Unverified50% confidenceFactExact time
The core idea of EWC is to first use the Fisher information matrix to estimate the importance of each parameter to the old task, and then apply a quadratic penalty to important parameters during subsequent fine-tuning so that they are not easily changed substantially.
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the field of continual learning, methods such as elastic weight consolidation (EWC) use Fisher information as a proxy for parameter importance, imposing stronger update penalties on important parameters to mitigate catastrophic forgetting.79% similarUnverifiedExpected Information Gain (EIG) is the core metric of Bayesian experimental design, used to quantify how much uncertainty about the hidden state a candidate question can reduce in expectation.72% similarUnverifiedEmpirical Game-Theoretic Analysis (EGTA) was systematically proposed by Michael Wellman and others at the University of Michigan; it constructs an empirical payoff matrix from match data collected over a limited strategy set in order to find Nash equilibria.68% similarUnverified研究者已提出弹性权重巩固(EWC)、渐进式神经网络等缓解灾难性遗忘的方案,但在超大规模模型上的有效性仍是活跃研究方向67% similarUnverifiedOutlines通过将EBNF文法预编译为有限状态机(FSM)来高效实现约束采样67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1149860API
curl https://kongchang.com/api/v1/knowledge/claims/1149860MCP
get_claim(id=1149860)