Personality traits are known to moderate treatment response and are often an essential add-on to a symptom picture when performing a patient’s systematic evaluation. However, personality measures are often long to administer due to their large number of items. Rammstedt and John (2007) abbreviated the Big Five Inventory (BFI-44) to a 10-item version (BFI-10) and found that the shortened scales retained reasonable levels of reliability and validity. The Italian adaptation of BFI-44 was administered to 645 subjects, together with a socio-demographic questionnaire. Psychometric properties (i.e., internal consistency and construct validity) of the BFI-44 and of BFI-10 were assessed through Confirmatory Factor Analyses. Psychometric properties of the BFI-44 and BFI-10 overlapped those of the English, Spanish and German version. Confirmatory analyses revealed that the factor structure based on responses to the items of BFI-10 was invariant with the factor structure based on responses to the items of BFI-44. We also modeled the effects of social desirability, age, gender and their interactions. The effects of such covariates were substantially invariant across factor structures of BFI-10 and BFI-44. Social desirability increased the goodness of fit of the measurement model while the linear component of age was positively correlated with Conscientiousness and negatively with Nevroticism, on which females scored higher than males. Though the BFI-10 scales showed acceptable levels of reliability and validity, they do not reach the depth of construct operazionalization provided by the scales of BFI-44, which thus should be employed in systematic evaluation in clinical settings.
Personality traits are known to moderate treatment response and are often an essential add-on to a symptom picture when performing a patient’s systematic evaluation. However, personality measures are often long to administer due to their large number of items. Rammstedt and John (2007) abbreviated the Big Five Inventory (BFI-44) to a 10-item version (BFI-10) and found that the shortened scales retained reasonable levels of reliability and validity. The Italian adaptation of BFI-44 was administered to 645 subjects, together with a socio-demographic questionnaire. Psychometric properties (i.e., internal consistency and construct validity) of the BFI-44 and of BFI-10 were assessed through Confirmatory Factor Analyses. Psychometric properties of the BFI-44 and BFI-10 overlapped those of the English, Spanish and German version. Confirmatory analyses revealed that the factor structure based on responses to the items of BFI-10 was invariant with the factor structure based on responses to the items of BFI-44. We also modeled the effects of social desirability, age, gender and their interactions. The effects of such covariates were substantially invariant across factor structures of BFI-10 and BFI-44. Social desirability increased the goodness of fit of the measurement model while the linear component of age was positively correlated with Conscientiousness and negatively with Nevroticism, on which females scored higher than males. Though the BFI-10 scales showed acceptable levels of reliability and validity, they do not reach the depth of construct operazionalization provided by the scales of BFI-44, which thus should be employed in systematic evaluation in clinical settings.
Xavier Amatriain, VP of Engineering, Quora at MLconf SEA - 5/01/15MLconf
Machine learning applications for growing the world’s knowledge at Quora: At Quora our mission is to “share and grow the world’s knowledge”. We want to do this by getting the right questions to the right people to answer them, but also by getting the existing answers to people who are interested in them. In order to accomplish this we need to build a complex ecosystem where we value issues such as content quality, engagement, demand, interests, or reputation. It is not possible to build a system like this unless most of the process are highly automated and scalable. We are fortunate though to have lots of very good quality data on which to build machine learning solutions that can help address all of the previous requirements.
In this talk I will describe some interesting uses of machine learning at Quora that range from different recommendation approaches such as personalized ranking to classifiers built to detect duplicate questions or spam. I will describe some of the modeling and feature engineering approaches that go into building these systems. I will also share some of the challenges faced when building such a large-scale knowledge base of human-generated knowledge.
The Science behind Viral Marketing is a look at the key factors that drive growth in viral marketing. (Hint, the most important factor is not the one everyone expects.) It also looks at what is needed to get virality to work, and how to create and optimize viral marketing campaigns or viral products.
One part of the presntation shows the key formulae behind viral marketing.
