A Note on Foundations of Bayesianism Stefan Arnborg, KTH Gunnar Sjödin, SICS.

Slides:



Advertisements
Similar presentations
Geometry Chapter 2 Terms.
Advertisements

With examples from Number Theory
Introduction to Proofs
Statistically motivated quantifiers GUHA matrices/searches can be seen from a statistical point of view, too. We may ask ‘Is the coincidence of two predicates.
1 In this lecture  Number Theory ● Rational numbers ● Divisibility  Proofs ● Direct proofs (cont.) ● Common mistakes in proofs ● Disproof by counterexample.
Cryptography and Network Security
Bayesian Inference Chris Mathys Wellcome Trust Centre for Neuroimaging UCL SPM Course London, May 12, 2014 Thanks to Jean Daunizeau and Jérémie Mattout.
Congestion Games with Player- Specific Payoff Functions Igal Milchtaich, Department of Mathematics, The Hebrew University of Jerusalem, 1993 Presentation.
BAYES RULES! -in finite models (ECCAI2000) -but not in infinite! (MaxEnt2000) Stefan Arnborg, KTH Gunnar Sjödin, SICS.
Fuzzy Sets and Applications Introduction Introduction Fuzzy Sets and Operations Fuzzy Sets and Operations.
Basic properties of the integers
Lecture 6 Hyperreal Numbers (Nonstandard Analysis)
Copyright © Cengage Learning. All rights reserved.
CS 454 Theory of Computation Sonoma State University, Fall 2011 Instructor: B. (Ravi) Ravikumar Office: 116 I Darwin Hall Original slides by Vahid and.
CMSC 250 Discrete Structures Number Theory. 20 June 2007Number Theory2 Exactly one car in the plant has color H( a ) := “ a has color”  x  Cars –H(
I The meaning of chance Axiomatization. E Plurbus Unum.
Foundations of Measurement Ch 3 & 4 April 4, 2011 presented by Tucker Lentz April 4, 2011 presented by Tucker Lentz.
Copyright © 2007 Pearson Education, Inc. Slide 8-1.
3-4 Algebra Properties Used in Geometry The properties of operations of real numbers that you used in arithmetic and algebra can be applied in geometry.
2.5 Reasoning in Algebra and Geometry
De Finetti’s ultimate failure Krzysztof Burdzy University of Washington.
Sets.
COMP14112: Artificial Intelligence Fundamentals L ecture 3 - Foundations of Probabilistic Reasoning Lecturer: Xiao-Jun Zeng
Inequalities and Proof
T. Mhamdi, O. Hasan and S. Tahar, HVG, Concordia University Montreal, Canada July 2010 On the Formalization of the Lebesgue Integration Theory in HOL On.
A Brief Summary for Exam 1 Subject Topics Propositional Logic (sections 1.1, 1.2) –Propositions Statement, Truth value, Proposition, Propositional symbol,
Fuzzy Logic. Lecture Outline Fuzzy Systems Fuzzy Sets Membership Functions Fuzzy Operators Fuzzy Set Characteristics Fuzziness and Probability.
Introduction to Proofs
Mathematics Review Exponents Logarithms Series Modular arithmetic Proofs.
MATH 224 – Discrete Mathematics
Methods of Proof. This Lecture Now we have learnt the basics in logic. We are going to apply the logical rules in proving mathematical theorems. Direct.
Chapter 4 : Similar Triangles Informally, similar triangles can be thought of as having the same shape but different sizes. If you had a picture of a triangle.
Introduction to the HyperReals An extension of the Reals with infinitely small and infinitely large numbers.
Homework Review notes Complete Worksheet #1. Homework Let A = {a,b,c,d}, B = {a,b,c,d,e}, C = {a,d}, D = {b, c} Describe any subset relationships. 1.
The Integers. The Division Algorithms A high-school question: Compute 58/17. We can write 58 as 58 = 3 (17) + 7 This forms illustrates the answer: “3.
Chapter 5 Existence and Proof by contradiction
Methods of Proof Lecture 3: Sep 9. This Lecture Now we have learnt the basics in logic. We are going to apply the logical rules in proving mathematical.
Data Security and Encryption (CSE348) 1. Lecture # 12 2.
On joint modelling of random uncertainty and fuzzy imprecision Olgierd Hryniewicz Systems Research Institute Warsaw.
Honors Geometry Intro. to Deductive Reasoning. Reasoning based on observing patterns, as we did in the first section of Unit I, is called inductive reasoning.
Building a System of Geometry Knowledge 2.4
Geometry Honors Section 2. 2
Chapter 2 Section 5. Objective  Students will make a connection between reasoning in Algebra and reasoning in Geometry.
1 Copyright, 1996 © Dale Carnegie & Associates, Inc. Chapter 2: Logic & Incidence Geometry Back To the Very Basic Fundamentals.
Making sense of randomness
Basic Principles (continuation) 1. A Quantitative Measure of Information As we already have realized, when a statistical experiment has n eqiuprobable.
Warm Up. Warm Up Answers Theorem and Proof A theorem is a statement or conjecture that has been shown to be true. A theorem is a statement or conjecture.
International Conference on Fuzzy Systems and Knowledge Discovery, p.p ,July 2011.
Objective: To prove and apply theorems about angles Proving Angles Congruent (2-6)
Proving Statements in Geometry Eleanor Roosevelt High School Geometry Mr. Chin-Sung Lin.
CS104:Discrete Structures Chapter 2: Proof Techniques.
1 Copyright, 1996 © Dale Carnegie & Associates, Inc. Chapter 2: Logic & Incidence Geometry Back To the Very Basic Fundamentals.
2.5 Reasoning in Algebra and Geometry Algebraic properties of equality are used in Geometry. –Will help you solve problems and justify each step. In Geometry,
Methods of Proof Lecture 4: Sep 20 (chapter 3 of the book, except 3.5 and 3.8)
Properties of Equality and Congruence
Chapter 2 1. Chapter Summary Sets (This Slide) The Language of Sets - Sec 2.1 – Lecture 8 Set Operations and Set Identities - Sec 2.2 – Lecture 9 Functions.
Week 8 - Wednesday.  What did we talk about last time?  Relations  Properties of relations  Reflexive  Symmetric  Transitive.
Intro to Proofs Unit IC Day 2. Do now Solve for x 5x – 18 = 3x + 2.
Uncertainty management Bayesianism found inadequate for lack of priors and means to compute posteriors Multitude of alternatives invented Modified after.
Objective - To use properties of numbers in proofs. Logical Reasoning Deductive ReasoningInductive Reasoning - process of demonstrating that the validity.
CHAPTER 5 Handling Uncertainty BIC 3337 EXPERT SYSTEM.
2.5 and 2.6 Properties of Equality and Congruence
Predicate Calculus Validity
Chapter 2.6 (Part 1): Prove Statements about Segments and Angles
Y. Davis Geometry Notes Chapter 2.
3-2 Angles & Parallel Lines
2.5 Reasoning in Algebra and Geometry
The Foundations: Logic and Proofs
Methods of Proof Rosen 1.7, 1.8 Lecture 4: Sept 24, 25.
Methods of Proof Rosen 1.7, 1.8 Lecture 4: Sept, 2019.
Presentation transcript:

