Combinets: Creativity via Recombination of Neural Networks
Full Text: iccc19-paper-11.pdf
One of the defining characteristics of human creativity is the ability to make conceptual leaps, creating something surprising from existing knowledge. In comparison, deep neural networks often struggle to handle cases outside of their training data, which is especially problematic for problems with limited training data. Approaches exist to transfer knowledge from models trained on one problem with sufficient data to new problems with insufficient data, but they tend to require additional training or a domain-specific method of transfer. We present conceptual expansion, a general approach for reusing existing trained models to derive new models without backpropagation. We evaluate our approach on few-shot variations of two tasks: image classification and image generation, and outperform standard transfer learning approaches.
Citation
M. Guzdial, M. Riedl. "Combinets: Creativity via Recombination of Neural Networks". International Conference on Computational Creativity (ICCC), (ed: Kazjon Grace, Michael Cook, Dan Ventura, Mary Lou Maher), pp 180-187, June 2019.| Keywords: | |
| Category: | In Conference |
BibTeX
@incollection{Guzdial+Riedl:ICCC19,
author = {Matthew Guzdial and Mark Riedl},
title = {Combinets: Creativity via Recombination of Neural Networks},
Editor = {Kazjon Grace, Michael Cook, Dan Ventura, Mary Lou Maher},
Pages = {180-187},
booktitle = {International Conference on Computational Creativity (ICCC)},
year = 2019,
}Last Updated: October 29, 2020Submitted by Sabina P
