Weighted and Truthful Predictions Sharad Goel Daniel M Reeves David M Pennock Presented by Nir Shabbat Outline Introduction Background and Related Work The Settings A Mechanism For Collective Revelation ID: 427322
Download Presentation The PPT/PDF document "Collective Revelation: A Mechanism for S..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Collective Revelation: A Mechanism for Self-Verified,Weighted, and Truthful Predictions
Sharad Goel, Daniel M. Reeves, David M. Pennock
Presented by: Nir ShabbatSlide2
Outline
IntroductionBackground and Related WorkThe SettingsA Mechanism For Collective RevelationThe Basic MechanismA General Technique for Balancing Budgets
SummarySlide3
Introduction
In many cases, a decision maker may seek to elicit and aggregate the opinions of multiple experts.Ideally, would like a mechanism that is:Incentive Compatible - Rewards participants to be
truthful
Information Weighted
-
Adjusts for the fact that some experts are better informed than
others
Self-Verifying
-
Works without the need for objective, “ground truth”
observations
Budget Balanced
-
Makes no net transfers to
agentSlide4
Background and Related Work
Proper Scoring Rules – Rewards the agent by assessing his forecast against an actual observed outcome.Agents are incentivized to be truthful.For example, using the Brier scoring rule:
We
get: Slide5
Background and Related Work
Peer Prediction Methods – Can be Incentive Compatible and Self-Verifying.For example, the Bayesian truth serum
asks agents to report both their own prediction and their prediction of other agents’ predictions.
Relies on the general property of information- scores, that a truthful answer constitutes the best guess about the most “surprisingly common” answer.Slide6
Background and Related Work
With the appropriate reward structure , this framework leads to truthful equilibria:
iff response of agent
is
- prediction of player
for the frequency of
- geometric average of predictions (or opinions) and meta predictions
Slide7
Background and Related Work
Predictions Markets – encourage truthful behavior and automatically aggregate predictions from agents with diverse information.For example, a prediction market for U.S. presidential elections. Agents buy and sell assets tied to an eventual Democratic or Republican win. Each share pay $1 if the corresponding event occurs.Slide8
Background and Related Work
Delphi Method – generates consensus predictions from experts generally through a process of structured discussion.In each round, each expert anonymously provides his forecast and reasons, at the end of the round, market maker summarize the forecasts and reasons.Process is repeated until forecasts converge.Slide9
Background and Related Work
Competitive Forecasting – elicits confidence intervals on predictions, thereby facilitating information weighting. Like prediction markets, competitive forecasting rewards accuracy, though is
not rigorously incentive compatible and relies on
benchmarking against
objective measurements.Slide10
Background and Related Work
Incentive Compatible
Information Weighted
Self-Verifying
Proper Scoring
Rules
●
Prediction Markets
○
●
Peer Prediction
○
●
Delphi Method
●
Competitive Forecasting○
●Polls●Collective Revelation
○
●
●Slide11
The Settings
Common prior – Agents & Market Maker have a common prior on the distribution of
Independent Private Evidence
– Each agent
privately observes
independent realization of the random variable
(independent both of each other and across agents).
Rationality
– Agents update their beliefs via Bayes’ rule.
Risk neutrality
– Agents act to maximize their expected payoff.
Slide12
A Mechanism For Collective Revelation
Agents are scored against one another’s private information (Self Verification).Agents report their subjective expectation of
, and their revised estimate in light of new hypothetical evidence (
HFS
).
From these 2 reports the subjective posterior, private information and also this agent’s prediction of other agents’ prediction is constructed.
Slide13
A Mechanism For Collective RevelationSlide14
The Basic Mechanism
LEMMA 1
An agent has subjective distribution
for
It is possible to elicit the k
th
moment of
via a proper scoring rule that pays based on the outcome
iff
Slide15
The Basic Mechanism
Simply put, scoring against a single Bernoulli observation can only reveal the agent’s expectation and not the uncertainty of it’s prediction (as quantified by the variance).So in order for a mechanism to be Information-Weighted
(for Bernoulli outcome) it must score agents against
multiple observations
.Slide16
The Basic Mechanism
For example, say we want to elicit an agent’s prediction regarding the distribution of
.
We want to reward the agent using a proper scoring rule based on a single outcome
:
To elicit the confidence, we
must
use another outcome
. For example, like this:
Slide17
The Basic Mechanism
The Beta distribution is the “conjugate prior” of the Bernoulli distribution.That is, given
with a prior
the posterior distribution of
, given
independent trials out of which
are successful, is:
Slide18
The Basic Mechanism
LEMMA 2Let
a prior on
Fix
(trials) and
(successes) integers
Define:
That is:
Slide19
The Basic Mechanism
LEMMA 2 (Cont’d)Then:
is a bijection from
to
And
Slide20
The Basic Mechanism
LEMMA 3Let
Let
a prior on
Let
Define:
Slide21
The Basic Mechanism
LEMMA 3 (Cont’d)Then:
is a bijection from
to
And:
Slide22
The Basic Mechanism
THEOREM 1Consider a Bayesian game according to the settings shown before with
players,
.
