Unverified50% confidenceFactExact time
AlexNet采用ReLU激活函数取代Sigmoid避免梯度消失、Dropout正则化防止过拟合、GPU并行计算加速训练
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedAlexNet的成功得益于ReLU激活函数取代Sigmoid、Dropout正则化和GPU并行计算三项工程突破83% similarUnverified完整保存优化器状态是断点续训的关键，仅保存模型权重会导致优化轨迹偏移影响收敛效果64% similarUnverifiedOC-SORT在经典SORT基础上引入观测中心的动量补偿和在线平滑机制，在目标被遮挡或暂时消失后能更好地恢复追踪62% similarUnverified将复杂任务拆解为小步骤并及时验证有助于规避迷失在中间效应，减少目标漂移60% similarUnverified自动追踪功能存在'追丢'和'反复摆动'问题，追踪算法依赖AI芯片，低端机型追踪延迟大、误触发多，对追踪要求高应选支持人形侦测+追踪的中高端型号60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/532898API
curl https://kongchang.com/api/v1/knowledge/claims/532898MCP
get_claim(id=532898)