CS281A: Statistical Learning Theory (Fall 2007)

Easy (answer on the announcements page): Dec 14: Final project writeups are due on Dec 17. Only

Hey Everyone, So, I hear today is the last class. I was thinking that we should all go to class today

As I'm in Vancouver, I won't be able to hold office hours this week. -Percy

When you do maximum likelihood estimation, you are minimizing KL(\hat p(x) || p_\theta(x)), where \

17th, monday poster on 14th ,friday On Nov 26, 5:04 pm, Will Chang <wch...@gmail.com> wrote:

There is apparently a neat way to take the derivatives of these sigmoid-bernoullis, as explained in

> If by unimodal, you mean "one hump", then it seems that way to me > too. I see no

Hi, I heard a friend of my said that I can drop one homework. Is that the class' policy? Thanks.

> In the "Introduction to MCMC for Machine Learning", section 3.4 "The > Gibbs

> I guess that gets at the heart of my question... from the reading in > the MCMC paper, as

For some reason, I cannot access the sample reports. Here is the error message returned by my browser

Unfortunately, Soda and Soda 306 are both locked, so we're going to have recitation in the Soda

> You don't need to compute it - just optimize it, which is gotten by > moment matching. Do

You're right...this is a problem with non-identifiability of the model, which means that two

Yes, derive the parameters for the p_k's as part of the EM algorithm... In practice, because of

The upside down V in the RHS means "AND" (an upright V is "OR"). So it's S_ij

Yes. On Nov 2, 2007, at 1:59 PM, karl wrote: > > Is there a closed form solution to the M-Step

Homework 4 will be due on Tuesday, Nov. 6.

> Problem #1: > * In the formula for p(y = 0 | x), should \xi have "j" as a >

> For Question 1.3, do we need to derive the derivative of the gamma > function? Put your

At the next recitation, I'll cover Iterative Proportional Fitting, an algorithm for doing maximum

Yes, the predictive distribution depends on the hyperparameter (a) use for the prior over the

> In problem 3, is it okay to write the corresponding conjugate > density for > p(x|theta) =

Just assume that the root node has some constant distribution p (x_root) that doesn't depend on

Yes, they should be μ_1, μ_2, ..., μ_N. -Percy On Oct 7, 2007, at 7:55 PM, Nathan wrote: > >

Homework 2 is graded. It's out of 12 and has equal weight to homework 1. You can pick up yours (

At tomorrow's recitation (4:30-6 in Soda 306), I plan to cover the following: - Derive the

The links to chapter 5 and 7 on the course website are currently pointing to chapter 4 - maybe a copy

Homework 1 has been graded. You can pick yours up outside 529 Soda. We graded 4 problems (2, 3, 4, 5)

Hi there, I am wondering whether we have recitation today. If yes, is it still in Soda 306? Given