Wednesday, February 10, 2016

Feature Selection - Data Mining

Aurélie C. Lozano  covers model selection, factor analysis, PCA., joint feature selection and estimation, .

Key discussion points covered in this webcast are:

Analytics Capacities Landscape

Why Dimension Reduction?
- Solution: Dimension Reduction
- Example: Document classification

Methods for Dimension Reduction and Applications

- Main classes of techniques for Dimension Reduction
- Feature Extraction/Reduction
- Unsupervised Feature Reduction

- Geometric view of PCA: 2-D Gaussian Scatter plot
- Example of Application for PCA: Clustering
- Non-linear PCA

- Multidimensional Scaling (MDS)
- Manifold Learning
- Manifold Learning: ISOMAP
- Example of Application: Hand
- Supervised Feature Reduction

- Linear Discriminant Analysis (LDA)
- LDA: Sample applications
- Supervised Principal Components
- Supervised PCA: Example of Application

- Feature/variable selection
- Feature Selection
- Feature Selection: Filter Methods
- Feature Selection: Wrapper Methods
- Sample Algorithms
- Sample Applications

- Sparse Learning
- Why Sparse Learning?
- Sparse Learning: Joint dimension reduction and estimation
- The Lasso: The most popular sparse learning method
- The Group Lasso: Extends the lasso to accommodate grouped selection
- Various types of sparsity on matrices
- Various types of sparsity (matrix factorization)
- Application to GWAS - Genetic Basis of Complex Diseases
- Application to climate change attribution
- The approach: Sparse Learning with spatio-temporal data
- Application: Key influencers in online communities
- Application: IBM Lotus Bloggers
- Sparse Learning on Matrices: Image denoising
- Sparse Learning on matrices: network inference



Data Mining (Advanced Analytics): Sparse Learning & Dimension Reduction
IBM Business Analytics

IBM Business Analytics

Register and download
A practical, three-step guide to planning your first data mining project and selling it internally

Honglak Lee, Assistant Professor - Computer Science and Engineering, University of Michigan

The 4th University of Michigan Data Mining Workshop

 This workshop will present techniques: models and technologies for statistical data analysis, Web search technology, analysis of user behavior, data visualization, etc. We speak about data-centric applications to problems in all fields, whether it is in the natural sciences, the social sciences, or something else.


Michigan Engineering

Mod-04 Lec-28 Feature Selection : Problem statement and Uses



Mod-04 Lec-29 Feature Selection : Branch and Bound Algorithm


Lec-30 Feature Selection : Sequential Forward and Backward Selection



Mod-04 Lec-32 Feature Selection Criteria Function: Probabilistic Separability Based



Mod-01 Lec-30 Principal Component Analysis (PCA)
Part of Multivariate Statistical Modeling



Mod-10 Lec-37 Feature Selection and Dimensionality Reduction; Principal Component Analysis


No comments:

Post a Comment