Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
This presentation is shown in RIMS conference, 
Kyoto, JAPAN 
ftanaka@sigmath.es.osaka-u.ac.jp 
Quantum Minimax Theorem 
i...
Prologue: 
What is Bayesian Statistics? 
(Non-technical Short Version) 
*It is for non-statisticians esp. researchers in q...
Mission of Statistics 
Mission of Statistics 
To give a desirable and plausible method according to statistical models, 
p...
What’s Bayesian Statistics? (1/2) 
Bayesian Statistics 
S i i Statistics d i i admitting h the usage of f i prior di ib i ...
What’s Bayesian Statistics? (2/2) 
Relation to frequentist’s statistics (traditional statistics) 
1 1. Bayesian statistics...
Misunderstanding 
Since priors are not objective, Bayesian methods should not be used for 
scientific result. 
Objective p...
Further Reference 
Preface in RIMS Kokyuroku 1834 (Theory and applications of statistical 
inference in quantum theory) 
P...
CONTENTS 
0. Notation 
1. Introduction 
2. Quantum Statistical Models and Measurements 
3. Quantum Statistical Decision Pr...
0. Notation
Description of Probability in QM 
Quantum Mechanics (QM): described by operators in a Hilbert space 
P b bilit Probability...
Density Operators 
Density operators (density matrices, state) 
S ( ) : { ll d Hilb t  } 
 all d.o.s on a Hilbert space...
Measurements in QM (1/2) 
Mathematical model of measurements 
= Positive Operator Valued Measure (POVM) 
POVM (def def. fo...
Measurements in QM (2/2) 
How to construct a probability space (Axiom of QM) 
d d.o o.   S ( ) 
{ , , , } 1 2 k sample ...
1. Introduction
Quantum Bayesian Hypothesis Testing(1/2) 
Alice chooses one alphabet (denoted by an integer) and sends one 
corresponding ...
Quantum Bayesian Hypothesis Testing(2/2) 
* The probability that Bob guesses j when Alice sends i. 
i j p (B  j | A  i) ...
Summary of Q-BHT 
Chooses one alphabet and sends one quantum state 
1,2,..., { , ,..., } (H) 1 2 S k 1 2 k     
Prop...
Minimax POVM 
1.Chooses one alphabet and sends one quantum state 
  { , ,..., } (H) 1 2 S k 1,2,..., k     
Guess th...
Quantum Minimax Theorem (Simple Ver.) 
1.Chooses one alphabet and sends one quantum state 
  { , ,..., } (H ) 1 2 S k 1,...
Example (1/2) 
states 
1 0 0 1/ 2 1/ 2 0 
 
 
 
 
 
 
 
 
 
 
0 0 0 
 
 
 
 
Quantum  
 
 
   , 
. 
 ...
states 
Example (2/2) 
1 0 0 1/ 2 1/ 2 0 
 
 
  
 
 
  
0 0 0 
 
 
  
Quantum , 
 
0 0 0 1 2 
 
, 1/ 2 1/ 2 ...
Rewritten in Technical Terms 
11.Nature 
Parameter space Quantum Statistical Model 
  
{ }  S (H)    1,2,..., k  
E...
Main Result (Brief Summary) 
Quantum Minimax Theorem 
      
R R 
inf sup ( )  sup inf ( ) (d ) M M ( ) 
M M 
 ...
Previous Works 
Wald (1950) 
Statistical Decision Theory 
H l Holevo (1973) Q t Quantum St titi Statistical lD ii Decision...
2. Quantum Statistical Models 
and Measurements
Formal Definition of Statistical Models 
(naïve ) Quantum statistical model = A parametric family of density operators 
 ...
Regular Operator-Valued Function (1/2) 
Definition 
 
K   
Locally compact metric space 
compact set 
: {( , ) : ( ,...
Regular Operator-Valued Function (2/2) 
Remark 1 
lim ( ) 0 
 K T K  
    
