Glimcher Decision Making

71
Glimcher Decision Making

Transcript of Glimcher Decision Making

Page 1: Glimcher Decision Making

GlimcherDecision Making

Page 2: Glimcher Decision Making

Signal Detection Theory

•With Gaussian Assumption•Without Gaussian Assumption•Equivalent to Maximum Likelihood

w/o Cost Function

Page 3: Glimcher Decision Making
Page 4: Glimcher Decision Making
Page 5: Glimcher Decision Making
Page 6: Glimcher Decision Making
Page 7: Glimcher Decision Making

Newsome Dot Task

Page 8: Glimcher Decision Making
Page 9: Glimcher Decision Making
Page 10: Glimcher Decision Making

•Max Likelihood for an Individual Neuron (w/o Gaussian Assumption)•Monkey’s Choice Behavior

Page 11: Glimcher Decision Making

Neurons vs Behavior

Page 12: Glimcher Decision Making
Page 13: Glimcher Decision Making
Page 14: Glimcher Decision Making
Page 15: Glimcher Decision Making

So How Should You Set Criterion?(What is the Loss Function)

Page 16: Glimcher Decision Making

More complicated theories of “Bias”

•Expected Value•Expected Utility•Prospect Theory

•Game Theory

Page 17: Glimcher Decision Making

The rise (and fall?) of expected utility• Why EU?

– EV-maximization Σxxg(x) easily disproved • St. Petersburg paradox

…but EU is a simple generalization of EV– Axioms:

• Completeness of preference (including transitivity)• Continuity (“solvability”)

– If ffgfh then there exists p such that pf+(1-p)h∼g

• Cancellation (independence)– If ffg then pf+(1-p)z f pg+(1-p)z for all z, p>0

Paul: Explain EV and EU Here

Page 18: Glimcher Decision Making

Cancellation is logically sensible but perceptually unnatural

• Allais common consequence problem:– A: (.33, $2.5 M; 0.1, $0; .67, $2.4M) or (.33, $2.4 M; 0.1 $2.4M; .67, $2.4M)

– B: (.33, $2.5 M; 0.1, $0; .67, $0) or (.33, $2.4 M; 0.1 $2.4M; .67, 0)

– Majority choose $2.4 M in A and $2.5 M in BViolates Cancellation because .01 payoff is identical in both experiments so it

can be cancelled, this reveals that preferences have reversed

Page 19: Glimcher Decision Making

Many variants of expected utility

• Source: Camerer JRiskUnc 88, Edwards (Ed) Utility Theories, 92

Page 20: Glimcher Decision Making

Useful tool: Marschak-Machina triangle for representing 3-outcome gambles

• Outcomes x3>x2>x1

• Each point is a gamble

• Theories dictate properties of indifference curves

Page 21: Glimcher Decision Making

• EU is equivalent to– Linear indiff curves– Parallel

• Allais common consequence effect curves get steeper toward upper left

Page 22: Glimcher Decision Making

Prospect theory, I • Key features:

– Reference-dependence, nonlinear w(p)• Nonlinear weighting of probability

– Zeckhauser bullet example, 6 5 and 1 0– Inflection (sensitivity), elevation (risk attitude)– KT (’92 JRU) w(p)=pγ//(pγ+(1-p)γ)1/γ

– Prelec (98 E’metrica) w(p)=1/exp((ln(1/p)γ) • Large overweighting on low probabilities• γ=.7 w(1/10)=.165, w(1/100)=.05, w(1/1,000,000)=.002

• Morgenstern 79 J Ec Lit:– [Like Newtonian mechanics] The domain of our axioms on utility theory

is also restricted…For example, the probabilities used must be within certain plausible ranges and not go to .01 or even less to .001, then be compared with other equally tiny numbers such as .02 etc.”

Page 23: Glimcher Decision Making

Empirical weighting functions

Page 24: Glimcher Decision Making

Prospect theory, I

• Key features:– Reference-dependence, nonlinear w(p)

• Reference-dependence– Extension of psychophysics (e.g. hot-cold)– U(x-r)

• Koszegi-Rabin u(x) + ωv(x-r)– U(.) “reflects” (gamble over losses)– Loss-aversion

Page 25: Glimcher Decision Making

Prospect theory value function: Note kink at zero and diminishing marginal sensitivity

(concave for x>0, convex for x<0)

Page 26: Glimcher Decision Making

Do Classical Decision VariablesInfluence Brain Activity in LIP?

LIP

Page 27: Glimcher Decision Making

Varying Movement Value

Page 28: Glimcher Decision Making
Page 29: Glimcher Decision Making
Page 30: Glimcher Decision Making
Page 31: Glimcher Decision Making

What Influences LIP?

Related to Movement Desirability• Value/Utility of Reward• Probability of Reward• Overall Desirability (EU)

NB: Utility is used here without any intended reference to welfare

Page 32: Glimcher Decision Making

Varying Movement Probability

Page 33: Glimcher Decision Making
Page 34: Glimcher Decision Making
Page 35: Glimcher Decision Making

What Influences LIP?

Related to Movement Desirability• Value/Utility of Reward• Probability of Reward• Overall Desirability (EU)

Page 36: Glimcher Decision Making

What Influences LIP?

Related to Movement Desirability• Value/Utility of Reward• Probability of Reward• Overall Desirability (EU)

Page 37: Glimcher Decision Making

Is LIP a Decision Making Map?

A Map of Movement Desirabilities?

Page 38: Glimcher Decision Making

Theory of Games:

By The Nash Equations:At Behavioral Equilibrium

EUworking=EUshirking

Page 39: Glimcher Decision Making

Paul: Show Equilib Computation here

Page 40: Glimcher Decision Making

Behavior FixedEU Changes

Behavior ChangesEU Fixed

Page 41: Glimcher Decision Making
Page 42: Glimcher Decision Making
Page 43: Glimcher Decision Making
Page 44: Glimcher Decision Making
Page 45: Glimcher Decision Making
Page 46: Glimcher Decision Making
Page 47: Glimcher Decision Making
Page 48: Glimcher Decision Making
Page 49: Glimcher Decision Making

Along Indifference LinesBoth Actions Have Equal

Expected Utility

Page 50: Glimcher Decision Making
Page 51: Glimcher Decision Making
Page 52: Glimcher Decision Making
Page 53: Glimcher Decision Making
Page 54: Glimcher Decision Making
Page 55: Glimcher Decision Making
Page 56: Glimcher Decision Making
Page 57: Glimcher Decision Making

Monkey Work of Shirk

Page 58: Glimcher Decision Making
Page 59: Glimcher Decision Making
Page 60: Glimcher Decision Making

In Lottery Task:

Behavior Held Constant 50% into Response FieldExpected Utility Varies High vs. Low

Page 61: Glimcher Decision Making
Page 62: Glimcher Decision Making

In Strategic Game:

Behavior Varies Nash Equilibrium Expected Utility Fixed On Average EU1 = EU2

Page 63: Glimcher Decision Making
Page 64: Glimcher Decision Making
Page 65: Glimcher Decision Making

LIP Tracks Average EU

Page 66: Glimcher Decision Making

Local Fluctuations in EU?

Page 67: Glimcher Decision Making

Perhaps

Page 68: Glimcher Decision Making

LIP Codes Absolute or RelativeExpected Utility?

Page 69: Glimcher Decision Making

Stochastic Decision Model

Page 70: Glimcher Decision Making

Stochastic Decision Model

Page 71: Glimcher Decision Making

Stochastic Decision Model