2022-02-08

This is the same as 2022-02-07 except for the equation numbers in sec 7.7.3 (solving underconstrained systems (least norm estimation)).
This section omits some erroneous material from an earlier version, so I have modified the numbering at the end of this section so remaining equation numbers match previous versions of the book.

2022-02-07

2022-02-07 release

2021-11-08 v1.1

Minor updates to the version sent to MIT Press. Closes all issues to date.
Details of changes are listed at #247

camera-ready 2021-08-30

This is the version I sent to MIT Press for the first printing :)

2021-08-03

• fixed many typos
• tweaked many figures to make CMYK version look nicer

2021-06-05

• moved functional AD to vol 2
• moved adversarial examples to vol 2 (since not specific to CNN chapter and no other home for it)
  -fixed typos from github

2021-05-21

• fixed numerous typos (see closed github issues)
• fixed eqns 11.128-11.29 for weight posterior in Bayesian linear regression (surprised this one slipped through the eyes of 52k readers!)
• added short section on weakly supervised learning to the very end of ch. 19
• moved the section on information criterion (5.1.7) to after the section on model selection (5.2.2)
• removed section on probability integral transform (2.8.8) since it will be covered in vol 2

2021-05-06

This is the version I sent to MIT Press for "final" copy editing.

2021-04-21

This version was sent to MIT Press, but they sent it back with some changes I need to make, so it is still a draft (pre-publication) version.

2021-03-08 (MAJOR UPDATE)

Changelog

• Moved some of the more advanced material on Bayesian inference (eg VI, MCMC) to vol 2.
• Moved material from the appendices (except for notation) into the main body, integrating some of the material to improve the flow
• Shortened the coverage of Bayesian and frequentist statistics to make more accessible for an intro ML audience
• Reduced page count from 930 to ~825 ("missing mass" will go into vol 2)
• Fixed typos
• Most of the code which is linked to in the figures now runs in colab