0 
Uniformly continuous w.r.t. trace no...
Quantum Statistical Models 
Definition 
S (H) : {X  L(H) : Tr X  1, X  0} 
 :   S (H) is called a quantum statistic...
Measurements (POVMs) 
U Decision space (locally compact space) 
  
AU Borel sets (the smallest sigma algebra containing ...
Born Rule 
Axiom of Quantum Mechanics 
For the quantum system described by 
 
and a measurement described by a POVM 
M(dx...
3. Quantum Statistical 
Decision Problems
Basic Setting 
Situation 
For the quantum system specified with the unknown parameter 
experimenters extract information f...
Decision Functions 
Example 
- Estimate the unknown parameter x 1   
 
a ( x )  
xn n 
x) - Estimate the unknown d.o. ...
Remarks for Non-statisticians 
Remarks 
1. If the quantum state is completely known, then the distribution of 
measurement...
Loss Functions 
Performance of decision functions is compared by adopting 
the loss functions (smaller is better). 
Ex: Pa...
Measurements Over Decisions 
Basic Idea 
From the g g, beginning, we y only consider POVMs over the action p 
space. 
Put ...
Quantum Statistical Decision Problems 
Formal Definitions 
 Parameter space (locally compact space) 
 ( ) Quantum stati...
4. Risk Functions and 
Optimality Criteria
Comparison of POVMs 
The loss (e.g. squared error) depends on both unknown parameter  
and 
our decision u u, which is a ...
D fi iti 
Risk Functions 
Definition 
The risk function of 
M  Po (U ) 
R ( ) : w( , u )Tr ( )M(d u ) M     E w( , ...
Optimality Criteria (1/2) 
Definition 
A POVM is said to be minimax if it minimizes the supremum of the risk function, i.e...
Optimality Criteria (2/2) 
P() All probability distributions (defined on Borel sets) 
In y Bayesian , statistics,   ( P...
Ex. of Loss Functions and Risk Functions 
Parameter Estimation 
{ ( ) :    R m } 
U   2 w( , u )    u 
R ( ) u...
5. Main Result
Main Result (Quantum Minimax Theorem) 
Theorem (*) 
Let  and U be a compact metric space. 
Then the following equality ho...
Key Lemmas For Theorem 
Compactness of the POVMs 
Holevo (1976) 
If the decision space U is compact, then P Po (U ) 
is al...
6. Minimax POVMs 
and Least Favorable Priors
Previous Works (LFPs) 
Wald (1950) Statistical Decision Theory 
Jeffreys (1961) 
Jeffreys prior 
Bernardo ( , 1979, ) 
200...
Main Result (Existence Theorem of LFP) 
Definition 
If a prior supremum, i.e., achieves the supremum i e the following hol...
Pathological Example 
Remark 
Even on a compact space, a bounded lower semicontinuous (and not 
continuous) loss function ...
Minimax POVM is constructed using LFP 
Corollary 
Let  and U 
be a compact space 
metric space. 
If a LFP exists for a qu...
7. Summary
Discussion 
1. We show quantum minimax theorem, which gives theoretical basis in 
quantum statistical decision theory, in ...
Upcoming SlideShare
Loading in …5
×

Quantum Minimax Theorem in Statistical Decision Theory (RIMS2014)

720 views

Published on

This is almost self-contained explanation of our recent result. The contents are based on our talk in the RIMS2014 conference.

Recently, many fundamental and important results in statistical decision theory have been extended to the quantum system. Quantum Hunt-Stein theorem and quantum locally asymptotic normality are typical successful examples.
In our recent preprint, we show quantum minimax theorem, which is also an extension of a well-known result, minimax theorem in statistical decision theory, first shown by Wald and generalized by LeCam. Our assertions hold for every closed convex set of measurements and for general parametric models of density operator. On the other hand, Bayesian analysis based on least favorable priors has been widely used in classical statistics and is expected to play a crucial role in quantum statistics. According to this trend, we also show the existence of least favorable priors, which seems to be new even in classical statistics.

