Update

Hi - I've just put a new version online. Since last time, i have done some work on the index (it's pretty good up to 'e' now). I've sorted out some problems with the description of sparse linear regression and relevance vector regression. I've also added sections on relevance vector classification; trees, forests and ferns; and a comparison to non-probabilistic classification methods. I've sorted out a lot of small problems, especially in chapters 8-14.

It's definitely going to be a little while before the book is complete (perhaps end of July?), so please if you read this, send me any comments / errata / feedback. A few people have already mailed me and its made a massive difference to the final product.

Back up

It should be back up again now. Still not completely sure this is resolved for good.

Link is down

I'm not sure why. I'm investigating.

New version

The new update is here - I've finished the first draft of the last chapter, and I've been working on the index. I've fixed about a thousand errors spread throughout the book since the last version as well. Since then someone posted a link here on a bulletin board and about 5000 people hit this site. If the download fails just wait a bit and try again. UCL's servers are pretty good and you'll definitely get a copy. Please don't torrent or repost to the internet; this is always going to be downloadable for free and I don't want half-baked versions out there. I'm updating this on a day to day basis and there's still a lot of mistakes in it.

The download is huge because I made the figures in Inkscape and it can't produce compressed PDFs. If anyone knows how to solve this problem, I'd be happy to hear about it...

Update - this is partially resolved now. I managed to compress it a bit using Quartz filters on the MAC without losing too much quality. Now its a mere 150Mb.

Hold on a second

Since a link to this book has just been posted to a prominent website, I've taken it down for a little while while I upload the most recent version that has a lot of small errors corrected. It will be back up in a couple of hours I expect. It's worth the wait...

Update

There's another update this morning. I've corrected a number of small errors throughout the manuscript and done some more work on chapter 18 which now features a roughly comprehensible discussion of latent Dirichlet allocation and the author-topic model. Thanks to people who have helped me eliminate errors.

Small edits

In the new version today, a lot of small mistakes have been eliminated from the later chapters and I've done some more work on the final chapter (chap 18). Trying to find a simple way to explain latent Dirichlet allocation. Not that easy. This chapter should reach levels of comprehensibility by the weekend (I hope).

Temporal models

Another small update last night. Lots of small errors eliminated from Chapter 10 and some work put into the temporal models chapter which was pretty sloppy.

More minor changes

Lots more tiny changes in chapters 11-14 especially but also in other places. I started making the figures for the final chapter. I've updated the main pdf and I'll put new versions of the individual sections up tomorrow with all of the changes made so far.... They're up now.

Minor updates

More minor updates today. No major changes, but a large number of tiny errors sorted out in chapters 1-11 and 15. Again, these can only be seen in the full version. I'll update the individual sections this weekend. Thank you very much to those who mailed me with comment -- you know who you are!

Minor changes

Some changes to the chapter on graphical models and the chapter on Markov random fields. Lots of minor errata corrected in the appendices on image preprocessing and optimization. I've only updated the full book as it is a bit of a bother splitting it into parts so often, so try to download that if you can.

Update

Just posted a new update. Many minor bug fixes in the first 9 chapters or so. I've also worked a bit on the regression chapter - there is an expanded discussion of non-linear regression and I think I've made the sparse regression model a little clearer. I've re-written a lot of the chapter on shape models (probably the weakest chapter, I think). It's still not great but I think it is moving in the correct direction. Thanks to all of the people who have mailed me errata so far!