%0 Conference Proceedings
%T Disentanglement of Color and Shape Representations for Continual Learning
%A David Berga
%A Marc Masana
%A Joost Van de Weijer
%B ICML Workshop on Continual Learning
%D 2020
%F David Berga2020
%O LAMP; 600.120
%O exported from refbase (http://refbase.cvc.uab.es/show.php?record=3506), last updated on Tue, 08 Feb 2022 12:04:17 +0100
%X 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.
%U http://refbase.cvc.uab.es/files/BMW2020.pdf