Search :
Logo

Cavendish Laboratory, Cambridge

The Inference Group.

David MacKay's group works on machine learning and information theory. Current projects involve neural networks, automated Go playing, the design of record-breaking error-correcting codes and quantum error-correcting codes, and the construction of human-computer interfaces that make use of adaptive language models.

Upcoming talks

More group meetings

People

Cavendish Lab Other departments
Faculty David MacKay
Alan Blackwell
Zoubin Ghahramani
Research
Associates
Per Ola Kristensson
Patrick Welche
Alan Lawrence
Emli-Mari Nel
John Winn
Graduate
students
Keith Vertanen
Christian Steinruecken
Philip Sterne
Carl Scheffler
Philipp Hennig
Recently
departed
people
Tamara Broderick
Ryan Adams
Oliver Stegle
Hanna Wallach
David Stern
Phil Cowans
Miriam Backens

Will Zou
See also the Former Members' page

Projects

Dasher logo
opengazer
tapir
DIMPY: Physical quantities in python and Sage
talks.cam

Software

John Skilling's BayeSys

music access for the disabled
bayesimage BayesImage by Venkat Ramesh

Teaching

Teaching by David, Steve, and Sanjoy

Wikis

Inference wiki | Internal wiki | Dasher Wiki

We enjoy collaborations with Cambridge Centre for Energy Studies, the Computational and Biological Learning Lab, CUED, TCM, Cavendish Astrophysics, Microsoft Research, and the Rainbow group.


What's old: Special One Day Meeting | wiki | slides


Logo

Inference group projects

The Inference Group works on machine learning and information theory. Current projects involve computation in neural networks, the modelling of DNA microarray data, the design of record-breaking error-correcting codes, quantum error correcting codes, the teaching of Physics, and the construction of information-efficient human-computer interfaces.

Eye-tracking

DNA microarray

Software packages

Dasher text entry interface   
Rate 1/4 code, channel signal to noise 0.84 -> decoded transmission, all errors corrected
Before decoding After decoding
qeccmove3
eye-tracker dasher experiments xybernaut wearable tcl dasher ipaq running dasher japanese dasher


Some contact details

Christian Steinruecken, Room 514, extension 37278
Keith Vertanen, Room 514, extension 37278
Dr Piotr Zielinski, Room 514, extension 37278
Philip Sterne, Room 515, extension 37460
Ryan Adams, Room 516, extension 37238
Oliver Stegle, Room 516, extension 37238
Carl Scheffler (from April 1), Room 517, extension 64164
Prof David MacKay, Room 518, extension 39852

Logo

Group meetings

The Machine Learning group at CUED and the Inference Group at the Cavendish Laboratory hold joint weekly group meetings, usually on Wednesdays at 2pm (but sometimes at other times), and alternating venues between Engineering and the Physics department.

To make sure that you get automated emails about all these joint group meetings, please visit talks.cam and click on "Send you e-mail reminders".

Future talks | Past talks


How this list of talks is generated

All inference group activities in one table


Logo
 
 

Short talks:

David Stern
Katherine Heller
Anitha Kannan
Ryan Turner
Philip Sterne
Finale Doshi
Sean Holden
Andrew Fitzgibbon
Miguel Lázaro Gredilla
Martin Szummer
Blaise Thomson
Ryan Adams
Chris Bishop
Frederik Eaton
Yunus Saatci
Phil Cowans
Sinead Williamson
Ricardo Silva
Andrew Naish
Philipp Hennig
Arik Azran
Philip Dawid
Christian Steinrücken
Andreas Vlachos
 
Ed Snelson
Joaquin Quinonero Candela

About talks.cam

Are you a seminar organizer? Do you maintain a webpage that lists upcoming talks, seminars, or journal clubs? Then talks.cam is for you.

At talks.cam, you can create your own series (one or more series, as you wish); you, the other managers of your series, and the speakers themselves can enter the content into the series database at talks.cam; and talks.cam automatically serves up live, up-to-date content into your webpages, using listings -- not only your webpages, but into lots of other people's webpages too, because your series content will now be automatically advertised to everyone who is interested. (Other people can add your series to their listings.) talks.cam integrates seamlessly into your existing websites (See this page for an example where the talk content comes from talks.cam.) Or if you would like to have a series website created for you by talks.cam, we can do that too (as shown by this example).

