Unverified50% confidenceFactExact time

Downsampling operations in intermediate layers of a CNN (such as pooling and strided convolutions) disrupt shift equivariance, making shift invariance impossible to guarantee.

1

Sources

50%

Confidence

Long-term

Relevance

8/30/2026

First Seen

Sources

A Single Pixel Shift Can Fool AI? A Deep Dive into Shift Invariance

Related Entities

下采样concept CNNconcept

Related Claims

Unverified反向传播机制要求整个计算图必须是可微分的，且在极深网络中存在梯度消失/爆炸问题69% similar Unverified网络中间层往往比输出层表现出更严重的叠加态67% similar Unverified传统实例分割网络如Mask R-CNN在高遮挡率下性能会显著下降66% similar Unverified状态离散化存在权衡：分桶粒度过粗导致信息损失，过细则引发维度灾难，这也是DQN用神经网络替代Q表的核心动机66% similar Unverified级联架构面临的核心挑战是误差传播，上游模型的偏差会被放大传递到下游65% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/825957

API

curl https://kongchang.com/api/v1/knowledge/claims/825957

MCP

get_claim(id=825957)