Let
a common prior on
Let
be the posterior dist. of agent
for
after updating according to its private info.
Fix
and
integers
Slide23
The Basic Mechanism
THEOREM 1 (Cont’d)Suppose that each agent
plays action
where
Slide24
The Basic Mechanism
THEOREM 1 (Cont’d)Define:
Where
is projection into the k
th
component.
Slide25
The Basic Mechanism
THEOREM 1 (Cont’d)For arbitrary constants
, set the reward of agent
to be:
Slide26
The Basic Mechanism
THEOREM 1 (Cont’d)Then:(*)is a strict Nash Equilibrium.Slide27
The Basic Mechanism
PROOFFix attention on agent , and suppose all agents
play according to (*).
By Lemma 2 -
parameters of
.
Slide28
The Basic Mechanism
PROOF (Cont’d)Let
be private observations of agent
. Then:
Consequently:
Slide29
The Basic Mechanism
PROOF (Cont’d)In particular,
and
If
plays according to (*) then:
Since we are using the Brier Proper scoring rule, strategy (*) maximizes
’s expected reward.
Slide30
The Basic Mechanism
PROOF (Cont’d)Moreover, since
is an injection, this is
’s unique best response, and so strategy (*) is a strict Nash equilibrium.
Slide31
The Basic Mechanism
CORROLARYSuppose all agents play the equilibrium strategy.Define:Slide32
The Basic Mechanism
CORROLARY (Cont’d)Define the aggregate information-weighted prediction as:
Let
be the posterior dist. Resulting from the cumulative private evidence of all agents.
Then
.
Slide33
A General Technique for Balancing Budgets
The idea is to reward agents via a Shared scoring rule.
A simple application would be:
And this is clearly budget balanced since
Slide34
A General Technique for Balancing Budgets
In our setting, the “observations”
are determined precisely by agents’ reports – simple application won’t work!
The solution proposed is to decouple
scoring
from
benchmarking
, e.g. agent
is rewarded according to its performance relative to a set of agents whose score cannot be affected by agent
’s action.
Slide35
A General Technique for Balancing Budgets
DEFINITIONLet
be a Bayesian game with
players
A
-projective family of
is a family of games
such that for each
with
,
is a Bayesian game restricted to the players
that preserves type spaces, action spaces, and players’ beliefs about types. In particular,
is determined by a family of reward functions
that specifies the expected reward for each player in
resulting from any given strategy profile of those players.
Slide36
A General Technique for Balancing Budgets
THEOREM 2G is a Bayesian game with n players
is a (strict) Nash equilibrium of G
Suppose
is a k-projective family of G
such that
Suppose that for each
,
is a (strict) Nash equilibrium for
.
Slide37
A General Technique for Balancing Budgets
THEOREM 2 (Cont’d)Then for any constant
, there are player rewards
for the n-player game G such that:
For any strategy profile s,
The original equilibrium q is still a (strict) Nash equilibrium for the modified game
In particular, by setting
, one can alter the rewards so that the game G is strongly budget balanced.
Slide38
A General Technique for Balancing Budgets
PROOFFor
, denote:
- reward function of player
in
- reward function of player
in the game
- reward of player
in game
, when
is restricted to game
.
Slide39
A General Technique for Balancing Budgets
PROOF (Cont’d)Denote the player sets
where the players wrap around for
.
Consider the modified rewards for
:
Slide40
A General Technique for Balancing Budgets
PROOF (Cont’d)Summing over all players we get:Slide41
A General Technique for Balancing Budgets
PROOF (Cont’d)Further more:
And since
, we have
, and so
for
.
Slide42
A General Technique for Balancing Budgets
PROOF (Cont’d)Consequently, if
is the strategy of player
and
is the strategy profile of all other players, then:
Where
is a function that does not depend on
.
Slide43
A General Technique for Balancing Budgets
PROOF (Cont’d)Since
is a (strict) Nash equilibrium for
:
And so, using
the new reward functions, the original equilibrium
is still a (strict) Nash equilibrium.
Slide44
Summary
We’ve seen a mechanism, i.e. Collective Revelation, for aggregating experts opinions which is:Incentive CompatibleInformation WeighedSelf-VerifyingWe’ve seen a general technique for constructing budget balanced mechanisms that applies both to collective revelation and to past peer-predictions method.Slide45
THE END