0.25
0.5
0.75
1.25
1.5
1.75
2
We need a BIT more GUTS (Grand Unified Theory of Statistics)
Published on Jan 25, 20124149 Views
A remarkable variety of problems in machine learning and statistics can be recast as data compression under constraints: (1) sequential prediction with arbitrary loss functions can be transferred to
Related categories
Chapter list
We need a BIT more GUTS00:00
The End before the Beginning01:01
Central Thesis02:15
In defense of (some) monocausotaxophilia - 103:09
In defense of (some) monocausotaxophilia - 204:13
In defense of (some) monocausotaxophilia - 304:45
Bits?04:52
Menu06:04
On-Line Probabilistic Prediction06:57
Logarithmic Loss07:40
prediction strategy = distribution09:07
Universal Prediction with log loss10:34
A Bayesian Strategy - 111:37
A Bayesian Strategy - 212:32
Evaluating Bayes12:40
Bayes is very good universal code - 115:40
Bayes is very good universal code - 216:05
Bayes is very good universal code - 317:08
Entropification (The Gauss Device) - 117:19
Entropification (The Gauss Device) - 218:12
Entropification (The Gauss Device) - 319:10
Entropification (The Gauss Device) - 419:37
Entropification (The Gauss Device) - 520:14
Entropification (The Gauss Device) - 621:05
Applying Bayes to general predictors21:43
Bayes goes Beyond Model - 122:31
Bayes goes Beyond Model - 224:57
The Most Important Notion:25:41
The Clue - 126:27
The Clue - 227:38
The Clue - 329:07
The Clue - 429:12
The Clue - 530:27
The Clue - 630:49
The Clue - 732:06
Apply to Statistical Learning Theory32:26
Relation to Tsybakov Condition - 134:09
Relation to Tsybakov Condition - 235:18
Relation to Tsybakov Condition - 335:40
Relation to Tsybakov Condition - 437:19
If Tsybakov Exponent Unknown37:59
and learning when model is wrong - 139:29
and learning when model is wrong - 239:46
and learning when model is wrong - 340:37
The “Lynchpin” of Bayesian Statistics - 140:57
The “Lynchpin” of Bayesian Statistics - 241:23
The “Lynchpin” of Bayesian Statistics - 342:20
Measuring Mixability Gap as Testing42:49
Two Historical Roles of Compression in Statistics and ML - 144:07
Two Historical Roles of Compression in Statistics and ML - 244:49
No Hypercompression! - 145:31
No Hypercompression! - 246:00
No Hypercompression! - 346:15
No Hypercompression & p-values - 146:38
No Hypercompression & p-values - 247:00
Significance Testing - 147:06
Significance Testing - 247:36
Significance Testing - 348:02
Significance Testing - 448:04
Significance Testing - 548:05
Some problems specific to p-values...48:07
P-value and Counterfactuals49:20
Lo and Behold50:05
Our Test is not Standard LR Test50:29
The Emerging Picture - 150:32
The Emerging Picture - 251:00
Why this is not Bayes as we know it - 151:08
Why this is not Bayes as we know it - 251:17
Towards a Theory of “Safe” Probability?51:55