A neural network model to solve analogical equations between strings of symbols is proposed. The method transforms the input strings into two fixed size alignment matrices. The matrices act as the input of the neural network which predicts two output matrices. Finally, a string decoder transforms the predicted matrices into the final string output. By design, the neural network is constrained by several properties of analogy. The experimental results show a fast learning rate with a high prediction accuracy that can beat a baseline algorithm.
|Number of pages||10|
|Journal||CEUR Workshop Proceedings|
|Publication status||Published - 2016|
- Neural networks
- Proportional analogy
ASJC Scopus subject areas
- Computer Science(all)