Disentanglement of Color and Shape Representations for Continual Learning
Lists
Only you see this
Abstract
We hypothesize that disentangled feature representations suffer less from catastrophic forgetting. As a case study we perform explicit disentanglement of color and shape, by adjusting the network architecture. We tested classification accuracy and forgetting in a task-incremental setting with Oxford-102 Flowers dataset. We combine our method with Elastic Weight Consolidation, Learning without Forgetting, Synaptic Intelligence and Memory Aware Synapses, and show that feature disentanglement positively impacts continual learning performance.
BibTeX
@misc{berga2020disentanglement,
author = {David Berga and Marc Masana and Joost Van de Weijer},
title = {{Disentanglement of Color and Shape Representations for Continual Learning}},
year = {2020},
eprint = {2007.06356},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2007.06356},
}Loading related content...