<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[qq_44763553的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/qq_44763553</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; qq_44763553]]></copyright><item><title><![CDATA[uv安装方法]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/156415559</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/156415559</guid><author>qq_44763553</author><pubDate>Tue, 30 Dec 2025 09:59:59 +0800</pubDate><description><![CDATA[主包用powershell的命令行一直下不下来，所以直接去github上下载发布版本了。]]></description><category></category></item><item><title><![CDATA[python读取ATL15.nc数据并保存为]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/145386483</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/145386483</guid><author>qq_44763553</author><pubDate>Tue, 28 Jan 2025 11:16:48 +0800</pubDate><description><![CDATA[ATL15数据是nc格式，分成了很多组，arcgis不能直接打开。]]></description><category></category></item><item><title><![CDATA[网络中的网络NiN]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/145145780</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/145145780</guid><author>qq_44763553</author><pubDate>Tue, 14 Jan 2025 20:03:43 +0800</pubDate><description><![CDATA[基本上所有的参数都在全连接层，会带来过拟合训练大头是卷积层之后的全连接层NiN的思想是完全不用全连接层一个卷积层后跟两个全连接层。]]></description><category></category></item><item><title><![CDATA[VGGhhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/145145641</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/145145641</guid><author>qq_44763553</author><pubDate>Tue, 14 Jan 2025 18:58:18 +0800</pubDate><description><![CDATA[多个VGG块后接全连接层，不同次数的重复块得到不同的架构。如何更好地更深和更大？]]></description><category></category></item><item><title><![CDATA[池化层hhhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/145022916</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/145022916</guid><author>qq_44763553</author><pubDate>Thu, 09 Jan 2025 09:54:56 +0800</pubDate><description><![CDATA[池化层返回窗口中最大或平均值缓解卷积层对位置的敏感性同样有窗口大小、填充、步幅作为超参数没有可以学习的参数。]]></description><category></category></item><item><title><![CDATA[卷积层hhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144989606</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144989606</guid><author>qq_44763553</author><pubDate>Thu, 09 Jan 2025 09:05:33 +0800</pubDate><description><![CDATA[卷积层将输入和和矩阵进行交叉相关，加上偏移后得到输出核矩阵和偏移是可学习的参数核矩阵的大小是超参数数据多样性大/学习率大每个通道的卷积核是一样的，不同通道的卷积核是不一样的。]]></description><category></category></item><item><title><![CDATA[数值稳定性+模型初始化和激活函数]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144895877</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144895877</guid><author>qq_44763553</author><pubDate>Mon, 06 Jan 2025 19:27:06 +0800</pubDate><description><![CDATA[所有h都是向量，向量对向量求导是矩阵，里面会做d-t次矩阵乘法。]]></description><category></category></item><item><title><![CDATA[丢弃法hhhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144890787</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144890787</guid><author>qq_44763553</author><pubDate>Thu, 02 Jan 2025 21:53:57 +0800</pubDate><description><![CDATA[一个好的模型需要对输入数据的扰动鲁棒丢弃法：在层之间加入噪音，等同于加入正则h2和h5变成0了dropout一般作用在全连接隐藏层的输出上。]]></description><category></category></item><item><title><![CDATA[权重衰退hhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144884234</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144884234</guid><author>qq_44763553</author><pubDate>Thu, 02 Jan 2025 16:59:50 +0800</pubDate><description><![CDATA[通过限制参数值的选择范围来控制模型容量。]]></description><category></category></item><item><title><![CDATA[模型选择+过拟合欠拟合]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144787530</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144787530</guid><author>qq_44763553</author><pubDate>Mon, 30 Dec 2024 23:08:51 +0800</pubDate><description><![CDATA[训练误差：模型在训练数据上的误差泛化误差：模型在新数据上的误差验证数据集：一个用来评估模型好坏的数据集测试数据集：只能用一次。]]