Maximum Likelihood vs. Sequential Normalized Maximum Likelihood in On-line Density Estimation  thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

Maximum Likelihood vs. Sequential Normalized Maximum Likelihood in On-line Density Estimation

Published on Aug 02, 20113775 Views

The paper considers sequential prediction of individual sequences with log loss (online density estimation) using an exponential family of distributions. We first analyze the regret of the maximum

Related categories

Chapter list

Maximum Likelihood vs. Sequential Normalized Maximum Likelihood in On-line Density Estimation00:00
Outline00:16
Sequential Prediction02:38
Sequential Prediction: Example04:43
Avoiding Infinite Regret05:59
Maximum Likelihood Strategy08:19
Regret Bound for ML Strategy10:18
Improving ML Strategy11:46
SNML: Examples (without projection)12:55
Regret Bound for SNML Strategy14:06
Example: ML vs. SNML - 114:41
Example: ML vs. SNML - 215:46
Example: ML vs. SNML - 315:47
Example: ML vs. SNML - 415:48
Example: ML vs. SNML - 515:49
Example: ML vs. SNML - 615:55
SNML and Bayes - 116:00
SNML and Bayes - 216:43
Conclusions17:58