A Note on Foundations of Bayesianism Stefan Arnborg, KTH Gunnar Sjödin, SICS

BAYES RULES! -in finite models (ECCAI2000) -but not in infinite! (MaxEnt2000) Stefan Arnborg, KTH Gunnar Sjödin, SICS

Normative claim of Bayesianism 4 EVERY type of uncertainty should be treated as probability 4 Aristotle, Sun Zi(300BC), Bayes(1763), Laplace, de Finetti, Jeffreys, Keynes, Ramsey, Adams, Lindley, Cheeseman, Jaynes,… 4 This claim is controversial and not universally accepted: Fisher(1922), Cramér, Zadeh, Dempster, Shafer, Walley(1999) …

Fundamental Justifications 4 Consistent Betting Paradigm: de Finetti, Savage(1950), Lindley(1982), … Snow(1999) 4 Information Based: Cox(1946), Aczél(1966), Jaynes(1994) Criticized by Paris(1994), Halpern(1999)

Main Result: 4 Cox information based justification can be derived with weak common sense assumptions. Difference between finite and infinite models. 4 Assumptions are: Refinability Information independence Strict monotonicity Infinite case: Model is closed

Caveats 4 Classical Bayesianism only for finite model 4 And for closed infinite model embeddable in the real line 4 Ordered closed infinite model: Extended probability (with infinitesimals) also permissible (non-monotonic and default logics).

Jaynes’s Desiderata on Uncertainty Management 4 Uncertainty is measured by real number, dependent on information subject possesses: A|C : plausibility of A given C. 4 Consistency. 4 Common sense.

Real Numbered uncertainties 4 Given set of statements (possible world sets) A, B, C, … 4 Plausibility A|C: plausibility of A given that C is known to be true - a real number 4 Conjunction: AB 4 Disjunction A+B, Difference A-B 4 AB|C=F(A|BC, B|C) 4 A+B|C=G(A|C,B-A|C) 4 not A|C=S(A|C)

RESCALABILITY THEOREMS 4 Under suitable assumptions there is a strictly monotone function w(x) such that 4 w(F(x,y))=w(x)w(y) 4 w(G(x,y))=w(x)+w(y) 4 I.E., by rescaling the plausibility measure by w, model becomes a probability model 4 I.E., if you accept the assumptions, then Bayes Rules!

Invariance under rescaling 4 * and + are strictly monotone, symmetric, associative and jointly distributive 4 These properties are invariant under strictly monotone rescaling 4 If F and G violate the properties, rescaling is impossible.

Consistency 4 AB|C==BA|C, thus F(A|BC,B|C)=F(B|AC,A|C) 4 A+B|C==B+A|C 4 (AB)C|D==A(BC)|D 4 (A+B)C|D==AC+BC|D 4 Does this mean that F,G must be associative, symmetric and jointly distributive??? 4 No, not without additional assumptions!

