Regularization in Directable Environments with Application to - - PowerPoint PPT Presentation

regularization in directable environments with
SMART_READER_LITE
LIVE PREVIEW

Regularization in Directable Environments with Application to - - PowerPoint PPT Presentation

Regularization in Directable Environments with Application to Tetris Jan Malte Lichtenberg zgr im ek Shrinkage Toward Equal Weights (STEW) Lichtenberg J.M. and im ek . Regularization in Directable Environments with


slide-1
SLIDE 1

Regularization in Directable Environments 
 with Application to Tetris

Jan Malte Lichtenberg Özgür Şimşek

slide-2
SLIDE 2

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Shrinkage Toward Equal Weights (STEW)

slide-3
SLIDE 3

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Shrinkage Toward Equal Weights (STEW)

0.0 0.1 0.2 0.3 10−4 10−2 100 102 104

λ (log scale) Weight estimates STEW (q = 2)

slide-4
SLIDE 4

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Shrinkage Toward Equal Weights (STEW)

Equal Weights

0.0 0.1 0.2 0.3 10−4 10−2 100 102 104

λ (log scale) Weight estimates STEW (q = 2)

slide-5
SLIDE 5

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Shrinkage Toward Equal Weights (STEW)

Equal Weights

0.0 0.1 0.2 0.3 10−4 10−2 100 102 104

λ (log scale) Weight estimates STEW (q = 2)

0.0 0.1 0.2 0.3 10−3 10−2 10−1 100 101 102

λ (log scale) Weight estimates Ridge regression:

slide-6
SLIDE 6

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

slide-7
SLIDE 7

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

“1 / N rule“ (DeMiguel et al., 2009)

slide-8
SLIDE 8

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

slide-9
SLIDE 9

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Are feature directions known?

slide-10
SLIDE 10

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Are feature directions known? “Direct” all features

slide-11
SLIDE 11

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Are feature directions known? “Direct” all features

slide-12
SLIDE 12

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Are feature directions known? “Direct” all features

slide-13
SLIDE 13

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

1 −2 −1 1 2 β

Density

slide-14
SLIDE 14

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

1 −2 −1 1 2 β

Density

  • 5

10 15 20 25 15 20 30 50 100 200

Training set size (log scale) MSE

  • EW

Ridge Lasso NNLasso STEW

slide-15
SLIDE 15

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

1 −2 −1 1 2 β

Density

1 −2 −1 1 2 β

Density

1 −2 −1 1 2 β

Density

Environment less directable

  • 2

4 6 8 15 20 30 50 100 200

Training set size (log scale) MSE

  • 2.5

5.0 7.5 10.0 15 20 30 50 100 200

Training set size (log scale) MSE

  • 5

10 15 20 25 15 20 30 50 100 200

Training set size (log scale) MSE

  • EW

Ridge Lasso NNLasso STEW

slide-16
SLIDE 16

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

  • 5

10 15 15 20 30 50 100 200

Training set size (log scale) MSE

  • Ridge

Lasso NNLasso STEW

1 −2 2 β

Density

slide-17
SLIDE 17

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

  • 5

10 15 15 20 30 50 100 200

Training set size (log scale) MSE

  • Ridge

Lasso NNLasso STEW 5 10 15 15 20 30 50 100 200

Training set size (log scale) Error component

  • Sq. Bias Ridge
  • Sq. Bias STEW

Variance Ridge Variance STEW

1 −2 2 β

Density

slide-18
SLIDE 18

Lichtenberg J.M. and Şimşek Ö. Regularization in Directable Environments with Application to Tetris.

Tetris

1000 2000 3000 4000 5000 3 5 10 20 50 100 200 300

Iteration (log scale) Mean score

M−learning + STEW M−learning + NN M−learning + Ridge M−learning Equal Weights

Poster #137 @ Pacific Ballroom