<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[sweet_Mary的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/sweet_Mary</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; sweet_Mary]]></copyright><item><title><![CDATA[保研机试之【动态规划子序列问题】四种问题+总结]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/138999994</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/138999994</guid><author>sweet_Mary</author><pubDate>Fri, 09 Aug 2024 19:41:49 +0800</pubDate><description><![CDATA[动态规划子序列问题]]></description><category></category></item><item><title><![CDATA[线性回归Pytorch方法]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/140999317</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/140999317</guid><author>sweet_Mary</author><pubDate>Fri, 09 Aug 2024 19:37:33 +0800</pubDate><description><![CDATA[在每个epoch中，使用一个样本进行更新的方法称为随机梯度下降，而使用所有样本进行更新的方法称为批量梯度下降。允许处理更大的数据集（无法容纳在内存中）随机梯度下降的问题：近似损失快速波动。超参数：lr、batch_size。使用验证数据集来决定超参数大小。替换criterion函数。有点网格搜索优化的感觉。]]></description><category></category></item><item><title><![CDATA[Pytorch实现线性回归Linear Regression]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/140802514</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/140802514</guid><author>sweet_Mary</author><pubDate>Thu, 01 Aug 2024 21:08:58 +0800</pubDate><description><![CDATA[Batch Gradient Descent：使用整个训练集来更新模型的参数。学习率过低，需要大量的迭代才能获得最小值。对于线性回归，特定类型的噪声是高斯噪声。创建自定义模块（内含一个线性回归）用PyTorch实现线性回归模块。学习率过高，可能错过参数的最佳值。与上文类似，只是多加了个b。]]></description><category></category></item><item><title><![CDATA[保研机试之【滑动窗口】]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/140666243</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/140666243</guid><author>sweet_Mary</author><pubDate>Thu, 01 Aug 2024 15:49:45 +0800</pubDate><description><![CDATA[本篇为的笔记。基于滑动窗口算法框架写出的代码，时间复杂度是O(N)，比嵌套 for 循环的暴力解法效率高。]]></description><category></category></item><item><title><![CDATA[张量Tensor]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/140671630</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/140671630</guid><author>sweet_Mary</author><pubDate>Fri, 26 Jul 2024 19:23:33 +0800</pubDate><description><![CDATA[张量运算的本质是向量和矩阵运算。神经网络的输入、输出、参数都将采用张量进行。Pytorch中的张量可以和Python中的numpy相互转换，这使得Pytorch在GPU上应用成为可能。神经网络中的参数是允许计算梯度或导数的张量。]]></description><category></category></item><item><title><![CDATA[Segment Anything]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139547780</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139547780</guid><author>sweet_Mary</author><pubDate>Sat, 08 Jun 2024 16:49:13 +0800</pubDate><description><![CDATA[参考：]]></description><category></category></item><item><title><![CDATA[BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Mode]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139544285</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139544285</guid><author>sweet_Mary</author><pubDate>Sat, 08 Jun 2024 12:04:10 +0800</pubDate><description><![CDATA[计算image transformer输出query表征Z（与可学习query长度相同）与text transformer输出文本表征 t 中【CLS】token相似性，选取最大值作为图像文本对相似度。为防止信息泄露，作者使用单模态self-attention mask，query与text不能互相可见，防止从文本直接学习。]]></description><category></category></item><item><title><![CDATA[CLIP--Learning Transferable Visual Models From Natural Language Supervision]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139377609</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139377609</guid><author>sweet_Mary</author><pubDate>Sat, 01 Jun 2024 22:58:47 +0800</pubDate><description><![CDATA[参考：]]></description><category></category></item><item><title><![CDATA[图像分割模型LViT-- (Language meets Vision Transformer)]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139246766</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139246766</guid><author>sweet_Mary</author><pubDate>Tue, 28 May 2024 23:12:53 +0800</pubDate><description><![CDATA[使得无标签数据可以像有标签数据一样为模型提供监督信息。这种方式能够有效利用大量的无标签数据，提高模型的泛化能力和鲁棒性。左面的红方框是Transformer支路，右面的红方框是CNN支路。双u型结构：u型CNN支路+u型Transformer支路。]]