OUR common sense assumptions 4 REFINABILITY: Assume B’|B=c was defined; It should be possible to refine another event A by A’ so that A’|A=c (cf Tribus, Jimison, Heckerman) 4 INFORMATION INDEPENDENCE: New events obtained by refinement of same event can be postulated independent: A|BC=A|C and B|AC=B|C ’Knowledge of one has no effect on plausibility of the other’

Halpern’s Example: 4 Worlds A B C D E G H I J K L M D|E=H|J B|C = L|M A|C = I|JE|G = A|B H|J≈K|M D|G = K|LM

Example: F(F(x,y),z)≈F(x,F(y,z))   C D E G H I J K L M D|E=H|J=x B|C = L|M=z A|C = I|JE|G = A|B=y H|J≈K|M D|G = K|LM

Refine:A’|A=D|E: INCONSISTE   C D E G H I J K L M D|E=H|J=x B|C = L|M=z A|C = I|JE|G = A|B=y H|J≈K|M D|G = K|LM A’ H|J=A’AB|C=K|M !!!!!!!!!!!!!

OBSERVATION 4 The functions F and G must be symmetric and associative if refinability and information independence accepted 4 F, G must likewise be jointly distributive F(G(x,y),z)=G(F(x,z),F(y,z)) 4 But only on the finite range of definition 4 Not enough for rescalability

Rescalability is solvability of LP L4+L4-La=0 L3+L5-La=0 L2+L4-Lb=0 L1+L5-Lb=0 L4+L6-Lc=0 L3+L7-Lc=0 L2+L6-Ld=0 L1+L8-Ld=0 L1<L2<… L8 F(x4,x4)=F(x3,x5)=a F(x2,x4)=F(x1,x5)=b F(x4,x6)=F(x3,x7)=c F(x2,x6)=F(x1,x8)=d x1<x2<…<x8 Find L such that ML=0 and DL>0 Unfortunately, the equalities imply L7=L8 !!!!

Theorem 4, finite case: 4 Rescalability follows for finite models from weak common sense assumptions: refinability and information independence using finite-dimensional duality theory 4 Conjecture: Savages and Lindleys consistent betting behavior analyses can be similarly strengthened

Probability model Counterexample, x^(i+1)=F(x^i,x) Log probability i INFINITE CASE: NON-SEPARABILITY x y

Extended Probability 4 Probability values are taken from ordered field. 4 An ordered field is generated by rationals, reals and infinitesimals (Conway)  Previous example explained by x=0.5, y=0.5+ . 4 Extended probability has been shown equivalent to non-monotonic reasoning schemes (Benferhat, Dubois, Prade, 1997).

Ordered rings and fields Integers Modular Rational Gaussian Algebraic reals Complex Reals Ordered infinitesimals Conway’s No No: a universal ordered field, extension of any ordered field.

Infinite Models 4 Theorem 6:Model is rescalable iff all plausibilities are separable !! - but separability somewhat contrived. 4 Assume plausibility model can be closed: F(x,y) defined on D^2 G(x,y) defined on D^2 if y<S(x) Range of F in D Range of G in D.

Theorems, Infinite Closed Models: 4 Theorem 9: Every plausibility measure is equivalent to extended probability. 4 Corollary 10:Every plausibility measure that can be embedded in the reals is equivalent to standard probability.

Proof Sketch 4 Assume plausibility measure closed in D F G S(x) Introduce subtraction (a,b) (a,b)≈(c,d) if G(a,d)=G(c,b) or G(S(a),S(d))=G(S(c),S(b)) D DxD/≈

Extend from -infty to +infty … (i+d)+(j+e)=i+j+d+e (i+d)(j+e)=ij+ie+jd+de We now have an ordered ring which is also an integral domain * and + are extensions of F and G D ZxD

Create an ordered field 4 Standard quotient construction for ordered integral domain gives an ordered field (MacLane-Birkhoff) 4 This field is a subfield of the universal ordered field No of Conway

Proof Sketch, continued 4 Ordered fields are generated by reals and infinitesimals, non-zero values smaller than any real number, i.e., are models of extended probability

Proof Sketch, continued... 4 If the closed model is required to consist of real numbers, the model is equivalent to standard probability: All subfields of No with lowest upper bounds are embeddable in the real number field, and if field is real, then D was also real (the embedding process does not introduce infinitesimals in separable model).

SUMMARY 4 With assumptions of refinability, independence and strict monotonicity, finite ordered plausibility models are equivalent to probability models 4 With further assumption of closability, (infinite) plausibility models are equivalent to extended probability models 4 And closed plausibility models embeddable in the reals are equivalent to probability models.

OPEN PROBLEMS 4 NONE? (for well-defined statements) 4 However, Bayesian analysis of imprecise statements interesting alternative to fuzzy/rough/possibilistic logics (cf Wittgenstein (1956) word games). 4 Meaning of A fuzzy -> meaning depends on context -> modelled as conditional meaning.