Auto-encoders typically require batch learning to be effective. There is a lack of online learning mechanisms for auto-encoders. To address this shortcoming in the literature, we propose an auto-encoder that can not only learn on a sample-by-sample basis without back-propagation but also has a memory to benefit from past learning. The memory can be adapted to fit the current state of the data by varying the memory length of the auto-encoder. Simulation supports our approach, especially when the data is nonstationary.
|Publication status||Published - Jul 2021|
|Event||The International Joint Conference on Neural Networks 2021 - Virtual|
Duration: 18 Jul 2021 → 22 Jul 2021
|Conference||The International Joint Conference on Neural Networks 2021|
|Abbreviated title||IJCNN 2021|
|Period||18/07/21 → 22/07/21|