人工智能技术格局导航.pptx

1、Exploring the Interconnected World of LogisticRegression,Neural Networks,and Computer Vision,Liliang ChenFreddie M,Agenda,Basics of Logistic Regression Understand from the Sum of VectorsGeometric Interpretation of a Logistic RegressionGoal of an Activation FunctionLogistic Regression vs.Neural Netwo

2、rk:One-Layer,Two Layer,Hidden LayerBack-propagation of Error vs.Forward ActivationApplication of Convolutional Neural Network(CNN)in Computer Vision Difference between Sigmoid and RELU Activation FunctionCNN vs.Neural Network CNNs Back-propagation,Basics of Logistic Regression,Binary Classification:

3、0,1 Predict the probability of being in a particular class:=1;)Could fit a linear model:,=Use the sigmoid function to force the output to lie in 0,1 range:,=()=1 1+,Understand from Geometry(the Sum of Vectors),A,1,2,=A+B=1 1+2 2=,=,1,2,3,B,w,w,A,B,C,1,2,1,2,3,=,=A+B=1 1+2 2=,=C+D=c+3 3=1 1+2 2+3 3=,

4、D,into a single vector with a deterministic direction and magnitude,We can think of w T x as tranforming from multiple independent vectors,Vector addition in 2D,Vector addition in 3D,=1 2+2 2,=1 2+2 2+3 2,Understanding the linear model shape=0 from geometry,defines a line in a two-dimension space vs

5、 a hyper-plane in a three-dimension space,0+1 1+2 2=0,=0,0+1 1+2 2+3 3=0,The unit vector normal to this plane is,The unit vector normal to this line is,The unit vector normal to the line/plane has the same direction as the sum of vectors based on individual vectors.,3,0,Geometric Interpretation of a

6、 Logistic Regression,Decision Boundary as a hyperplane in 3D,Logistic regression seeks the decision boundary to perfect linear separate positive and negative points;,Decision Boundary as a line in 2D,Classification depends on comparing relative distance from the origin to the data points vs.the deci