></description><category></category></item><item><title><![CDATA[多层感知机]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144771333</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144771333</guid><author>qq_44763553</author><pubDate>Sat, 28 Dec 2024 11:24:41 +0800</pubDate><description><![CDATA[感知机就是一个二分类问题max对应if语句，如果分类正确，y需要大于0，梯度是一个常量感知机是一个二分类模型本质上是学习率为1的梯度下降不能解决XOR问题。]]></description><category></category></item><item><title><![CDATA[softmax回归]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144763212</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144763212</guid><author>qq_44763553</author><pubDate>Fri, 27 Dec 2024 14:08:02 +0800</pubDate><description><![CDATA[回归：估计一个连续值分类：预测一个离散类别只有一个yi为1，其他都为0分类问题关心的是对分类问题置信度更大，而不是关心具体的值。]]></description><category></category></item><item><title><![CDATA[基础优化方法]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144753862</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144753862</guid><author>qq_44763553</author><pubDate>Thu, 26 Dec 2024 23:12:59 +0800</pubDate><description><![CDATA[学习率代表每一次沿着这个方向走多远，]]></description><category></category></item><item><title><![CDATA[线性回归a]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144753618</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144753618</guid><author>qq_44763553</author><pubDate>Thu, 26 Dec 2024 21:46:57 +0800</pubDate><description><![CDATA[求平方损失的平均值1/n，目标求解w，b使得最小。]]></description><category></category></item><item><title><![CDATA[导数hhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144662819</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144662819</guid><author>qq_44763553</author><pubDate>Tue, 24 Dec 2024 11:18:18 +0800</pubDate><description><![CDATA[内存复杂度：反向传播O(1)，正向累计O(n)，需要记录所有操作子。]]></description><category></category></item><item><title><![CDATA[用于卫星影像间接RPC模型精化的通用光束法平差方法]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/144405710</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/144405710</guid><author>qq_44763553</author><pubDate>Wed, 11 Dec 2024 21:58:10 +0800</pubDate><description><![CDATA[介绍了通用RPC模型的表达式，which has been down to death描述了RPC模型产生误差的原因——主要与定义传感器方位的姿态角有关。每个影像都会对应一个三维点云，但是对同一地物拍摄的不同影像对应出来的三维点云是不一样的，所以才需要进行BA来约束示例说明RPC精化对多视图三维重建的好处。( a )从四组不同的卫星影像中计算出独立的点云。( b )模型在对象空间中的相对位置，采用未精化的RPC模型。( c )模型在对象空间中的相对位置，采用本文方法修正后的RPC模型。]]></description><category></category></item><item><title><![CDATA[NeRFhh]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/143673076</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/143673076</guid><author>qq_44763553</author><pubDate>Mon, 11 Nov 2024 22:27:31 +0800</pubDate><description><![CDATA[一个NeRF神经网络只能存储一个三维物体或者三维场景的信息在沿光线射线的路径上都有对应的RGBA值加位置编码，通过concat方式全连接层得到最终结果位置编码为什么是20*3？相当于做了特征增强T(t)不透光率，关注近边界，不关注远边界密度高的地方多采样一些点，密度低的地方少采样一些点。]]></description><category></category></item><item><title><![CDATA[Spring Cloud微服务]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/141230532</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/141230532</guid><author>qq_44763553</author><pubDate>Mon, 11 Nov 2024 20:36:47 +0800</pubDate><description><![CDATA[网关转发之前做登录校验因为每一个微服务都需要获得用户的登录信息，比如购票、选座什么的，不可能让每一个微服务都去做JWT校验，所以把JWT校验放在【网关】中。网关向微服务发起的是HTTP请求，所以把JWT校验信息保存在请求头中。]]></description><category></category></item><item><title><![CDATA[Docker]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/143476566</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/143476566</guid><author>qq_44763553</author><pubDate>Mon, 04 Nov 2024 21:16:30 +0800</pubDate><description><![CDATA[/ -t 是tag的意思// myImage:1.0 冒号前面是repository，后面是版本号，版本号可以省略，默认是latest// 强烈注意最后的“ .”用于指定Dockerfile中的目录，当前目录是“ .”]]></description><category></category></item><item><title><![CDATA[Redis登录校验]]></title><link>https://blog.youkuaiyun.com/qq_44763553/article/details/142747543</link><guid>https://blog.youkuaiyun.com/qq_44763553/article/details/142747543</guid><author>qq_44763553</author><pubDate>Mon, 14 Oct 2024 11:00:38 +0800</pubDate><description><![CDATA[第一个拦截器只是确保一切请求都要进行token刷新的大动作第二个拦截器从ThreadLocal中获取user用户。]]></description><category></category></item></channel></rss>