></description><category></category></item><item><title><![CDATA[医学图像分割--U-net变种]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139218767</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139218767</guid><author>sweet_Mary</author><pubDate>Mon, 27 May 2024 15:52:50 +0800</pubDate><description><![CDATA[这里以3*256*256的图片为例，手动推算了一遍，有问题的小伙伴可以对照着代码自己手推一遍。这是我自己手推的示意图，没明白的朋友建议先补一下注意力机制的Q、K、V，然后来看看图鸭~也是类似的，朋友们一定要自己推导一遍U-net，这个后面的内容看起来就会很轻松。类（水个字哈哈哈）---两个卷积层、批量归一化和ReLU激活函数。由两个连续卷积层（可选批量归一化）和ReLU激活函数组成的模块。网络结构，包括编码（下采样）和解码（上采样）部分。普通的卷积模块（两个卷积层和批归一化层）类---上采样卷积块。]]></description><category></category></item><item><title><![CDATA[吉林大学软件工程简答题整理]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139184828</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139184828</guid><author>sweet_Mary</author><pubDate>Sun, 26 May 2024 16:33:43 +0800</pubDate><description><![CDATA[期末复习，好运来，好运来]]></description><category></category></item><item><title><![CDATA[吉林大学软件工程易错题]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139184007</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139184007</guid><author>sweet_Mary</author><pubDate>Fri, 24 May 2024 21:09:21 +0800</pubDate><description><![CDATA[15.UML四种基本关系：关联关系（实线，聚合是白色菱形，组合是黑色菱形）、依赖关系（A------->B，说明A依赖B）、泛化关系（实线，白色三角，三角指向的是父类）、实现关系（虚线，三角，A----层次图中的一个矩形框代表一个模块，方框间的连线表示调用关系，层次图很适于在自顶向下设计软件的过程中使用。精化阶段--生命周期架构；极限编程是采取必要的手段，充分挖掘软件开发团队人员的极限能力，在最短的时间内交付软件的开发方法。10.【单选题】（ ）着重反映的是模块间的隶属关系，即模块间的调用关系和层次关系。]]></description><category></category></item><item><title><![CDATA[LAMDA面试准备（2024-05-23）]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139156289</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139156289</guid><author>sweet_Mary</author><pubDate>Fri, 24 May 2024 20:29:44 +0800</pubDate><description><![CDATA[LAMDA准备，谢谢老师给的机会，为了后面更顺利的面试吧]]></description><category></category></item><item><title><![CDATA[医学图像分割]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139111286</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139111286</guid><author>sweet_Mary</author><pubDate>Thu, 23 May 2024 17:36:27 +0800</pubDate><description><![CDATA[医学影像]]></description><category></category></item><item><title><![CDATA[吉林大学软件工程UML+OOA]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139058698</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139058698</guid><author>sweet_Mary</author><pubDate>Mon, 20 May 2024 12:11:01 +0800</pubDate><description><![CDATA[面向对象=对象+类+继承+通过消息进行通信。]]></description><category></category></item><item><title><![CDATA[训练扩散模型---史密森尼蝴蝶数据集]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/138975063</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/138975063</guid><author>sweet_Mary</author><pubDate>Sun, 19 May 2024 14:15:45 +0800</pubDate><description><![CDATA[训练扩散模型---史密森尼蝴蝶数据集]]></description><category></category></item><item><title><![CDATA[吉林大学软件工程1~2]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/139008967</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/139008967</guid><author>sweet_Mary</author><pubDate>Sat, 18 May 2024 17:18:04 +0800</pubDate><description><![CDATA[软件开发退化修改软件出故障：无法用备用零件替换来解决，是因为设计开发过程中存在错误软件本身软件开发与维护的方法。]]></description><category></category></item><item><title><![CDATA[扩散模型自动管道AutoPipeline]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/138973932</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/138973932</guid><author>sweet_Mary</author><pubDate>Thu, 16 May 2024 20:57:08 +0800</pubDate><description><![CDATA[自动管道]]></description><category></category></item><item><title><![CDATA[保研机试之【子序列】]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/138954783</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/138954783</guid><author>sweet_Mary</author><pubDate>Thu, 16 May 2024 11:49:15 +0800</pubDate><description><![CDATA[子序列]]></description><category></category></item><item><title><![CDATA[保研机试之【execve函数】]]></title><link>https://blog.youkuaiyun.com/sweet_Mary/article/details/138891774</link><guid>https://blog.youkuaiyun.com/sweet_Mary/article/details/138891774</guid><author>sweet_Mary</author><pubDate>Wed, 15 May 2024 10:30:36 +0800</pubDate><description><![CDATA[execve]]></description><category></category></item></channel></rss>