Suitable for marketers or for product designers.
Quora is a powerful marketing and outreach tool, if you resist the urge to spam the crap out of it. At Pubcon 2016, I went over the stuff that's made it a useful platform for me, as a Quora amateur.
Presentation at "Emerging problems in particle phenomenology" workshop held at CUNY on April 11, 2010. Has sensitivity of Jets+MET searches for 7 TeV LHC.
Computational complexity and simulation of rare events of Ising spin glasses Martin Pelikan
We discuss the computational complexity of random 2D Ising spin glasses, which represent an interesting class of constraint satisfaction problems for black box optimization. Two extremal cases are considered: (1) the +/- J spin glass, and (2) the Gaussian spin glass. We also study a smooth transition between these two extremal cases. The computational complexity of all studied spin glass systems is found to be dominated by rare events of extremely hard spin glass samples. We show that complexity of all studied spin glass systems is closely related to Frechet extremal value distribution. In a hybrid algorithm that combines the hierarchical Bayesian optimization algorithm (hBOA) with a deterministic bit-flip hill climber, the number of steps performed by both the global searcher (hBOA) and the local searcher follow Frechet distributions. Nonetheless, unlike in methods based purely on local search, the parameters of these distributions confirm good scalability of hBOA with local search. We further argue that standard performance measures for optimization algorithms---such as the average number of evaluations until convergence---can be misleading. Finally, our results indicate that for highly multimodal constraint satisfaction problems, such as Ising spin glasses, recombination-based search can provide qualitatively better results than mutation-based search.
These are the slides from the review session. THE FILE IS BIG AND MAY HAVE BEEN CORRUPTED. IF YOU CAN'T SEE IT THROUGH THE FLASH INTERFACE, JUST CLICK THE "DOWNLOAD" LINK and view it on your own computer.
Xavier Amatriain, VP of Engineering, Quora at MLconf SEA - 5/01/15MLconf
Machine learning applications for growing the world’s knowledge at Quora: At Quora our mission is to “share and grow the world’s knowledge”. We want to do this by getting the right questions to the right people to answer them, but also by getting the existing answers to people who are interested in them. In order to accomplish this we need to build a complex ecosystem where we value issues such as content quality, engagement, demand, interests, or reputation. It is not possible to build a system like this unless most of the process are highly automated and scalable. We are fortunate though to have lots of very good quality data on which to build machine learning solutions that can help address all of the previous requirements.
In this talk I will describe some interesting uses of machine learning at Quora that range from different recommendation approaches such as personalized ranking to classifiers built to detect duplicate questions or spam. I will describe some of the modeling and feature engineering approaches that go into building these systems. I will also share some of the challenges faced when building such a large-scale knowledge base of human-generated knowledge.
The Science behind Viral Marketing is a look at the key factors that drive growth in viral marketing. (Hint, the most important factor is not the one everyone expects.) It also looks at what is needed to get virality to work, and how to create and optimize viral marketing campaigns or viral products.
One part of the presntation shows the key formulae behind viral marketing.
Suitable for marketers or for product designers.
Quora is a powerful marketing and outreach tool, if you resist the urge to spam the crap out of it. At Pubcon 2016, I went over the stuff that's made it a useful platform for me, as a Quora amateur.
Presentation at "Emerging problems in particle phenomenology" workshop held at CUNY on April 11, 2010. Has sensitivity of Jets+MET searches for 7 TeV LHC.
