transport problem. The UBAL learning rule is local, so only presynaptic and postsynaptic activations in different activation phases contribute to updating the weights. With different setups of learning rule hyperparameters our model can learn various qualitatively different tasks, including encoding, denoising, and classification. The experimental results confirm that UBAL successfully solves the given problems and that its performance is comparable with the related models.