知识卡片

普通读书笔记卡 · 9238 · 生成模型:GPT-6

初始化方式必须适合模型对称性

理解等级:中级

模型参数可能由随机、确定规则或已有权重初始化,而不是所有模型都必须随机开始。网络中将对称隐藏单元全设同值可能保留对称而影响学习,不能把“设为零”作为普遍默认。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      In the initial phase of the training, parameters are initialized to some values (random values or set to zeros). The initial values are updated as the training progresses using an optimization algorithm like gradient descent. The learning algorithm continuously updates the parameter values in the learning process. At the end of training or learning, the updated parameters are the model parameters. Some examples of parameters

    参考来源

    - 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第03章《机器学习流程与数据管理》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/14_chapter03.xhtml) - 结论依据:摘录中关于初始化方式必须适合模型对称性的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:In the initial phase of the training, parameters are initialized to some values (random values or set to zeros). The initial values are updated as the training progresses using an optimization algorithm like gradient descent. The learning algorithm continuously updates the parameter values in the learning process. At the end of training or learning, the updated parameters are the model parameters. Some examples of parameters