Pot -Luck Bring Your Ow n Problem Isabelle Guyon, Clopinet - - PowerPoint PPT Presentation

▶
pot luck
SMART_READER_LITE
LIVE PREVIEW

Pot -Luck Bring Your Ow n Problem Isabelle Guyon, Clopinet - - PowerPoint PPT Presentation

Nov 01, 2023 43 likes •184 views

Causality Challenge Pot -Luck Bring Your Ow n Problem Isabelle Guyon, Clopinet Constantin Aliferis and Alexander Statnikov, Vanderbilt Univ. Andr Elisseeff and Jean-Philippe Pellet, IBM Zrich Gregory F. Cooper, Pittsburg University Peter

slide-1
SLIDE 1

1

Causality Challenge

Pot -Luck

Bring Your Ow n Problem

Isabelle Guyon, Clopinet Constantin Aliferis and Alexander Statnikov, Vanderbilt Univ. André Elisseeff and Jean-Philippe Pellet, IBM Zürich Gregory F. Cooper, Pittsburg University Peter Spirtes, Carnegie Mellon

slide-2
SLIDE 2

2

Causality Workbench

  • Goal: Benchmark causal discovery algorithms.
  • Method:

– Challenges. – Repository of datasets, tasks, models, software, etc. – Interactive workbench. – Weekly teleconference seminar.

  • So far…

– Causality Challenge #1: Causation and Prediction (WCCI 2008). – Causality Challenge #2: Pot-luck (NIPS 2008).

  • Winter, 2008: Start developing an interactive workbench.
  • June, 2009: KDD workshop on causality in time series?
slide-3
SLIDE 3

3

Why a new challenge?

  • Causality challenge #1

– Favor “depth”

  • Single well defined task
  • Rigorous performance assessment
  • Causality challenge #2

– Favor “breadth”

  • Many different tasks
  • Encourage creativity
slide-4
SLIDE 4

4

http://clopinet.com/causality

5

slide-5
SLIDE 5

5

artif

Pot-Luck challenge

self eval self eval real real artif artif artif

0 (109)

Stemmatology

5 (218)

CauseEffectPairs

1 (330) TIED 2 (415) SIGNET 3 (570) PROMO 10 (558) LOCANET 2 (394) CYTO Type Participants

(views)

Task

self eval real real

slide-6
SLIDE 6

6

Other donated datasets

self eval real real real

35 SEFTI 59 SECOM 43 NOISE 65 MIDS 90 WebLogs Type Views Task

artif artif real

http://clopinet.com/causality

slide-7
SLIDE 7

7

Winners

PRIZES Best benchmark result: Best contributed task: MENTIONS Significant advance on REGED: SIDO: SIGNET: SIGNET: TIED:

Ernest Mwebaze and John Quinn You Zhou, Changzhang Wang, Jianxin Yin, Zhi Geng Mehreen Saeed Cheng Zheng and Zhi Geng Advanced Analytics, Intel, LTD Kun Zhang and Aapo Hyvärinen Guido Nolte

slide-8
SLIDE 8

8

MEK3/ 6 MAPKKK

P LCγ Erk1/ 2 Mek1/ 2 Raf PKC p38 Akt

MAPKKK MEK4/ 7

JNK

L A T

Lck

VAV SLP- 76

RAS PKA 1 2 3

CD28 CD3

PI 3K

LFA- 1 Cytohesin

Zap70

PI P3 PI P2 JAB- 1

Act ivat ors

  • 1. α- CD3
  • 2. α- CD28
  • 3. I CAM- 2
  • 4. PMA
  • 5. β2cAMP

I nhibitors

  • 6. G06976
  • 7. AKT inh
  • 8. Psitect
  • 9. U0126
  • 10. LY294002

10 5 4 6 7 9 8

Causal discovery from real m anipulations

The CYTO problem

Karen Sachs et al

slide-9
SLIDE 9

9

What if w e cannot experim ent?

The LOCANET problem

  • Experiments may be infeasible, costly
  • r unethical.
  • Using only observations we may want

to predict the effect of new policies.

  • Policies may consist in manipulating

several variables.

  • Task: Find the local causal structure

around a given target variable (depth 3 network) in four datasets (REGED, CINA, SIDO, and MARTI).

slide-10
SLIDE 10

10

Multiple alternative solutions

The TIED problem

Two disjoint subsets of variables V1 and V2 are Target Information Equivalent w.r.t. target Y TIEY(V1, V2), iff:

– V1⊥Y – V2⊥Y – V1⊥Y | V2 – V2⊥Y | V1

Alexander Statnikov & Constantin Aliferis X2 X3 X11 Y 1 2 1 2 2 1 2 3 3 3 1 1 2 3 X1 3

TIEY(X1, X2) TIEY(X1, X3) TIEY(X1, X11) TIEY(X2, X3) TIEY(X2, X11) TIEY(X3, X11)

slide-11
SLIDE 11

11

A dynam ical system : SIGNET

Abscisic Acid Signaling Netw ork

Donated by Jenkins and Soni Resimulated from Li, Assmann, Albert, PLOS, 2006

47

slide-12
SLIDE 12

12

Causality in tim e series: PROMO

A sim ulated m arketing task

  • 100 products
  • 1000 promotions
  • 3 years of daily data
  • Goal: quantify the

effect of promotions

  • n sales.

The difficulties include:

  • non iid samples
  • seasonal effects
  • promotions are binary, sales

are continuous

  • ther

1000 100

Jean-Philippe Pellet & André Elisseeff

slide-13
SLIDE 13

13

Causal direction am ong only tw o variables?

The CauseEffectPairs problem

  • Many causal discovery

methods rely on tests of conditional independence between 3 or more variables.

  • Task: Find the causal

direction among pairs of variables (real data, e.g. temperature and altitude).

Dominik Janzing

slide-14
SLIDE 14

14

What’s next?

  • Proceedings of NIPS workshop (JMLR, early 2009):

– Submit revised 10-page paper by December 19, 2008.

  • Depth vs. breadth → focus months:

– Teleconference presentations on one particular challenge. – Deadline for submission; result analysis and debate.

  • Causality challenge #3:

– Focus on time series. – Target KDD, June 2009.

  • Interactive workbench:

– Under development; target next NIPS.

http://clopinet.com/causality