BoltzCONS: dynamic symbol structures in a connectionist network
Artificial Intelligence - On connectionist symbol processing
Mapping part-whole hierarchies into connectionist networks
Artificial Intelligence - On connectionist symbol processing
Recursive distributed representations
Artificial Intelligence - On connectionist symbol processing
Artificial Intelligence - On connectionist symbol processing
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Learning internal representations by error propagation
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Learning word meaning and grammatical constructions from narrated video events
HLT-NAACL-LWM '04 Proceedings of the HLT-NAACL 2003 workshop on Learning word meaning from non-linguistic data - Volume 6
2005 Special issue: Recursive principal components analysis
Neural Networks - Special issue on neural networks and kernel methods for structured domains
A Neurolinguistic Model of Grammatical Construction Processing
Journal of Cognitive Neuroscience
Vector Symbolic Architectures: A New Building Material for Artificial General Intelligence
Proceedings of the 2008 conference on Artificial General Intelligence 2008: Proceedings of the First AGI Conference
Exact solutions for recursive principal components analysis of sequences and trees
ICANN'06 Proceedings of the 16th international conference on Artificial Neural Networks - Volume Part I
Semi-supervised recursive autoencoders for predicting sentiment distributions
EMNLP '11 Proceedings of the Conference on Empirical Methods in Natural Language Processing
Representing objects, relations, and sequences
Neural Computation
| Hi-index | 0.00 |
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.