If this isn't enough to convince you that you should switch over your seminar list management to talks.cam, here are some additional features:

  1. talks.cam will automate the task of communicating with the speaker, asking for his/her title and abstract, and ensuring that all websites are immediately updated when that information is provided.
  2. talks.cam provides personalized browsing and searching of the University-wide database.
  3. in the next planned version, talks.cam will also handle the task of sending email announcements of talks to people who want to receive emails.

Examples

Pages in which embedded content is delivered live by talks.cam
Pages entirely generated by talks.cam
Website managers may be interested in viewing the source of the above pages to see how simple they are.


How our talks listings are generated

Our talks listings (on for example our home page) are generated automatically by a talks-listing service created by Phil Cowans, in collaboration with Alan Blackwell, Duncan Simpson, and David MacKay, with support from Cambridge University Research Exchange (CURE).

To get the content included, all that I need to put in the webpage is this

 <script language="javascript"
src="http://talks.cam.ac.uk/talks/external/embed_feed.php?id=1"></script>

plus, to get the `look' delivered too, I add the following rules to my style sheet:

.talklist_date{ background-color: #dddddd; font-weight: bold; }
.talklist_header{ font-weight: bold; }
.talklist_entry{ }
.talklist_block{ clear: right; }
.talklist_block_small{ font-size: 0.8em; }
.talklist_date_small{ background-color: #dddddd; font-weight: bold; }
.talklist_header_small{ font-weight: bold; }

The content of our series database is managed by our talk series organizer through a nice web interface.

To get the content in the smaller box on the group home page, I used:

 <script language="javascript"
src="http://talks.cam.ac.uk/talks/external/embed_feed.php?id=30&suffix=small"></script>

Alternatively you can use this style sheet
.talklist_date{ border-top-style: dashed; border-bottom-style: dashed;
border-left-style: none; border-right-style: none; border-width: 1px;
border-color: #cccccc; margin-top: 8px; margin-bottom: 8px; margin-left:
16px; }
.talklist_week{ border-top-style: solid; border-bottom-style: solid;
border-left-style: none; border-right-style: none; border-width: 2px;
border-color: #cccccc; margin-top: 8px; margin-bottom: 8px; font-weight:
bold; }
.talklist_header{ font-weight: bold; }
.talklist_speaker{ font-style: italic; }
.talklist_entry{ margin-bottom: 16px; margin-left: 32px; }
which will make it look like this.

Here is Mike Roses' style-sheet cocktail:


.talklist_entry{ }
.talklist_block{ clear: right; }
.talklist_block_small{ font-size: 0.8em; }
.talklist_date_small{ background-color: #dddddd; font-weight: bold; }
.talklist_header_small{ font-weight: bold; }
.talklist_date{ border-top-style: dashed; border-bottom-style: dashed;
border-le
ft-style: none; border-right-style: none; border-width: 1px; border-color:
#cccc
cc; margin-top: 8px; margin-bottom: 8px; margin-left: 16px; font-weight:
bold; }
.talklist_week{ border-top-style: solid; border-bottom-style: solid;
border-left
-style: none; border-right-style: none; border-width: 2px; border-color:
#cccccc
; margin-top: 8px; margin-bottom: 8px; font-weight: bold; }
.talklist_header{ font-weight: bold; }
.talklist_speaker{ font-style: italic; }
.talklist_entry{ margin-bottom: 16px; margin-left: 32px; }



Logo

Information about the group


Bibliography files of MacKay group members

  • bib entries for all publications submitted or published while anyone is in the group should be deposited in files in this directory (/home/ftp/pub/www/is/bibs/).
  • One paper copy of the final printed paper should be given to DJCM and two copies to the RA secretary so that the paper is included in the RA database. If a large delay is expected between submission and publication then please give me a copy of the submitted paper.
  • The RA database form should be filled in and given to the RA secretary, and the RA database number should be included in the .bib entry in this directory.
  • If you wish, put postscript files for papers on the web server either under your own home page or in the group postscript directory.

