This page lists errors or typos appearing in the second edition and
printing 1 of the second edition of the
book Foundations of
Machine Learning as well as their corresponding
corrections. We are grateful to all readers who kindly bring those to
our attention.
- p. 34, Equation 3.19: the set $X$ should be in calligraphic
font.
- p. 35, proof of Theorem 3.7, 3rd line: it should be r, not r^2.
- p. 41, third line: subscript $2$ near the end of the line is a
typo and should be instead $\cG_2$; bottom line: = d -1 should be <= d - 1.
- p. 44, sentence after Equation 3.32: X is missing \overline.
- p. 65, display equation (4.4) and 4 lines below: the Rademacher complexity term should be in \mathfrak font.
- p. 93, displayed equation in statement of Lemma 5.7: the
extra parenthesis after $x_i$ should be removed;
bottom line: the first \Phi_m should be replaced with \Phi_i.
- p. 101, Exercise 5.4 (b): \alpha_2 that minimizes -> \alpha_2 that maximizes.
- p. 107, Figure 6.2: poor rendering in this printing.
- p.108, Equation 6.2: sum should be from 1 to m, not from 1 to n.
- p.158, sentence after corollary 7.6: in the additional term, it
should be $2/\rho$ and not $2/\delta$.
- p.170, exercise 7.2: the inequality should be $\Phi(x) \leq 1$,
not $\Phi(x) \geq 1$.
- p. 196, Theorem 8.12: it should not be square-root of the norm-2
of the vector of losses, but just the norm-2 of the vector of losses.
- p. 270, statement of Theorem 11.3: L(x, y) should be L(h(x), y)
and similarly with the empirical loss term; third line from the
bottom: empirical Rademacher complexity notation is missing \widehat;
third line in the proof of the theorem: supremum over $h$ in $H$
missing in both expectations.
- p. 271, definition 11.4: The set $\set{z_1, \ldots, z_m}$
should be included in $\cZ$ not $\cX$.
- p. 274, definition 11.9: The set $\set{z_1, \ldots, z_m}$
should be included in $\cZ$ not $\cX$.
- p. 410, in definition of group norm: it should be M_j and not M_i.
- p. 439, proof of Theorem D.3: the third and fourth lines are repeated;
sentence before Equation D.7: the choice should be q = 1 - p +
\epsilon, not p - \epsilon; statement of Theorem D.4: 1/p - 1 should
be min{1, 1/p - 1}; proof of Theorem D.4: log(1 + x) < x should be
log(1 + x) <= x.
- p. 440, proof of Theorem D.7: Sum_i V_k should be sum_i V_i; bold
V_m should simply be V_m.
- p. 445, Equation D.26: it should be r_i, not r_j.
- p. 451, Equation E.8: p_1(x) in the last term should be replaced with p_2(x).
- Index: "Talagrand's lemma" is missing.