Regularization: extreme case 1 define the loss...

Regularization: extreme case 1 define the loss function J_{n, lambda}left( heta, heta_0 ight)=frac{1}{n} sum_{t=1}^n frac{left(y^{(t)}- heta cdot x^{(t)}- heta_0 ight)^2}{2}+frac{lambda}{2}| heta|^2 where lambda is the regularization factor. In the figure above, the blue dots are the training examples. If we increase lambda to infty, where does f(x)= heta cdot x+ heta_0 converge to? line 1 line 2 line 3

Transcript

00:01 Hello students here are the step to solve the perception learning algorithm for the given training sample.

00:07 So here this is given x1 x2 in the class.

00:11 So x1 x2 training samples are given with class 11 plus 1 minus 1 minus 1 then minus 1 y 0 .5 minus 1 x 1 .1 x 2 .5 minus 1.

00:28 So all of these sample data is given to us and assuming weight vector of the initial decision boundaries w tx.

00:36 So here we will equal we will assume 0 and w 11 as a weight matrix weight.

00:43 So the boundaries 1 in how many step will the perception learning algorithm coverage number 2 what will be the final decision boundary and show the response stepwise update.

00:52 So here we have to do this task.

00:55 So initialize the first i am going with the perception algorithm initialize the weight vector to 11 for each training sample x1 x2 and y if y into w tx which is a weight vector is less than 0 update the weight vector as follows w plus w plus y into x1 x2.

01:13 So repeat the step 2 until the perception coverage that is until no training sample is misclassified in this case the perception will coverage in the 4 steps.

01:23 The final decision boundary will be a line that separate the positive and the negative example here is the stepwise update of the weight vector i am writing.

01:34 So here initial vector i will take w 11.

01:39 So here x1 x2 for the first one is 111 all are the one and if now i am using this if it is less than 0 then i have to update...

Question

Please give Ace some feedback