Recommended format

Please use the bibtex style and include all relevant information in the .bib entry.
  • In the case of papers published in conference proceedings, include all authors names with initials, full title, page numbers (start-end), editor of proceedings, title of proceedings, publisher, address of publisher.
  • In an `annote' environment, include the information you put in the RA database form regarding date submitted, date accepted, date published, collaborating institutes.
  • Also include the RA database number.
  • If it is easy to do so, include the abstract too.
Example entry:
 AUTHOR		={L. Gavard and H. K. D. H. Bhadeshia and
		  D. J. C.  MacKay and S. Suzuki},
 TITLE		={Bayesian Neural Network Model for Austenite
		  Formation in Steels},
 journal        ={Materials Science and Technology},
 vol            =12,
 pages          ={453-463},
 YEAR           =1996,
 ANNOTE         ="Date submitted: May 1995; Date accepted: n/k 1995;
                  Collaborating institutes:
		  Cambridge University Department of Metallurgy and
		  Materials Science. MRAO reprint number 1941."}

David MacKay <mackay@mrao.cam.ac.uk>
Last modified: Tue Jul 17 12:55:43 2001

PhD Theses submitted

TitleExaminersExamination date
Mark Gibbs Bayesian Gaussian Processes for Regression and Classification Steve Gull(?) and Chris Williams ~ Sep 3 1997
Matthew Davey Error-Correction using Low-Density Parity-Check Codes. Steve Gull and Paddy Farrell Friday 17th December, 1999
Simon Wilson Applications of cyclic belief propagation Ross Anderson and Paddy Farrell May 29th 2000
James Miskin Ensemble Learning for Independent Component Analysis. Steve Gull and Stephen Roberts ~ April 19, 2001
David Ward Adaptive Computer Interfaces Tadashi Tokieda and Peter Littlewood ~ 18 December 2001
John Winn Variational Message Passing and its Applications Zoubin Ghahramani and David Spiegelhalter November 2003 (Submitted September 2003)
Ed Ratzer Error-correction on non-standard communication channels November 2003 (Submitted September 2003)
Seb Wills Computation with Spiking Neurons Carlos Brody and Graeme Mitchison November 2004 (Submitted October 1 2004)
Phil Cowans Probabilistic Document Modelling Zoubin Ghahramani and Steve Renals April 28th 2006 (Examined 9 May 06)
David Stern Modelling Uncertainty in the Game of Go Stephen Eglen and Yee-Whye Teh Examined Nov 28 2007
Hanna Wallach Structured Topic Models for Language Zoubin Ghahramani and Yee-Whye Teh Submitted 9 June 2008, Examined June 22 2008
Ryan Adams Nonparametric Bayesian Inference for Probability Densities and Point Processes Carl Edward Rasmussen and Neil Lawrence Submitted 30th April 2009, Examined 1 June 2009
Keith Vertanen Efficient Correction Interfaces for Speech Recognition John Bridle and Phil Woodland Submitted 30th April 2009, Examined 1 June 2009
Oliver Stegle Probabilistic models in computational biology Wolfgang Huber and Brendan Frey Submitted 30th April 2009, Examined 1 June 2009
Philip Hennig Approximate Inference in Graphical Models Zoubin Ghahramani and Neil Lawrence Submitted 14 November 2010, Examined 11 January 2011
Philip Sterne Distributed Associative Memory Carl Rasmussen and David Willshaw Examined 21 January 2011; Approved 19 April 2011.
Carl Scheffler Applied Bayesian Inference: Natural Language Modelling and Visual Feature Tracking Phil Woodland and Yee-Whye Teh Submitted 20 December 2010, Examined ... February 2011
Christian Steinrücken

MPhil Theses submitted

TitleExaminersExamination date
John Barry Polyphonic Music Transcription using Independent Component Analysis Steve Gull and Mark Plumbley Submitted 25.4.2003
Tamara Broderick

Inference group in the news


(Original article by Jon Pendergast (2002))
WeighNS
back
(The story also made it into die Zeit)

Cambridge Evening News, 18.Dec.00

Logo

Resources for the IS Group

This page contains a short list of the available online and other resources. This page will offer you the passwords if you have JavaScript enabled on your browser.

Information for visitors

How to get a bank account

Barclays offers a free account with an ATM card and internet banking. They have a branch at 15 Benet street, CB2 3PZ in Cambridge, at the market square close to Kings College.

