<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[selectnothing的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/selectnothing</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; selectnothing]]></copyright><item><title><![CDATA[FaceChain学习笔记]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/133215530</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/133215530</guid><author>selectnothing</author><pubDate>Sat, 23 Sep 2023 19:45:06 +0800</pubDate><description><![CDATA[大体上来看，FaceChain所做的是训练出一个包含用户面部信息的LoRA模型，然后在推理阶段将该面部LoRA与风格LoRA进行结合并注入，最终实现风格化照片的生成。注：个人在应用时偶尔会出现NoteBook中Gradio地址无法点击的情况，换去Terminal中重新运行即可。P.S.之所以使用人脸融合，大概率是SD生成的人脸不像，为了保证相似度，需要从原始图像中切出人脸进行融合。至于训练和推理流程，在官方README中均有介绍，这里不做过多赘述。按流程操作即可，阿里云会送5000核时的A10使用时间。]]></description><category></category></item><item><title><![CDATA[Task03-大模型的有害性]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/132866017</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/132866017</guid><author>selectnothing</author><pubDate>Wed, 13 Sep 2023 22:18:42 +0800</pubDate><description><![CDATA[性能差异主要指大模型在特定群体或任务场景下表现优秀，在其他场景下表现较差的情况。]]></description><category></category></item><item><title><![CDATA[动手学深度学习笔记---4.6 环境与分布偏移]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129582666</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129582666</guid><author>selectnothing</author><pubDate>Thu, 16 Mar 2023 11:27:43 +0800</pubDate><description><![CDATA[在实际应用场景中，导致模型预测异常的原因不仅仅来自模型参数或结构异常，还可能来自数据自身。神经网络不会理解数据，只会学习数据的特征。因此当数据出现分布异常时，很容易使得模型学到异常的特征，导致灾难性的预测结果.]]></description><category></category></item><item><title><![CDATA[动手学深度学习笔记---4.5 数值稳定性与模型初始化]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129582564</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129582564</guid><author>selectnothing</author><pubDate>Thu, 16 Mar 2023 11:25:37 +0800</pubDate><description><![CDATA[梯度爆炸/消失通常由模型层数过深或激活函数选择不当所导致，应尽量避免梯度过大或者过小。]]></description><category></category></item><item><title><![CDATA[动手学深度学习笔记---4.4 前向传播、反向传播与计算图]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129582180</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129582180</guid><author>selectnothing</author><pubDate>Thu, 16 Mar 2023 11:20:16 +0800</pubDate><description><![CDATA[归根到底，正向传播与反向传播是一个学习参数，计算误差，然后求导对参数更新的过程。可以理解为一次作业的完成过程：写作业（前向传播）-> 老师审阅（计算误差）-> 打回修改（反向传播） -> 完成作业（训练完成）]]></description><category></category></item><item><title><![CDATA[动手学深度学习笔记---4.3 解决过拟合_权重衰减与Dropout]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129582043</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129582043</guid><author>selectnothing</author><pubDate>Thu, 16 Mar 2023 11:05:32 +0800</pubDate><description><![CDATA[L1L2wblwbminLwbsubjectto∣∣w∣∣2≤θbb0θθλminlwb2λ​∣∣w∣∣2λλ0θλ→∞θ→0w∗→0上图中，绿线部分为Lwb,在未引入正则惩罚项2λ​∣∣w∣∣2时，L的最优解在圆心（绿点）处，而引入正则惩罚项后，绿点处的值相对于黄线部分来说非常大，此时不再是最优解。当w∗向原点处移动时，Lwb。]]></description><category></category></item><item><title><![CDATA[动手学深度学习笔记---4.2 模型选择与 过/欠拟合]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129581913</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129581913</guid><author>selectnothing</author><pubDate>Thu, 16 Mar 2023 11:02:52 +0800</pubDate><description><![CDATA[模型在训练集上的误差模型在新数据（测试集）上的误差。用来完成模型训练用来评估模型好坏（切忌和训练集出现数据重复，否则会导致模型性能虚高）的泛化性能，不可使用测试集调整模型超参数。]]></description><category></category></item><item><title><![CDATA[动手学深度学习笔记--4.1 多层感知机]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129580721</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129580721</guid><author>selectnothing</author><pubDate>Thu, 16 Mar 2023 10:47:51 +0800</pubDate><description><![CDATA[与输出实数的线性回归模型以及输出分布概率的Softmax多分类模型不同，感知机（神经元）模型使用Sign函数进行输出处理，输出结果是离散变量，本质属于二分类模型。对于给定的一组输入特征向量xi∈T​，权重w以及偏差bOσ(⟨wxi​⟩bσx1ifxi​0−1otherwise​。]]