Computational complexity and simulation of rare events of Ising spin glasses Martin Pelikan
We discuss the computational complexity of random 2D Ising spin glasses, which represent an interesting class of constraint satisfaction problems for black box optimization. Two extremal cases are considered: (1) the +/- J spin glass, and (2) the Gaussian spin glass. We also study a smooth transition between these two extremal cases. The computational complexity of all studied spin glass systems is found to be dominated by rare events of extremely hard spin glass samples. We show that complexity of all studied spin glass systems is closely related to Frechet extremal value distribution. In a hybrid algorithm that combines the hierarchical Bayesian optimization algorithm (hBOA) with a deterministic bit-flip hill climber, the number of steps performed by both the global searcher (hBOA) and the local searcher follow Frechet distributions. Nonetheless, unlike in methods based purely on local search, the parameters of these distributions confirm good scalability of hBOA with local search. We further argue that standard performance measures for optimization algorithms---such as the average number of evaluations until convergence---can be misleading. Finally, our results indicate that for highly multimodal constraint satisfaction problems, such as Ising spin glasses, recombination-based search can provide qualitatively better results than mutation-based search.
These are the slides from the review session. THE FILE IS BIG AND MAY HAVE BEEN CORRUPTED. IF YOU CAN'T SEE IT THROUGH THE FLASH INTERFACE, JUST CLICK THE "DOWNLOAD" LINK and view it on your own computer.
1. Model Independent Searches for Jets + MET
or
Discovering Gluinos in an Uncertain Universe
Jay Wacker
SLAC
CDF Exotics
July 17, 2008
0803.0019 & work in progress
with J. Alwall, M-P. Le, M. Lisanti
2. Outline
Introduction
Modules vs Models
Model Independent Searches
Projected Sensitivity
Two Miscellaneous Items
3. High Energy Frontier
No “sure thing” theory to discover
Tevatron, Flavor, Precision EW, Higgs
LHC may not burst into a superfire
Many BSM possibilities to search for
Supersymmetric Standard Model
Universal Extra Dimensions
Randall-Sundrum
Little Higgs
Different TeV scale physics, but similar signals
Inverse problem hard
Discovery first
4. Jets plus Missing Energy
A common signature
New Colored Particle Decays to WIMP
Existing searches based upon MSSM
˜˜ ˜˜
˜˜ gg qg
qq
Very general template to start from
Can find SSM, UED, RS/LH w/ T-parity
5. Jets + Missing Energy Cuts at D0
1fb-1 analysis
˜˜
˜˜ ˜˜
qg
qq gg
Gg
1j + ET 2j + ET 3j + ET 4j + ET
≥ 150 ≥ 35 ≥ 35 ≥ 35
ET j1
≥ 35 ≥ 35 ≥ 35
ET j2 < 35
≥ 35 ≥ 35
ET j3 < 35 < 35
≥ 20
ET j4 < 20 < 20 < 20
≥ 150 ≥ 225 ≥ 150 ≥ 100
ET
≥ 150 ≥ 300 ≥ 400 ≥ 300
HT
(Not exclusive searches)
T=
HTable 1: Summary of the cuts used by DO
ET j
Will these discover anything visible in these channels?
cascade into Standard Model particles plus some particles th
6. What a theorist knows about gluino limits
DØ Preliminary, 0.96 fb-1
600
Squark Mass (GeV)
Slepton & Chargino
tan!=3, A =0, µ <0
0
CDF II
limits
500
400
DØ IA
UA1
UA2
CDF IB
no mSUGRA
solution
300
DØ IB
200
100
LEP
0
0 100 200 300 400 500 600
Gluino Mass (GeV)
7. mSugra is not representative of the MSSM
mg : mB = 6 : 1
˜
˜
Anomaly Mediation
Mirage Mediation
non-Minimal Gauge Mediation
Never varies decay kinematics
Want a model independent search
But first, a brief survey of models...