Setting up the account takes about 40 minutes, so it is best to call them and make an appointment in advance.

Call Miss Carly Taylor at +44 (0)1223-542203 (or try +44(0)1223 542139). To open the account you will need:

  • Valid Identification (such as a passport)
  • Proof of your address in the UK
  • Proof of your address for the past 3 years (like a document with your home address)
  • Address of your employer or host (Cambridge University)
they will ask a ton of questions and check your british debt record.

When K.S. opened her account, it helped to have a letter of introduction from the university department (a letter from DJCM was good enough). This letter should have the name exactly as it appears in the passport, and both the addresses of the visitor (last address and current address).



| MSR seminars |

All inference group activities - talks, lunches, journal clubs, department seminars


Travel insurance

From 1 August 2007 the University is now providing free comprehensive travel insurance for employees travelling abroad on University business as long as the trip has been registered.

Trips should be registered online.

Inference Group Research Grants / Accounts

Accounts: x37412, Mr Matthew Burgess mb476@phy.cam

AEGIS Research Grant - RG50866 KMZA/061 with EC FP7 CP - Collaborative Projects

SET UP AS PER CONTRACT - EURO 561,721.65 AT EXCHANGE RATE 1.34 = STERLING ?419,195.26 - SUBJECT TO CHANGE WHEN ACTUAL EXCHANGE RATE IS KNOWN RG Number: 50866 UFS Project: KMZA/061 pFACT ID: 7910 PI Name: MacKay, DJC Department: Physics - Theory of Condensed Matter Sponsor: EC FP7 CP ? Collaborative Projects Sponsor Ref: 224348 Start Date: 2008-09-01 End Date: 2012-02-29 Full Title: AEGIS - Open Accessibility Everywhere: Groundwork, Infrastructure, Standards
	* Payments are expressed in person-months
	* Month 1 = Now, i.e.  1st September 2008

Period         Person-months
------	       -------------
M 1-12            26
M13-24            30
M25-36            30
M37-44/end	  11.90
----------     -------------
Total             97.90

Dasher donations

KM.KMGB.EGDT - (24.7K on Thu 23/8/07)

Nokia donations

KM.KMGC.HBDT - (18K on Thu 23/8/07)

Lab consumables

KM.KMGB.AAAA - (1K on Thu 23/8/07)

MacKay General Fund

KM.KMGB.EFBQ - (20.9K on Thu 23/8/07)
Per Ola K RG52032 with EC FP7 MC IEF - Intra-European Fellowships for Career Development. STERLING 120,317.90
started 1 July 2008.
PIEF-GA-2008-220793

microsoft

RG47219 (studentship for Christian S) activated Wed Jan 3 12:36:55 2007 Wed Jan 3 12:36:55 2007 RG NUMBER: 47219 pFACT ID: 3399 UFS PROJECT: The project number has now been changed From KKZA/047 - Astrophysics To KMZA/036 - Theory of condensed Matter Physics PI NAME: MacKay, Doctor David JC DEPARTMENT: Physics - Astrophysics SPONSOR: Microsoft Research Limited (MRL) SPONSOR REF: 2007-011 START DATE: 01/10/2006 END DATE: 30/09/2009 FULL TITLE: MRL Studentship for Christian Steinruecken: "Learning to recognise hierarchies of objects and scenes"

Nokia

KMGC.HBDT

SAMSUNG: grant for work on communication and tracking RG47069 RG47069 KKZA/044 (transferred 2000 equipment money to AP)

Distributed phase codes, and Beyond Dasher

RG43517 with Gatsby Charitable Foundation RG 43517 KMZA/051
Please note I have reduced the consumables budget by ?284 to keep in line with the +Gatsby allocation for the 1st year of ?98,374. Jamie Horsley Applications and Awards Manager Research Services Division 16 Mill Lane Cambridge CB2 1SB T: +44 (0) 1223 765018 F: +44 (0) 1223 332988
RG NUMBER:	43517
UFS PROJECT:	was KKZA/038  now  RG 43517 KMZA/051
PI NAME:	MacKay, Prof D J C
DEPARTMENT:	Physics - Astrophysics
SPONSOR:	Gatsby Charitable Foundation
SPONSOR REF:	GAT2785
START DATE:	01/01/2005
END DATE:	31/12/2007
FULL TITLE:	Distributed Phase Codes and 'Beyond Dasher' projects

