Learning Transformation Invariance for Object Recognition

Jens Hocke, Thomas Martinetz


Based on Tomaso Poggio’sM-theory, we propose a method to learn transformation invariant representations. Using an artificial dataset, we demonstrate that our supervised method learns invariance to shifts, and on the MNIST data we show first results for learning the unknown transformations underlying handwritten digits.
Original languageEnglish
Title of host publicationWorkshop New Challenges in Neural Computation 2014
EditorsBarbara Hammer, Thomas Villmann
Number of pages6
Publication date02.09.2014
Publication statusPublished - 02.09.2014
Event36th German Conference on Pattern Recognition - Münster, Germany
Duration: 02.09.201405.09.2014


Dive into the research topics of 'Learning Transformation Invariance for Object Recognition'. Together they form a unique fingerprint.

Cite this