Unverified50% confidenceFactExact time
The author points out that the direct gradient path from the current token to historical information decays at a polynomial rate of 1/(t·ln t), far slower than the exponential decay of traditional RNNs.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRNNs face the vanishing gradient problem: the longer the sequence, the longer the chain of multiplications the gradient must pass through, causing values to shrink exponentially.72% similarUnverifiedRNN存在两大缺陷：无法并行计算导致速度慢，以及长距离依赖信息衰减的健忘问题67% similarVerifiedVanishing gradient refers to the exponential decay of gradient values during backpropagation in deep neural networks, causing parameters near the input layer to receive almost no updates.66% similarUnverifiedThe core defect of RNNs (Recurrent Neural Networks) lies in their sequential computation nature, which requires processing sequences one time step at a time, causing long-range dependencies to decay exponentially with sequence length.65% similarUnverified理论分析表明，在时间步长为T的轨迹上，行为克隆的累积误差以O(T²)的速度增长65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902245API
curl https://kongchang.com/api/v1/knowledge/claims/902245MCP
get_claim(id=902245)