en-de
en-es
en-fr
en-sl
en
en-zh
0.25
0.5
0.75
1.25
1.5
1.75
2
MOA Concept Drift Active Learning Strategies for Streaming Data
Published on Nov 11, 20115981 Views
We present a framework for active learning on evolving data streams, as an extension to the MOA system. In learning to classify streaming data, obtaining the true labels may require major effort and m
Related categories
Chapter list
MOA Concept Drift Active Learning Strategies for Streaming Data00:00
Examples of data streams00:58
Mining data streams02:24
Predictive models for data streams (1)02:36
Predictive models for data streams (2)02:53
Predictive models for data streams (3)02:53
Predictive models for data streams (4)03:06
Active learning for data streams (1)03:40
Active learning for data streams (2)03:56
Active learning for data streams (3)04:05
Problem setting summary04:33
Contributions05:05
active learning strategies for data streams05:39
Do we need this label? (1)05:45
Do we need this label? (2)05:51
Random strategy (naive) - 105:52
Random strategy (naive) - 206:25
Do we need this label? - Fixed threshold06:31
Online active learning in the data stream setting?06:39
Fixed uncertainty06:52
Online active learning in the data stream setting?07:21
What is needed?07:36
Budget: we have resources to label up to 1/3 of the incoming examples07:58
Need to label but no resources08:13
Nothing is labelled very certain, but not necessarily accurate, as data evolves08:24
What is needed?08:38
Changes in the regions where classifier is very certain should not be missed08:52
Do we need this label? - Our strategies09:14
What is needed?09:21
Adaptive uncertainty strategy09:35
Randomized uncertainty10:10
empirical results10:39
Massive Online Analysis - MOA10:46
Experimental evaluation14:05
REUTERS data (1)15:07
REUTERS data (2)15:35
REUTERS data (3)17:08
Accuracy with different budgets17:08
Accuracy at different budgets17:50
conclusion18:20
Conclusion18:23
Thanks!19:17