Direct tel:	64191	Fax: 01223 332988
E-mail	karen.hewitt@rsd.cam.ac.uk
	

Receiving donations

The IBAN number is: IBANGB96BARC20171910921084 The Swift code is: BARCGB22 The IBAN number includes the bank sort code and account number but just for completeness these are: Sort Code: 20-17-19 A/c No.: 10921084 When the transfer is made could you ask for the Reference/Comment " Physics KM (MacKay donation) " to be added so that we can pick the item up in the listings and transfer it to your account?

Martin King donation to support Disabled-related research

The sterling amount is 15,523.13. account is KK.KKBO.HABQ

5000USD (3129.89UKP) by the Nine Tuna Foundation to support rehabilitative engineering projects. This will shortly be credited to KK.KKBO.HABQ. The Martin King funds are now in that account. This has changed to KK.KKBO.EFBQ which also includes the 15K from MS.

Funds from RG32843 and RG32054 (CASE awards for Ed and Seb)

These funds were transferred to the AP CASE account (KK.KKBM.GBAA/B) Tue 23/9/03.

Microsoft Studentship to support David Stern

(supervised by Dr Mackay and co-supervised by Thore Graepel of Microsoft Research) reference number RG 38193. Microsoft Studentship for David Stern - Machine Learning For Go
        UFS Project/Task        KKZA/024
 RG38193
 Karen E M Hewitt
Awards Administrator
Research Finance Group
Research Collaboration Office
16 Mill Lane
Cambridge
CB2 1SB

Tel No: 01223 764191
Fax No: 01223 332988
email Karen.Hewitt@rsd.cam.ac.uk

IBM award

was CRA 0171, is now KKCE/HBDT. [bal Fri 28/6/02 7007]

John Winn

Further to your enquiry regarding possibly funding from Microsoft for conference attendance, please see the message below from Sharon Renault. The money is in account number KKBM GBAA - Emma knows about it if you want to query anything. As the contract for funding is between students and funding bodies, I haven't seen any correspondence from Microsoft. I understand from you, John, that you also haven't seen any paperwork. >I imagine it will be the same next year. You should check with Robert next >year. > >Regards > >Sharon Renaut

Phil

2,400 (1,000 of which is from MS) available for travel, etc, in AP CASE account KKBM GBAB. KKZA/005

Ed

Award No. RG32054, "Error Correcting Communications (CASE)

Seb

British Telecommunications plc ML845741: "Industrial CASE - Mr S Wills:
Inference and learning using simple hardware"

    Award No.       RG32843
    Project No.    KKZA/005
    Task No.              1

The award has an announced period of 1 October 2000 to 30 September 2003.
The award provides Industrial CASE support for Mr S A Wills for 36 months.

Gatsby charitable foundation: Senior Research Fellowship for DJCM

GAT 2277 (starting Sep 2002)
 RG 36363
 KKZA/018

COGAIN

RG40404
UFS project: KKZA/034
Ec FP6 (NoE - Partner) SPONSOR REF: 511598 START DATE: 01/08/2004 END DATE: 31/07/2007 FULL TITLE: COGAIN: Communication by gaze interaction Administrator: karen.hewitt@rsd.cam.ac.uk x64191

The Dasher Project

 RG NUMBER	36325
 UFS Project/Task nos.	KKZA/019
contact karen.hewitt@rsd.cam.ac.uk if any questions

Gatsby charitable foundation: Associate Lectureship for Sanjoy

GAT 2251 (starting 1 sep 2002) KKZA/013

Gatsby charitable foundation computer support

RG 21564. (closed)

Gatsby charitable foundation - Teaching Physics

was RG 26397. Now RG: 36075

EPSRC Withers/MacKay

EPSRC grant number L 10239. Departmental ref: RG 23369.