8. Outline
Introduction
Modules vs Models
Model Independent Searches
Projected Sensitivity
Two Miscellaneous Items
9. ˜˜
Examining g g more carefully
“Modules”
Minimal set of particles and interactions
necessary to do a search over a class of models
10. ˜˜
Examining g g more carefully
“Modules”
Minimal set of particles and interactions
necessary to do a search over a class of models
The minimal “gluino” module
˜ ¯˜
Turn on one decay mode g → q q χ0
Keep masses and total cross section free
˜˜
σ(p¯ → g g X)
mg mχ p
˜ ˜
11. ˜˜
Examining g g more carefully
“Modules”
Minimal set of particles and interactions
necessary to do a search over a class of models
The minimal “gluino” module
˜ ¯˜
Turn on one decay mode g → q q χ0
Keep masses and total cross section free
˜˜
σ(p¯ → g g X)
mg mχ p
˜ ˜
Captures many models (MSSM, UED, etc)
Misses heavy flavor and cascades
12. Where has the Tevatron probed “gluinos”?
mSugra has focused on one mass ratio
Q = mg − mχ
˜ ˜
Q=0
˜
χ
m
=
Q = mχ
˜
g ˜
m
mχ
˜
mSugra
mg
˜
13. Degenerate Search
j1
j3
˜
g ˜
˜ B
B
˜
g
ET
j2 j4
Useful when not phase space limited Q = mg − mB > mB
˜ ˜
˜
If Q < mB
˜
Bino carries away energy but not momentum
1
As gluinos get boosted, jets become
∆Φ j ET
∼
collinear and ET aligned with jets
γg
˜
14. Producing Degenerate Gluinos
˜
B
˜ j2
g
j1 ET
˜
g j3
˜
B
Need additional hard jets
Want the spectrum as well
ET > 150 GeV
j1
PT > 150 GeV
D0
j2
PT < 50 GeV ∆ΦjET > 30◦
j1
PT > 150 GeV ET > 120 GeV
CDF
j2
PT < 60 GeV
∆Φj2ET > 0.3
j3
PT < 20 GeV
15. Searches useful in gluino searches
Q=0
˜ j ISR + ET
χ
m
=
Q = mχ
˜
g ˜
m
mχ
˜
n j + ET
mSugra
mg
˜
17. For one cascade this is tractable
2 additional parameters from the minimal module
mW Br(˜ → W )
g
˜
Know the results in the following limits
Br(˜ → W ) → 0
g
mW → mB
˜ ˜
mW → mg
˜ ˜
Ask “when does the cascade hurt the search the most?”
18. Outline
Introduction
Modules vs Models
Model Independent Searches
Projected Sensitivity
Two Miscellaneous Items
19. Should be a better way of searching
Don’t want to miss a visible signal
Jets plus MET Searches are effectively:
Jet classification criterion
Visible Energy and Missing Energy Cuts
As parameters in a module vary,
visible and missing energy change dramatically
20. Exclusive Jets + MET Search
4 Separate Searches, Individually Optimized
1j + ET 2j + ET 3j + ET 4j + ET
≥ 150 ≥ 35 ≥ 35 ≥ 35
ET j1
≥ 35 ≥ 35 ≥ 35
ET j2 < 35
≥ 35 ≥ 35
ET j3 < 35 < 35
≥ 20
ET j4 < 20 < 20 < 20
≥ 150 ≥ 225 ≥ 150 ≥ 100
ET
≥ 150 ≥ 300 ≥ 400 ≥ 300
HT
Table 1: Summary of the cuts used by DO
Leave Free
21. One Proposal
For each jet multiplicity
d2 σ
∆HT ∆ET
Set a limit on
dHT dET
e.g.
