Unverified50% confidenceFactExact time
一段数据参与神经网络训练后，其影响以分布式方式弥散在数十亿参数中，难以被真正删除
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
Unverified深度神经网络天然倾向于过拟合训练分布，对分布外样本的处理能力往往大幅下降77% similarUnverifiedCatastrophic forgetting is a classic problem in the field of neural networks; as early as the 1990s, researchers discovered that the performance on old tasks drops sharply when neural networks are continuously trained on new tasks69% similarUnverifiedThe understanding of text-to-image models is essentially large-scale statistical association, and generation quality drops significantly when dealing with long-tail concepts that appear rarely in the training set.69% similarUnverified过度域随机化会引入与真实分布相差过大的不可能样本，稀释有效训练信号，随机化范围的设定需要大量领域专家经验69% similarUnverified当生成模型反复在自身生成的数据上训练时，输出分布会逐渐退化，低概率事件被系统性遗忘69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/687691API
curl https://kongchang.com/api/v1/knowledge/claims/687691MCP
get_claim(id=687691)