Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Learning optimized features for hierarchical models of invariant object recognition.

Heiko Wersing1, Edgar Körner

  • 1HONDA Research Institute Europe GmbH, 63073 Offenbach/Main, Germany. heiko.wersing@honda-ri.de

Neural Computation
|June 21, 2003
PubMed
Summary

This study introduces a new feedforward neural network model for object recognition. It uses sparse coding to learn features, achieving high performance even in cluttered scenes without precise object segmentation.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Toward Self-Referential Autonomous Learning of Object and Situation Models.

Cognitive computation·2016
Same author

Structural synaptic plasticity has high memory capacity and can explain graded amnesia, catastrophic forgetting, and the spacing effect.

PloS one·2014
Same author

Does spike-timing-dependent synaptic plasticity couple or decouple neurons firing in synchrony?

Frontiers in computational neuroscience·2012
Same author

A life-long learning vector quantization approach for interactive learning of multiple categories.

Neural networks : the official journal of the International Neural Network Society·2012
Same author

Cortext: a columnar model of bottom-up and top-down processing in the neocortex.

Neural networks : the official journal of the International Neural Network Society·2009
Same author

A model for learning topographically organized parts-based representations of objects in visual cortex: topographic nonnegative matrix factorization.

Neural computation·2009

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Machine Learning

Background:

  • Hierarchical neural feedforward architectures are debated for real-world invariant object recognition.
  • Supervised and unsupervised learning methods for these models remain an active research area.

Purpose of the Study:

  • To propose a novel feedforward model for recognition using optimized intermediate feature learning.
  • To investigate the application of sparse coding principles to intermediate feature extraction in hierarchical networks.

Main Methods:

  • Developed a feedforward model incorporating weight sharing, pooling, and competitive nonlinearities.
  • Applied sparse coding principles to learn optimized intermediate complex features under convolutional architecture constraints.
  • Enforced symmetry constraints like translation invariance for efficient feature optimization and dimension reduction.

Related Experiment Videos

  • Analyzed recognition performance on object and face databases, comparing different pooling nonlinearities.

    • Main Results:

    • The proposed model demonstrates effective learning of intermediate complex features using sparse coding.
    • Hierarchical networks with features learned on object data achieved competitive performance on face recognition without parameter changes.
    • A combination of strong competitive nonlinearities and sparse coding yielded the best recognition performance in cluttered environments.
    • Precise object segmentation was found to be unnecessary for both learning and recognition.

    Conclusions:

    • Sparse coding can be effectively employed for learning optimized intermediate features in hierarchical feedforward networks.
    • The developed model offers a competitive approach to invariant object recognition, robust to clutter and without requiring segmentation.
    • The findings highlight the potential of integrating sparse coding with specific nonlinearities for enhanced recognition capabilities.