andrew ng deep learning specialization

When should do we use fine-tuning?It is typically used only if you have a large labeled training set; In this setting, fine-tuning can significantly improve the performance of your classifier. However, if you had a large unlabeled dataset (for unsupervised feature learning/pre-training) and only a relatively smal L labeled training Set, then fine-tuning was significantly less likely to help.Stacked Autoencoders (Training):Equivalent to capturing the c

CLA Ssifier. But no worries, you'll build a even better classifier next week! Also, you'll be clearly the model is overfitting the training data. Later in this specialization you would learn to reduce overfitting and for example by using regularization. Using the code below (and changing the "index variable") You can look at the in predictions on pictures of the test set. gradient descent gradual optimization loss function Interpretation:You can dec

This semester has been to follow up on the Coursera Machina learning public class, the teacher Andrew Ng is one of the founders of Coursera, machine learning aspects of Daniel. This course is a choice for those who want to understand and master machine learning. This course

what is linear regression. The so-called linear regression (taking a single variable as an example) is to give you a bunch of points, and you need to find a straight line from this pile of points. Figure below This screenshot is from Andrew Ng's What you can do when you find this line. Let's say we find A and b that represent the line, then the line expression is y = a + b*x, so when a new x is present, we can know Y.

rate to characterize the model.mly--12. Takeaways:setting up development and test sets1. Your validation set and test set should be captured as much as possible from the data in your actual application scenario. Validation sets and test sets do not have to be distributed identically to your training data. (I think it's best to have a similar distribution between the training set and the validation set, if the training data and the validation data are distributed too much, you may be able to tra

it is easy to cause the overflow. This is because X and ln (x) have the same monotonicity, and both sides take the logarithmSo this is the J (Theta) that Andrew gave, and the only difference is that Andrew has a negative coefficient in front of it, which makes the maximum value a minimum, so that the gradient descent algorithm can be used.But in fact, with this formula can also complete the task, just use

Recently is a period of idle, do not want to waste, remember before there is a collection of machine learning link Andrew ng NetEase public class, of which the overfiting part of the group will report involved, these days have time to decide to learn this course, at least a superficial understanding.Originally wanted to go online to check machine

of the weights is (0,1).The main ideas of local weighted linear regression are:Where weights are assumed to conform to the formulaThe weight size in the formula depends on the distance between the predicted point X and the training sample. If |-x| is smaller, then the value is close to 1, and vice versa is close to 0. The parameters tau, called bandwidth, are used to control the amplitude of the weights.The advantage of local weighted linear regression is that it is less dependent on feature se

Week 2 gradient descent for multiple variables [1] multi-variable linear model cost function Answer: AB [2] feature scaling feature Scaling Answer: d 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: 【] Answer: [Original] Andrew Ng chose to fill in the blanks in Coursera

"linear regression, gradient descent" The regular equationThe training features are represented as X-matrices, the results are expressed as Y-vectors, and the linear regression model is still the same, and the loss function is unchanged. Then θ can be derived directly from the following formula: The derivation process involves the knowledge of linear algebra, where the linear algebra knowledge is not expanded in detail. Set m as the number of training samples; x is the independent variable in

method provides a method for finding the θ value of the f (θ) =0. How to maximize the likelihood function ? What is the maximum value of the first derivative at the corresponding point? (θ) to zero. So let f (θ) =? ' (θ), maximized ? (θ) can be converted to: Newton's method of seeking ? (θ) The problem of =0 Theta . The expression of the Newton method, the iterative update formula forθ is:Newton-Slavic iteration (Newton-raphson method)in the logistic regression, θ is a vector, so we generalize

build the model.In the exponential distribution family expression of the Bernoulli distribution we have known:, thus obtained.Three assumptions for building a generalized linear model: Assuming that the Bernoulli distribution is met, , in Bernoulli distribution The derivation process is as follows:As with the least squares model, the next work is done by gradient descent or Newton's method.Note the above push to the result, recall, in the logistic regression, we choose th

the gradient descent, when we calculate the derivative term, we need to do the summation, so, in each individual gradient descent, we finally have to calculate such a thing, this item needs to sum all the m training samples. In the following lesson, we will also talk about a method that can solve the minimum value of the cost function J without the need for multi-step gradient descent, which is another called normal equation (normal equations) . The method. In fact, the gradient descent method

This semester has been to follow up on the Coursera Machina learning public class, the teacher Andrew Ng is one of the founders of Coursera, machine learning aspects of Daniel. This course is a choice for those who want to understand and master machine learning. This course

continuously updating theta. Map Reduce and Data Parallelism: Many learning algorithms can be expressed as computing sums of functions over the training set. We can divide up batch gradient descent and dispatch the cost function for a subset of the data to many different machines So, we can train our algorithm in parallel. Week 11:Photo OCR: Pipeline: Text detection Character segmentation Ch

)/m; at End - End - -%size (J,1) -%size (J,2) - ind3 = A3-Ty; -D2 = (D3 * THETA2 (:,2: End)). *sigmoidgradient (z2); toTheta1_grad = Theta1_grad + d2'*a1/m; +Theta2_grad = Theta2_grad + d3'*a2/m; - the% ------------------------------------------------------------- *jj=0; $ Panax Notoginseng forI=1: Size (Theta1,1) - forj=2: Size (Theta1,2) theJJ = JJ + Theta1 (i,j) *theta1 (i,j) *lambda/(m*2); + End A End theSize (Theta1,1); +Size (Theta1,2); - $ forI=1: Size (THETA2,1) $

-validation approach. Cross-validation A simple idea to solve the above model selection problem is that I use 70% of the data to train each model, with 30% of the data for training error calculation, and then we compare the training errors of each model, we can choose the training error is relatively small model. If you do not refer to these errors (learn the theory of experience risk minimization--andrew ng