harvard-ml-courses / cs181-textbook Goto Github PK

Chapter 8, page 109:

"In this setup, we say that information from A to C is being ‘blocked’ by the unobserved variable. B. Thus A and C are currently conditionally independent. However, once we’ve observed B, as shown in Figure 8.17, the information flow changes.
Now, information is flowing between A and C through the observed variable B, and thus A and C are now conditionally independent. This phenomenon, where the observation of the random variable in the middle creates conditional dependence is known as explaining away."

Should the bolded text be "conditionally dependent" instead of "conditionally independent"?

Chapter 7, page 100: I think the d after lambda should be a subscript based on the formula description. Attached is a screenshot.

In Derivation 2.8.1, it should say, "random variable of $(\mathbf{x}, y)$ pairs sampled from a distribution $F$," rather than "sample"

In Ch. 3, on pg. 29 there is a typo: "problem-" in the sentence "In this chapter, we’re going to think about a different problem- one where our target output is discrete-valued."

List of typos in Ch. 6:

• Incorrect $x''$ in Equation 6.1 on pg. 78, replace with $x'$.
• Disjoint box on pg. 81.
• Period after "different initializations will produce different results" on pg. 84.
• On pg. 89, "Notice also that we now have many layers of clustering: if we’re only only interested in clusters whose elements are at least $k$ units away from each other, we can ‘cut’ the dendrogram at that height and examine all the clusters that exist below that cut point." The word "only" is repeated. Also, this point is only true if we use the min-linkage criterion, which is not immediately clear since min-linkage criterion is brought up in a later paragraph.

Chapter 2, page 7: unwieldy is spelled incorrectly in the section on merging of bias (extra L).

Chapter 7, page 97: first two lines of simplification are identical, so one can be removed. Screenshot attached.

Chapter 7, page 93: In rewriting equation 72, min_w is applied on the left side of the equation but not the right.

Chapter 2, page 4: w!e're is spelled incorrectly in the introduction/first paragraph, in line 3.

"they can be written on their own as margin probabiilties"

At the start of section 4.4.3, the first paragraph on backprop reads, "weights that lie in the middle of our networ." rather than "network"

Chapter 2, page 10: parameter is spelled incorrectly in the section on matrix differentiation.

In the first paragraph, "Loss" is unnecessarily capitalized; i.e., "This is sometimes referred to as \textit{L1 Loss}" should be "This is sometimes referred to as \textit{L1 loss}."

Activation is spelled "Activaiton" in the second sentence of the page.

On chapter 2, pg. 8, the reader note suggests:

Although our input data points $\textbf{x}$ can take on multiple dimensions, our output data $y$ is always a 1-dimensional real number when dealing with regression problems.

However, this is not true in the case of multivariate regression, right?

Chapter 2, page 27: There is an integral sign missing in equation 2.14.

Chapter 7, page 96 (7.3.2, the equation between (7.2) and (7.3))

Is the 1 / M suppose to be 1 / N

Chapter 2, page 20: the Normal prior in Bayesian Regularization Deviation is written as when it should be .

Chapter 3, page 35: the distinction between batch gradient descent and stochastic gradient descent is not clear to me, could someone provide more explanation.