4 jets
800
< 5fb 5−2 fb < 2fb
+2
600
10−3 fb
+3
< 3fb
HT < 10fb
400
< 20fb 10+8 fb < 5fb
−8
200
100 200 300 400
ET
22. Backgrounds
Want to vary cuts to maximize discovery potential
Generate SM events and compare to D0
→ → PGS
Madgraph Pythia
23. Backgrounds
Want to vary cuts to maximize discovery potential
Generate SM events and compare to D0
→ → PGS
Madgraph Pythia
Three Dominant Backgrounds
W/Z + jets
t tbar
QCD
Subdominant Backgrounds
Diboson
Single top
24. W/Z + jets Backgrounds
Hit Z+jets to within QCD K-factors
W+jets need a ~30% MET-independent scaling
probably PGS efficiency at losing a lepton
Top Background
Worked at ~30% MET -independent level
QCD Background
No attempt to simulate ET > 100 GeV
25. Before HT ET cuts
Very useful plot for theorists
2 jet analysis
DØ Preliminary
4
10
Events / 20
Events / 5
Data
W quot; l ! + jets
300
Z quot; !! + jets
tt
103 WW,WZ,ZZ -
Z quot; l + l + jets
250
single-t
Signal
200
102
150
10
100
1 50
0
0 50 100 150 200 250 300 350 400 450 500 0
ET (GeV)
27. Outline
Introduction
Modules vs Models
Model Independent Searches
Projected Sensitivity
Two Miscellaneous Items
28. Calculating Additional Jets
Need additional radiation for signal
Parton Showering Matrix Elements
QCD Bremstrahlung Necessary for well-separated jets
Soft/Collinear Approximation Includes quantum interference
Resums large logs Fixed order calculation
Computationally Cheap Computationally expensive
Unlimited number of partons Limited number of partons
Matching merges best of both worlds
Necessary to avoid double counting
29. Calculating Additional Jets
Matrix Elements
g g 0j
˜˜ g g 1j
˜˜ g g 2j
˜˜
Parton Shower
ve
vet
to
oi
if
fp
ve
pT
to
T >
Q
>
if
cu
Qc
pT
t
ut
>
Qc
ut
g g 0j
˜˜ g g 1j
˜˜ g g 2j
˜˜
Decay
g ∗ g ∗ 0j
˜˜ g ∗ g ∗ 1j
˜˜ g ∗ g ∗ 2j
˜˜
31. First can try a cuts based search
mg = 210 GeV mB = 100 GeV
˜
˜
Dijet most effective channel
but signal merges with QCD
HT ≥ 150 GeV E T ≥ 100 GeV
HT ≥ 225 GeV E T ≥ 300 GeV
38. Sensitivity of the Worst Case Scenario
mg ∼ 120 GeV mg ∼ 130 GeV
˜ ˜
150
Bino Mass GeV
100
Out[27]=
X
50
0
100 200 300 400 500
Gluino Mass GeV
39. Outline
Introduction
Modules vs Models
Model Independent Searches
Projected Sensitivity
Two Miscellaneous Items
40. Monophotons vs Monojets
In degenerate limit
˜ ˜ ˜ ˜
g g g g
g
γ
vs
¯ ¯
q q
q q
Trend moving to Monophoton searches
Monojet rate is much larger for gluinos
(gluinos radiate gluons, but not photons)
Monojet discovery potential is significantly better
41. Leptons from Cascade Decays
Lepton Rich Cascades
˜
q
˜
g Lots of constraints
2j
˜ from flavour on
W squarks and sleptons
˜
˜
B
Leptons + Jets + MET very effective
42. Leptons from Cascade Decays
Lepton Poor Cascades
˜
q
˜
g
˜
2j
˜
W jj(70%) ν(30%)
W±
jj(70%) (10%)
˜ Z0
B
Leptons + Jets + MET not nearly as effective
43. Have implicitly been assuming lepton poor cascades
But, we’ve seen gluinos can be mg > 120 GeV
˜∼
σg ∼ O(100 pb)!
˜
O(400, 000) gluino pairs could have been produced
Even lepton poor spectra can have lots of leptons
σg ∼ 10 pb
˜ Lots of Hard Same Sign Dileptons
˜
g 200 GeV ˜
g2
2j
˜ 160 GeV
W W + W − Z0
W ±Z 0
˜ W+
B 80 GeV
˜
g1 W −
Z0
44. We are probing the Energy Frontier
Don’t know what we are looking for
Models are just motivation
Should strip models to their modules
Leads to more model-independent searches
Worst tragedy is to not discover a visible signal