NIPS ´08 Workshop: New Challenges in Theoretical Machine Learning: Learning with Data-dependent Concept Spaces

Study of Classification Algorithms using Moment Analysis

author:Amit Dhurandha, University of Florida
published: Dec. 20, 2008, recorded: December 2008, views: 89

You might be experiencing some problems with Your Video player.

Slides

Slides
0:00	Study of Classification Models and Model Selection Measures based on Moment Analysis based Analysis
0:25	Background
0:38	Problem: Classification Model Selection
0:50	Goal
1:05	Problem - 1
1:43	Ideally
2:01	E
2:31	Problem - 2
2:46	Applications of Studying Moments
4:42	Considering the Applications
4:53	Roadmap
5:36	Note
5:50	Concept of Generalization Error
6:03	Moments of GE
6:24	Can E Be Computed?
7:06	Too Many !!!
7:25	Optimizations
7:32	Number of Terms Optimization
8:26	Example
8:44	With m Inputs
8:59	Theorem 1
10:19	Moments
10:42	Optimization in Term Calculation
10:50	Efficiently Computing
11:05	Naive Bayes Classifier
11:10	NBC with d = 2
11:57	Exact Computation Too Expensive
12:28	We Let ...
13:09	Partial Derivatives
13:30	Our problem has Reduced
13:38	Preferred Solution
14:03	LP Dual
14:08	Subject to ...
14:14	Convex but Equation of Boundary Unknown
14:29	Solutions
15:17	Monte Carlo vs RS
16:28	RS Better than MC
17:09	Collapsing Joint Cumulative Probabilities
18:08	Optimization in Term Computation
18:37	Summary
18:54	Other Classification Algorithms - 1
19:24	Other Classification Algorithms - 2
19:35	Analysis of Model Selection Measures
19:38	Relationships between Moments
19:44	Cross Validation
20:09	Pair-Wise Covariances vs Number of Folds
20:20	Able to Explain Trends
20:33	Convergence in Real Dataset Sizes
20:50	Conclusion
21:37	Thank You

Related content

Visitors who watched this lecture also watched...