TY  - CONF
AU  - David Berga
AU  - Marc Masana
AU  - Joost Van de Weijer
A2  - ICMLW
PY  - 2020//
TI  - Disentanglement of Color and Shape Representations for Continual Learning
BT  - ICML Workshop on Continual Learning
N2  - 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.
L1  - http://refbase.cvc.uab.es/files/BMW2020.pdf
N1  - LAMP; 600.120
ID  - David Berga2020
ER  -