Published in: Engineering
  • DOWNLOAD THIS BOOKS INTO AVAILABLE FORMAT (Unlimited) ......................................................................................................................... ......................................................................................................................... Download Full PDF EBOOK here { https://soo.gd/qURD } ......................................................................................................................... Download Full EPUB Ebook here { https://soo.gd/qURD } ......................................................................................................................... Download Full doc Ebook here { https://soo.gd/qURD } ......................................................................................................................... Download PDF EBOOK here { https://soo.gd/qURD } ......................................................................................................................... Download EPUB Ebook here { https://soo.gd/qURD } ......................................................................................................................... Download doc Ebook here { https://soo.gd/qURD } ......................................................................................................................... ......................................................................................................................... ................................................................................................................................... eBook is an electronic version of a traditional print book THIS can be read by using a personal computer or by using an eBook reader. (An eBook reader can be a software application for use on a computer such as Microsoft's free Reader application, or a book-sized computer THIS is used solely as a reading device such as Nuvomedia's Rocket eBook.) Users can purchase an eBook on diskette or CD, but the most popular method of getting an eBook is to purchase a downloadable file of the eBook (or other reading material) from a Web site (such as Barnes and Noble) to be read from the user's computer or reading device. Generally, an eBook can be downloaded in five minutes or less ......................................................................................................................... .............. Browse by Genre Available eBooks .............................................................................................................................. Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, Cookbooks, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult, Crime, Ebooks, Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, ......................................................................................................................... ......................................................................................................................... .....BEST SELLER FOR EBOOK RECOMMEND............................................................. ......................................................................................................................... Blowout: Corrupted Democracy, Rogue State Russia, and the Richest, Most Destructive Industry on Earth,-- The Ride of a Lifetime: Lessons Learned from 15 Years as CEO of the Walt Disney Company,-- Call Sign Chaos: Learning to Lead,-- StrengthsFinder 2.0,-- Stillness Is the Key,-- She Said: Breaking the Sexual Harassment Story THIS Helped Ignite a Movement,-- Atomic Habits: An Easy & Proven Way to Build Good Habits & Break Bad Ones,-- Everything Is Figureoutable,-- What It Takes: Lessons in the Pursuit of Excellence,-- Rich Dad Poor Dad: What the Rich Teach Their Kids About Money THIS the Poor and Middle Class Do Not!,-- The Total Money Makeover: Classic Edition: A Proven Plan for Financial Fitness,-- Shut Up and Listen!: Hard Business Truths THIS Will Help You Succeed, ......................................................................................................................... .........................................................................................................................
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • DOWNLOAD THIS BOOKS INTO AVAILABLE FORMAT (Unlimited) ......................................................................................................................... ......................................................................................................................... Download Full PDF EBOOK here { http://shorturl.at/krvUW } ......................................................................................................................... Download Full EPUB Ebook here { http://shorturl.at/krvUW } ......................................................................................................................... Download Full doc Ebook here { http://shorturl.at/krvUW } ......................................................................................................................... Download PDF EBOOK here { http://shorturl.at/krvUW } ......................................................................................................................... Download EPUB Ebook here { http://shorturl.at/krvUW } ......................................................................................................................... Download doc Ebook here { http://shorturl.at/krvUW } ......................................................................................................................... ......................................................................................................................... ................................................................................................................................... eBook is an electronic version of a traditional print book THIS can be read by using a personal computer or by using an eBook reader. (An eBook reader can be a software application for use on a computer such as Microsoft's free Reader application, or a book-sized computer THIS is used solely as a reading device such as Nuvomedia's Rocket eBook.) Users can purchase an eBook on diskette or CD, but the most popular method of getting an eBook is to purchase a downloadable file of the eBook (or other reading material) from a Web site (such as Barnes and Noble) to be read from the user's computer or reading device. Generally, an eBook can be downloaded in five minutes or less ......................................................................................................................... .............. Browse by Genre Available eBooks .............................................................................................................................. Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, Cookbooks, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult, Crime, Ebooks, Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, ......................................................................................................................... ......................................................................................................................... .....BEST SELLER FOR EBOOK RECOMMEND............................................................. ......................................................................................................................... Blowout: Corrupted Democracy, Rogue State Russia, and the Richest, Most Destructive Industry on Earth,-- The Ride of a Lifetime: Lessons Learned from 15 Years as CEO of the Walt Disney Company,-- Call Sign Chaos: Learning to Lead,-- StrengthsFinder 2.0,-- Stillness Is the Key,-- She Said: Breaking the Sexual Harassment Story THIS Helped Ignite a Movement,-- Atomic Habits: An Easy & Proven Way to Build Good Habits & Break Bad Ones,-- Everything Is Figureoutable,-- What It Takes: Lessons in the Pursuit of Excellence,-- Rich Dad Poor Dad: What the Rich Teach Their Kids About Money THIS the Poor and Middle Class Do Not!,-- The Total Money Makeover: Classic Edition: A Proven Plan for Financial Fitness,-- Shut Up and Listen!: Hard Business Truths THIS Will Help You Succeed, ......................................................................................................................... .........................................................................................................................
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Be the first to like this

