Verified75% confidenceFactExact time
EWC uses the diagonal elements of the Fisher information matrix to measure the importance of each parameter to old tasks, adding a penalty term L_new + λ·Σ_i F_i·(θ_i - θ*_i)² to the loss function, without requiring storage of old data.
3
Sources
75%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
BERT新增分类类别：增量微调的三种策略与灾难性遗忘解法
Related Claims
UnverifiedEWC等正则化方法无需存储旧数据，适合数据隐私敏感或存储受限的场景，但计算Fisher信息矩阵需在旧数据上进行一次完整前向传播64% similarUnverifiedOOP的典型实现存在性能缺陷：对象分散在堆内存导致缓存不友好、虚函数调用引入间接跳转破坏分支预测、深层继承树增加数据冗余59% similarUnverifiedAny combination of bounded storage plus an eviction policy implicitly contains a value judgment about "what is worth keeping"; the default LRU or FIFO presupposes that "time" is a proxy variable for importance, but this fails in scenarios such as retries, deduplication, and anomaly detection.56% similarUnverifiedKV Cache reduces inference computational complexity from O(n²) to O(n) by caching and reusing the Key and Value vectors of historical tokens; if the cache is incorrectly truncated, overwritten, or subject to cross-session interference, it causes context confusion and triggers loops.55% similarUnverifiedBranch-based restores reduce the complexity of recovery operations from O(data size) to near O(1) metadata writes.55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502640API
curl https://kongchang.com/api/v1/knowledge/claims/502640MCP
get_claim(id=502640)