A First Course in Machine Learning errata

Simon Rogers Mark Girolami

April 3, 2014
p9, para1, x
1
w
1
p18, Comment 1.7, the second row of the matrix multiplication example is erroneously a copy of the rst.
The nal matrix should be:
_
a
11
b
11
+ a
12
b
21
a
11
b
12
+ a
12
b
22
a
11
b
13
+ a
12
b
23
a
21
b
11
+ a
22
b
21
a
21
b
12
+ a
22
b
22
a
21
b
13
+ a
22
b
23
_
p15, section1.3, f(x, s
1
, . . . , s
5
; w
0
, . . . , w
6
)f(x, s
1
, . . . , s
8
; w
0
, . . . , w
9
)
p52, Equation 2.13 and the equation block below it should read (an X had been replaced with an x!):
var{X} = E
p(x)
_
(X E
p(x)
{X})
2
_
var{X} = E
p(x)
_
(X E
p(x)
{X})
2
_
= E
p(x)
_
X
2
2XE
p(x)
{X} +E
p(x)
{X}
2
_
= E
p(x)
_
X
2
_
2E
p(x)
{X} E
p(x)
{X} +E
p(x)
{X}
2
p34, second line, reguarised regularised
p60, rst equation, =
p80, . . . is the negative inverse of something called the Fisher information matrix . . . is the inverse of
something called the Fisher information matrix
p98, para2, This is much lower than that for r = 5This is much lower than that for r = 0.5.
p103, para1, . . . will gave a. . . will give a
p103, equation 3.6. The power of the (1 r) term is missing a 1:
p(r|y
N
) =
( + + N)
( + y
N
)( + N y
N
)
r
+y
N
1
(1 r)
+Ny
N
1
p123, below Equation 3-16. . . . doesnt depend on w
new
and so. . . . . . doesnt depend on x
new
and
so. . .
p130, para3, . . . Gaussian density (N(x
T
new
w,
2
) is. . .
p137, nal reference, mulitple multiple . . . Gaussian density (N(x
T
new
w,
2
)) is. . .
p137, ref[1], mcmcMCMC
p143, there is an issue with the normalising constant. As:
Z
1
= p(t|X,
2
)
then the posterior should be:
p(w|X, t,
2
) = Zg(w; X, t,
2
).
p145, In the second line of the rst set of equations, the log operates on both fractions. I.e.:
=
N

n=1
log
_
_
1
1 + exp(w
T
x
n
)
_
t
n
_
exp(w
T
x
n
)
1 + exp(w
T
x
n
)
_
1t
n
_
1
p168, ref[10], Alex Smola was an editor not a co-author of this paper.
p141 , para2, prior density, p(w) exercises.prior density, p(w).
p154, para4, the printing process seems to have removed the white areas from the dartboard. These should
be the two large areas in each segment.
p188, in comment 5.1, the b in the penultimate line should be an a.
p233, last paragraph, However, it is comes with However, it comes with
p242, third paragraph, . . . must be orthogonal to w
1
(w
T
1
x
2
= 0) must be orthogonal to w
1
(w
T
1
w
2
= 0)
p244, comment7.1 eigen valueseigenvalues
p255, Equation block 7.11 is formatted badly and misses a db it should be:
E
p(a)p(b)
{f(a)f(b)} =
__
p(a)p(b)f(a)f(b) da db
=
__
p(a)f(a) da p(b)f(b) db
=
_
E
p(a)
{f(a)}p(b)f(b) db
= E
p(a)
{f(a)}E
p(b)
{f(b)}
p255, E
Q
x
n
(x
n
)Q
w
m
(w
m
)
_
x
T
m
w
m
w
T
m
x
n
_
E
Q
x
n
(x
n
)Q
w
m
(w
m
)
_
x
T
n
w
m
w
T
m
x
n
_
Acknowledgements: Hat-tip to the observant individuals who have found these: Gavin Cawley, Luisa Pinto,
Andy OHanley, Joanathan Balkind, Mark Mitterdorfer, Charalampos Chanialidis, Juan F. Saa . . .
2