Quantum Minimax Theorem in Statistical Decision Theory (RIMS2014)

  1. 1. This presentation is shown in RIMS conference, Kyoto, JAPAN ftanaka@sigmath.es.osaka-u.ac.jp Quantum Minimax Theorem in Statistical Decision Theory November 10, 2014 PPuubblliic VVerrsiionn 田中冬彦((Fuyuhiko TANAKA)) Graduate School of Engineering Science, Osaka University If you are interested in details, please see quant-ph 1410.3639 ; citations are also welcome!
  2. 2. Prologue: What is Bayesian Statistics? (Non-technical Short Version) *It is for non-statisticians esp. researchers in quantum information.
  3. 3. Mission of Statistics Mission of Statistics To give a desirable and plausible method according to statistical models, purpose of inference (point estimation, prediction, hypothesis testing, model selection, etc.), and loss functions. *Precise evaluation of estimation error is secondary. Difference from Pure Math Statistics is required to meet various needs in the world → Significant problems in statistics have changed according to the times (This trivial fact does not seem to be known so much except for statisticians.)
  4. 4. What’s Bayesian Statistics? (1/2) Bayesian Statistics S i i Statistics d i i admitting h the usage of f i prior di ib i distributions prior distribution= prob. dist. on the unknown parameter (We do not care about the interpretation of this prob.) *Interpretation heavily depends on individual researcher.
  5. 5. What’s Bayesian Statistics? (2/2) Relation to frequentist’s statistics (traditional statistics) 1 1. Bayesian statistics has enlarged the framework and satisfies various needs. (Not an alternative of frequentists’ statistics!) 2 2. Sticking to frequentists frequentists’ methods like MLEs is the preference to a specific prior distribution (narrow-minded!) (Indeed, MLEs often fail; their independence of loss is unstuaible ) 3. Frequentists’ results give good starting points to Bayesian analysis For details, see, e.g., C. Robert, Bayesian Choice Chap.11 Frequentists (T diti l Bayesian Statistics Traditional statistics)
  6. 6. Misunderstanding Since priors are not objective, Bayesian methods should not be used for scientific result. Objective priors (Noninformative priors) are proposed for such usage. Bayesian statistics do not deny frequentists’ way and statistical inference based on an objective prior often agrees with frequentists’ one. (ex: MLE = MAP estimate under the uniform prior) It seems unsolved which noninformative prior should be used. The essence of this problem is how to choose a good statistical method with small sample. It is NOT inherent to Bayesian statistics, but has been a central issue in traditional statistics. Frequentists and mathematicians avoid this and only consider mathematically tractable models and/or asymptotic theory (it’s large-sample approximation!) In Bayesian statistics, this problem is transformed into the choice of noninformative prior
  7. 7. Further Reference Preface in RIMS Kokyuroku 1834 (Theory and applications of statistical inference in quantum theory) POINT - Point out some misunderstanding of Statistics g for non-experts (esp. Physicists) - Define the word “quantum statistical inference” as an academic term Sorry but it is written in Japanese.
  8. 8. CONTENTS 0. