Semilinear Predictability Minimization Produces Well-Known Feature Detectors

Jürgen Schmidhuber, Martin Eldracher, Bernhard Foltin

Research output: Contribution to journalArticlepeer-review

34 Scopus citations


Predictability minimization (PM - Schmidhuber 1992) exhibits various I intuitive and theoretical advantages over many other methods for unsupervised redundancy reduction. So far, however, there have not been any serious practical applications of PM. In this paper, we apply semilinear PM to static real world images and find that without a teacher and without any significant preprocessing, the system automatically learns to generate distributed representations based on well-known feature detectors, such as orientation-sensitive edge detectors and off-center-on-surround detectors, thus extracting simple features related to those considered useful for image preprocessing and compression.
Original languageEnglish (US)
Pages (from-to)773-786
Number of pages14
JournalNeural Computation
Issue number4
StatePublished - May 15 1996
Externally publishedYes

ASJC Scopus subject areas

  • Cognitive Neuroscience


Dive into the research topics of 'Semilinear Predictability Minimization Produces Well-Known Feature Detectors'. Together they form a unique fingerprint.

Cite this