NeuroCOLT

Neural Networks and Computational Learning Theory

 

About NeuroCOLT

Papers Archive

Research Areas

Partners

Coordinator

Events

info@neurocolt.org

 

NeuroCOLT Workshop
on
Generalisation Bounds Less than 0.5
Windsor, 29 April - 2 May 2002
Cumberland Lodge

Summary
The workshop draws together leading researchers in the drive to give non-trivial bounds on the generalisation of classifiers trained on real-world data. There will also be space for contributed presentations on related topics in the analysis of learning systems. The workshop will be the final meeting of the NeuroCOLT 2 project.

HOW TO GET TO CUMBERLAND LODGE

Workshop Schedule
Monday April 29th:

12.30 - 2.00pm

Lunch at Cumberland Lodge

2.00 - 2.15pm

Introduction/Overview

2.15 - 3.00pm

John Langford
"Basic stuff, Occam's Razor Bound"

3.00pm - 3.30pm

David McAllester
"PAC Bayes Bounds"

3.30pm - 4.00pm

Coffee Break

4.00pm - 5.00pm

John Langford
"Stochastic Neural Networks"

7.15pm Dinner

Tuesday April 30th:

9.00 - 10.15am

Matthias Seeger
"PAC Bayes for Bayesian Architectures"

10.15 - 10.45am

Coffee Break

10.45 - 11.45am

Thore Graepel
"PAC-Bayesian Bounds for Kernel Classifiers based on Compression and Margin"

11.45 - 12.45pm

John Langford
"Combining Holdout and Training Set Based Procedures for Sample Complexity Bounds"

12.45 - 2.00pm

Lunch

2.00 - 3.15pm

Peter Grunwald
"Bayes, MDL, SRM and \sqrt{n} in Classification"

3.15 - 3.45pm

Coffee Break

4.00pm - 5.00pm

Ashutosh Garg
"On generalization bounds, projection profile and margin distribution"

5.00pm Walk to Valley Gardens or Windsor
7.15pm Dinner

Wednesday May 1st:

9.00 - 10.15am

David McAllester
"Avoiding Union Bounds"

10.15 - 10.45am

Coffee Break

10.45 - 11.45am

John Shawe-Taylor
"Compression Bounds and the Set Covering Machine"

11.45 - 12.45pm

Ralf Herbrich and Bob Williamson
"Algorithmic Luckiness"

12.45 - 2.00pm

Lunch

2.00 - 3.15pm

Gert Lanckriet
"The Robust Minimax Probability Machine"

3.15 - 3.45pm

Coffee Break

4.00pm - 5.00pm

Shai Ben-David
"A Theoretical Framework for Learning from a Pool of Disparate Data Sources "

5.00 - 6.00pm

Volodya Vovk
"On-line Confidence Machines are well-calibrated"

7.15pm Dinner

Thursday May 2nd:

9.00 - 10.15am

Andreas Maurer
"Learning to Learn"

10.15 - 10.45am

Coffee Break

10.45 - 11.45am

Jose Balcazar
"Training SVMs by random sampling"

11.45 - 12.45pm

Neil Lawrence
"A Compression Scheme for Gaussian Process Classifiers: The Informative Vector Machine"

12.45 - 2.00pm

Lunch



Accommodation:
Residential accommodation for the workshop is being provided at Cumberland Lodge but is limited. The address for Cumberland Lodge is:

Cumberland Lodge, The Great Park Windsor Berkshire SL4 2HP Telephone: 01784 432316 Fax: 01784 438507

Day accommodation is less restricted, though there will be some practical  limitations which we may need to look into if demand is really high.

Travel Instructions:

BY RAIL - there are fast trains from London Waterloo to Egham (not Windsor). The journey time is approximately 35 minutes. From Reading and the west there is a direct service Reading - Egham - Waterloo. Egham station is 3 miles from Cumberland Lodge.

BY AIR - Heathrow Airport is about 20 minutes by car - it is best to book a taxi from the list below. From Gatwick there is the choice of the train into London and then via Waterloo to Egham or the airport link coach to Heathrow and from there by taxi.

TAXIS - there is a taxi waiting area at Egham Station and a taxi office near the traffic lights on Station Road, 250 yards from the station. Otherwise telephone - All Point Cars (01784) 432468 or Arrow Cars (01784) 436533 or A2B (01784) 432222 or Egham Taxis (01784) 433933 or Gemini Cars (01784) 471111. The approximate costs of taxis to Cumberland Lodge are - £5 from Egham station, £20 from Heathrow airport and £50 from Gatwick airport. It is often best to book a taxi in advance to await your arrival at Egham Station or at an airport.

Attendees are advised not to use the black London taxi's since they can be very expensive.