en-de
en-sl
en
en-zh
en-es
en-fr
0.25
0.5
0.75
1.25
1.5
1.75
2
Online Learning and Bregman Divergences
Published on 2007-02-2511762 Views
L 1: Introduction to Online Learning (Predicting as good as the best expert, Predicting as good as the best linear combination of experts, Additive versus multiplicative family of updates)\\ L 2: Breg
Related categories
Presentation
Online Learning and Bregman Divergences00:06
Bregman Divergences [Br,CL,Cs]01:01
Exponential Family of Distributions01:42
Expectation parameter04:10
Primal & Dual Parameters05:55
Gaussian (unit variance)06:50
Bernoulli08:39
Poisson10:40
Manfred1_Page_2512:07
Bregman Div. as Rel. Ent. between Distributions16:26
Area unchanged When Slide Flipped17:23
Dual divergence for Bernoulli19:56
Dual divergence for Poisson22:35
Dual matching loss for sigmoid transfer func.24:02
Example: Gaussian density estimation27:15
Derivation of Updates29:12
On-line Algorithm [AW]31:52
Alternate Motivation of Same On-Line Update32:27
Shrinkage Towards Initial34:46
Key Lemma [AW]37:29
Main Theorem37:55
Bounds for the Forward Algorithm38:15
Why Bregman divergences?42:17
General setup of on-line learning43:15
Minimax Algorithm for T Trials43:52
Gaussian45:51
Last-step Minimax47:35
Last-step Minimax: Bernoulli49:23
Synopsis of methods52:12
Content of this tutorial56:47
Simple conversions56:54
Expected loss bounds [HW]57:00
Tail bound [CCG]01:03:20
Application: Adaptive Channel Equalization01:04:22
Application: Caching [GBW]01:07:02
Caching Policies01:07:58
Which Policy to Choose?01:09:02
Characteristics Vary with Time01:09:44
Randomly Permuted Request Stream01:11:12
Want “Adaptive” Policy01:12:27