Geodesics of learned representations

Olivier J. Hénaff, Eero P. Simoncelli

Research output: Contribution to conferencePaperpeer-review


We develop a new method for visualizing and refining the invariances of learned representations. Specifically, we test for a general form of invariance, linearization, in which the action of a transformation is confined to a low-dimensional subspace. Given two reference images (typically, differing by some transformation), we synthesize a sequence of images lying on a path between them that is of minimal length in the space of the representation (a “representational geodesic”). If the transformation relating the two reference images is linearized by the representation, this sequence should follow the gradual evolution of this transformation. We use this method to assess the invariance properties of a state-of-the-art image classification network and find that geodesics generated for image pairs differing by translation, rotation, and dilation do not evolve according to their associated transformations. Our method also suggests a remedy for these failures, and following this prescription, we show that the modified representation is able to linearize a variety of geometric image transformations.

Original languageEnglish (US)
StatePublished - Jan 1 2016
Event4th International Conference on Learning Representations, ICLR 2016 - San Juan, Puerto Rico
Duration: May 2 2016May 4 2016


Conference4th International Conference on Learning Representations, ICLR 2016
Country/TerritoryPuerto Rico
CitySan Juan

ASJC Scopus subject areas

  • Education
  • Computer Science Applications
  • Linguistics and Language
  • Language and Linguistics


Dive into the research topics of 'Geodesics of learned representations'. Together they form a unique fingerprint.

Cite this