TY - GEN
T1 - A Brain-Inspired Distributed Long-Term Memory Guided Online Continual Learning Method
AU - Han, Yuyang
AU - Li, Xiuxing
AU - Wang, Qixin
AU - Jia, Tianyuan
AU - Wu, Xia
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.
PY - 2025
Y1 - 2025
N2 - Online Continual Learning (CL) is a challenging scenario that focuses on enabling models to incrementally learn new knowledge from sequential, non-i.i.d. data with only a single pass through the data. Replay-based methods have shown great potential in this scenario, but they still suffer from catastrophic forgetting. In contrast, the human brain exhibits great advantages in online CL scenario, with memories swiftly encoded as short-term memories in the hippocampus and subsequently consolidated into distributed long-term memories within the neocortex. Inspired by this neural mechanism, we introduce the Distributed Long-Term Memory Guided CL(DLTMG-CL) method, which guides the network to retain old knowledge while learning new information by consolidating and preserving distributed long-term memories. This approach facilitates efficient learning in online CL scenario. Our algorithm has achieved state-of-the-art (SOTA) performance among various CL methods.
AB - Online Continual Learning (CL) is a challenging scenario that focuses on enabling models to incrementally learn new knowledge from sequential, non-i.i.d. data with only a single pass through the data. Replay-based methods have shown great potential in this scenario, but they still suffer from catastrophic forgetting. In contrast, the human brain exhibits great advantages in online CL scenario, with memories swiftly encoded as short-term memories in the hippocampus and subsequently consolidated into distributed long-term memories within the neocortex. Inspired by this neural mechanism, we introduce the Distributed Long-Term Memory Guided CL(DLTMG-CL) method, which guides the network to retain old knowledge while learning new information by consolidating and preserving distributed long-term memories. This approach facilitates efficient learning in online CL scenario. Our algorithm has achieved state-of-the-art (SOTA) performance among various CL methods.
KW - Brain-inspired Method
KW - Catastrophic Forgetting
KW - Continual Learning
KW - Online Continual Learning
UR - http://www.scopus.com/inward/record.url?scp=105003630595&partnerID=8YFLogxK
U2 - 10.1007/978-981-96-4001-0_23
DO - 10.1007/978-981-96-4001-0_23
M3 - Conference contribution
AN - SCOPUS:105003630595
SN - 9789819640003
T3 - Communications in Computer and Information Science
SP - 331
EP - 343
BT - Human Brain and Artificial Intelligence - 4th International Workshop, HBAI 2024, Proceedings
A2 - Liu, Quanying
A2 - Qu, Youzhi
A2 - Wu, Haiyan
A2 - Qi, Yu
A2 - Zeng, An
A2 - Pan, Dan
PB - Springer Science and Business Media Deutschland GmbH
T2 - 4th International Workshop on Human Brain and Artificial Intelligence, HBAI 2024
Y2 - 3 August 2024 through 3 August 2024
ER -