></description><category></category></item><item><title><![CDATA[从DTW到DILATE---时序预测损失函数总结]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/129193760</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/129193760</guid><author>selectnothing</author><pubDate>Fri, 24 Feb 2023 09:06:06 +0800</pubDate><description><![CDATA[时间序列预测损失函数]]></description><category></category></item><item><title><![CDATA[Day 7 --- OpenMMLab系列值图像分割算法基础（二）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128990781</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128990781</guid><author>selectnothing</author><pubDate>Sat, 11 Feb 2023 23:54:10 +0800</pubDate><description><![CDATA[openmmlab 图像分割（二）]]></description><category></category></item><item><title><![CDATA[Day 6 --- OpenMMLab系列之图像分割算法基础（一）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128976614</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128976614</guid><author>selectnothing</author><pubDate>Fri, 10 Feb 2023 20:16:21 +0800</pubDate><description><![CDATA[openmmlab图像分割基础（一）]]></description><category></category></item><item><title><![CDATA[Day 5 --- OpenMMLab系列之目标检测基础（二）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128936196</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128936196</guid><author>selectnothing</author><pubDate>Wed, 08 Feb 2023 15:15:53 +0800</pubDate><description><![CDATA[OpenMMLab 目标检测第二部分]]></description><category></category></item><item><title><![CDATA[Day 4 --- OpenMMLab系列之目标检测基础（一）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128927932</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128927932</guid><author>selectnothing</author><pubDate>Tue, 07 Feb 2023 23:39:57 +0800</pubDate><description><![CDATA[openmmlab detection]]></description><category></category></item><item><title><![CDATA[Day 3 --- MMClassification工具包]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128881322</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128881322</guid><author>selectnothing</author><pubDate>Sat, 04 Feb 2023 14:01:24 +0800</pubDate><description><![CDATA[OpenMMLab图像分类工具包使用说明]]></description><category></category></item><item><title><![CDATA[Day 2 --- OpenMMLab系列之图像分类算法基础]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128870559</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128870559</guid><author>selectnothing</author><pubDate>Fri, 03 Feb 2023 16:31:19 +0800</pubDate><description><![CDATA[OpenMMLab系列之图像分类基础笔记]]></description><category></category></item><item><title><![CDATA[Day 1 --- 计算机视觉与OpenMMLab开源算法体系]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128844584</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128844584</guid><author>selectnothing</author><pubDate>Thu, 02 Feb 2023 09:38:04 +0800</pubDate><description><![CDATA[CV基础与OpenMMLab体系]]></description><category></category></item><item><title><![CDATA[跨模态搜索系列 --- 二、Jina-跨模态搜索实现]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128765696</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128765696</guid><author>selectnothing</author><pubDate>Thu, 26 Jan 2023 15:34:51 +0800</pubDate><description><![CDATA[基于Sentence-BERT和Jina框架实现文本内容检索]]></description><category></category></item><item><title><![CDATA[一、语音识别方法（3）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128473780</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128473780</guid><author>selectnothing</author><pubDate>Wed, 28 Dec 2022 20:37:31 +0800</pubDate><description><![CDATA[语音识别方法（3）]]></description><category></category></item><item><title><![CDATA[一、语音识别方法（2）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128473658</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128473658</guid><author>selectnothing</author><pubDate>Wed, 28 Dec 2022 20:33:18 +0800</pubDate><description><![CDATA[语音识别方法（2）]]></description><category></category></item><item><title><![CDATA[一、语音识别方法（1）]]></title><link>https://blog.youkuaiyun.com/selectnothing/article/details/128473567</link><guid>https://blog.youkuaiyun.com/selectnothing/article/details/128473567</guid><author>selectnothing</author><pubDate>Wed, 28 Dec 2022 20:18:44 +0800</pubDate><description><![CDATA[语音识别基础（1）]]></description><category></category></item></channel></rss>