Inferential sciences general fund

Mon 26/2/07: The general donations account is now KMGB EGDT and there is currently 20,914.32 in there. I will ask Tina to credit this account with the 320.00. [(Money from Parsons, Piresearch) -- was AL01 - 0007, KKCJ GAAA Nippon steel 2000.00]

This page is restricted to Cambridge access

The Inference Group Computer System

a computer user

How to install packages on Debian machines

From: Phil Cowans

As you may know, the new machine, eigg, is running Debian 3.0 rather
than RedHat Linux. Here are a few notes on installing new packages on
a Debian machine. Debian uses the 'apt' system for package
management. This is has the advantage over RPM of being able to
install dependencies automatically, which normally makes installing
new packages very easy.

Here are the commands needed for some common tasks:

1. Use 'apt-cache search' to give a list of available packages
matching a given search string, for example:

kern:~$ apt-cache search gnumeric
gnumeric - A GNOME spreadsheet application
gnumeric-doc - Documentation for Gnumeric, the GNOME spreadsheet application
....

2. Use 'apt-cache show' to give a detailed description of a package, eg:

apt-cache show gnumeric-doc

The following commands all require you to be root in order to carry
them out:

3. Use 'apt-get update' to refresh the list of available
packages. This should really be done before upgrading or installing
anything.

4. Use 'apt-get upgrade ' to upgrade a given package to the latest available version. If is omitted then all installed packages will be upgraded.

Note that security patches count as upgrades in Debian, and that the stable release (which we have installed) is very rarely changed for any other reason. It is therefore wise to do 'apt-get update' followed by 'apt-get upgrade' every time a security patch is released. Security patches are announced on the debian-security-announce mailing list, which you can subscribe to here.

5. Use 'apt-get install ' to install the specified package and anything that the package depends on. You will be asked to confirm if this means installing or changing a lot of things. You will also be asked questions to configure the new package if necessary.

ps. I'll leave it as an exercise to the reader to determine the function of the command 'apt-get moo' (no need to be root) :-)


Make your own website

For tips, see this website.

Computer and Software Resources


Photo albums

COGAIN 2005

Scientific Organisations


Logo

Want to join us?

Graduate places

Applications are handled through the "Mott" system in the Cavendish. Please apply to the Cavendish and also get in touch with David MacKay. Candidates are interviewed by a panel of `wise men' from the Mott building who decide to whom offers can be made. Each year typically one PhD place is available. British students in the group have typically been funded by CASE awards or EPSRC awards. Non-British students have been funded by Gates studentships, by Microsoft research studentships, and by other scholarships, especially ones awarded by the Cambridge Trusts. Questions about joining the group?

Postdoctoral people

A helpful list of funding sources for aspiring postdocs, compiled by Robert MacKay

If you can raise your own money, get in touch!

We are hoping to have funding (from Oct 2002) for a postdoc to work in biological data-modelling, looking for patterns in DNA microarray data. Total competence in Bayesian data modelling will be required. Contact David MacKay if interested.


Figures (pounds) as of Thu 20/3/03, for admission September 2003

        Maintenance   Overhead (@70%) Travel etc    University Fee  College Fee    Total
Year 1  12,000         8,400          1,100         3000            1,950          26,450
Year 2  13,500         9,450          1,150         3100            2,020          29,220
Year 3  15,000         10,500         1,200         3210            2,090          32,000
                                                                                 ---------
                                                                           Total   87,670

Notes

1. The maintenance is at the CASE rate (EPSRC stipends are to rise sharply in the next couple of years)
2. Travel etc. is equivalent to the RTSG of an EPSRC studentship
3. University fees are 19,546 GB pounds per year for non-european union students

Got Questions? (Implemented using MetaFAQ) metafaq

If you are a Physics student: If you are anyone: If you are in the group:
ask about ask about ask about

About this server, and nearby sites

The main organization on this server is the Inference Group, led by David MacKay.

Other organizations who have websites here are:

This web server: | Old Search index | Statistics | Everything stats | mirroring |

The new name for this webserver: www.inference.phy.cam.ac.uk/ - November 2001.
(Formerly known as wol.ra.phy)
(New direct route is coll.ra.phy, deprecated)


Nearby groups and web servers

Logo

Inference group Former members

Theses submitted

Other old links


The Inference Group is supported by the Gatsby Foundation
and by a partnership award from IBM Zurich Research Laboratory
David J.C. MacKay
Site last modified Thu Oct 20 16:12:27 BST 2011