Skip to Main content Skip to Navigation
Journal articles

Linear Recursive Distributed Representations

Thomas Voegtlin 1 Peter Dominey 2
1 CORTEX - Neuromimetic intelligence
INRIA Lorraine, LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications
Abstract : Connectionist networks have been criticized for their inability to represent complex structures with systematicity. That is, while they can be trained to represent and manipulate complex objects made of several constituents, they generally fail to generalize to novel combinations of the same constituents. This paper presents a modification of Pollack's Recursive Auto-Associative Memory (RAAM), that addresses this criticism. The network uses linear units and is trained with Oja's rule, in which it generalizes PCA to tree-structured data. Learned representations may be linearly combined, in order to represent new complex structures. This results in unprecedented generalization capabilities. Capacity is orders of magnitude higher than that of a RAAM trained with back-propagation. Moreover, regularities of the training set are preserved in the new formed objects. The formation of new structures displays developmental effects similar to those observed in children when learning to generalize about the argument structure of verbs.
Document type :
Journal articles
Complete list of metadata

https://hal.inria.fr/inria-00000108
Contributor : Thomas Voegtlin <>
Submitted on : Monday, June 13, 2005 - 2:25:27 PM
Last modification on : Friday, February 26, 2021 - 3:28:03 PM
Long-term archiving on: : Thursday, April 1, 2010 - 9:40:38 PM

Files

Identifiers

  • HAL Id : inria-00000108, version 1

Collections

Citation

Thomas Voegtlin, Peter Dominey. Linear Recursive Distributed Representations. Neural Networks, Elsevier, 2005, 18 (7), pp.878-895. ⟨inria-00000108⟩

Share

Metrics

Record views

377

Files downloads

403