Hello,
Can anyone tell me more about what extra content I can get from Prof. Ng’s MLS course Advanced Learning Algorithms if I sign up at $49/month?
Thanks.
Hello, Stephen,
An introduction to Neural Network and its terminology.
Course 1 covered g \circ f = g(f(\vec{x}; \vec{w}, b)) where f(\vec{x}; \vec{w}, b) = \vec{w} \cdot \vec{x} +b and g(x) = \frac{1}{1 + e^{-x}} (logistic regression) or g(x) = x (linear regression), whereas Course 2 covers g^{[L]} \circ f^{[L]} \circ g^{[L-1]} \circ f^{[L-1]} \circ \dots \circ g^{[1]} \circ f^{[1]}, where f^{[l]}(X; W^{[l]}, b^{[l]}) = XW^{[l]} +b^{[l]} and X, W^{[l]} and b^{[l]} are all matrices.
While the sigmoid function and the linear function are the candidates for g^{[L]} , other g^{[l]} take the very popular ReLU function, or, mathematically, g(x) = \begin{cases} 0, & \text{if}\ x \lt 0 \\ x, & \text{otherwise} \end{cases} or simply g(x) = max(0, x). These will be covered in the course as well.
From course 1, we know \vec{w} and b are optimized with gradient descent, and we also know \alpha (the learning rate), \lambda (the L2 regularization parameter) and the choices of polynomial features (e.g. to what degree) are not optimized that way but more like, which I would call it, “chosen heristically”. We call \vec{w} and b as trainable parameters, whereas those chosen heristically as hyperparameters. In course 2, we will see how we use “Bias and Variance” to tune hyperparameters, including the L in the above function composition, from some initial choice.
The last part of the course is an introduction to decision tree, covering the basic “how-it-works”.
Besides my comment above, we can also audit the course’s videos and reading items for free, which will give us another angle.
Cheers,
Raymond
2 Likes
Hi Raymond,
Thanks for that explanation.
I think my problem was more to do with thinking I wouldn’t get to complete the specialization unless I subscribed for $49 / month. But I have now applied for financial aid so I will see what happens.
1 Like
As Raymond mentioned in the last paragraph of his post, the lectures are available for free by taking the course in “audit” mode. What you miss by doing that is access to the graded part of the course: the quizzes, the programming assignments and the ability to get a certificate for completing the course if that is relevant to you. In order to access the quizzes and the programming assignments you need to pay for the course, either by subscribing or by applying for financial aid. If you do the financial aid method, note that they require you to apply for financial aid individually for each course rather than allowing you to do it for the specialization as a whole.
How did you handle that for MLS C1? Did you get access to the assignments?
The other way to see the assignments is that they give you a 7 day free trial when you first start one of the courses. Or maybe that only happens for the first course of a multicourse specialization. I’m not sure what happens if you actually complete all the assignments within the 7 day free trial period. If I had to guess, I would bet that they only give you the certificate if you pay by one of the above methods.
1 Like
Very nice rendering of the math here Raymond.
Stephen.