0%

经验误差与过拟合

##错误率
错误率(error rate):分类错误的样本数占样本总数的比例
精度(accuracy):精度=1-错误率

如果在m个样本中有a个样本分类错误,
则错误率:
$$E=a/m$$
精度:
$$1-a/m$$

误差error:学习器的实际预测输出与样本的真实输出之间的差异;
训练误差/经验误差:学习器在训练集上的误差;
泛化误差:在新样本上的误差。

我们希望得到泛化误差小的学习器,即在新样本上能表现得很好的学习器。。

过拟合和欠拟合

Read more »

Introduction

  • Give a definition of “Computer Vision”.
  • What type of information is extracted in each of the following applications.
  • Vision is often described as an “ill-posed, inverse problem”. Briefly describe what is meant by the terms “ill-posed” and “inverse problem” and their opposites “well-posed” and “forward problems”.
  • What is a “prior” and how does it help solve the ill-posed, inverse problem of vision? Give examples of priors.
Read more »

Overview of image formation

  • Light from a radiation source is reflected by surfaces in the world..
  • Refleted light passes through some optics and is recorded by a sensor
  • Images are formed when a sensor registers radiation that has interacted with physical objects
Read more »

Introdution to Linear Programming

Definitions

  • Objective function: linear function,$z$, we want to optimise
  • Variables: $x_j$ amounts we seek to assign values to, in order to optimise the objective function.
  • Constraints: linear (in)equalities that must be satisfied by an assignment of variables
    • Non-negativity constraints: constraints of the type $x_j≥0$
    • Resource constraints: the other constraints are often termed as resource constraints(or just constraints when it’s clear from the context).
  • Optimal solution: feasible point(s) at which the objective value is the largest (in case of maximisation) or smallest(in case of minimisation) among all feasible points.
  • Feasible point: point $x$ is feasible for a linear program if it satisfies all its constraints.
  • Feasible region: set of all feasible points of the problem.
Read more »

统计学习方法三要素

统计学习方法都是由模型、策略和算法构成的,即统计学习方法由三要素构成,可以简单地表示为:
$$方法=模型+策略+算法$$

Read more »

写在前面

主要四道题目,1道easy,2道medium,1道hard
主要题目如下:

  • Special Array With X Elements Greater Than or Equal X
  • Even Odd Tree
  • Find Valid Matrix Given Row and Column Sums
  • Minimum One Bit Operations to Make Integers Zero
Read more »

滑动窗口算法

该算法是通过使用特定大小的子列表,在遍历完整列表的同时进行特定的操作,以达到降低了循环的嵌套深度。

Read more »

贪心算法

在对问题求解时,总是做出在当前看来是最好的选择。也就是说,不从整体最优上加以考虑,做出的是某种意义上的局部最优解。

思想

  • 建立数学模型来描述问题;
  • 把求解的问题分成若干个子问题;
  • 对每个子问题求解,得到子问题的局部最优解;
  • 把子问题的解局部最优解合成原来解问题的一个解。
Read more »

写在前面

一共四题,1题easy,2题medium,1题hard
第一道,简单的系统设计题,要清楚object function;
第二道,同样应用题,但是通过率不高;
第三道,给出制定行和&列和的矩阵;
第四道,

  • Design Parking System
  • Alert Using Same Key-Card Three or More Times in a One Hour Period
  • Find Valid Matrix Given Row and Column Sums
  • Find Servers That Handled Most Number of Requests
Read more »

卡诺图 karnaugh map

卡诺图是由美国工程师karnaugh首先提出的一种用来描述逻辑函数的特殊方格图。
在这个方格图中,每个方格代表逻辑函数的一个最小项,而且几何相邻(在几何位置上,上下或左右相邻)的小方格具有逻辑相邻性,即两相邻小方格代表的最小项只有一个变量取值不同。

对于有n个变量的逻辑函数,其最小项有$2^n$个。
因此该逻辑函数的卡诺图又$2^n$个小方格构成,每个小方格都满足逻辑相邻项的要求。
可以运用于逻辑函数的化简。

Read more »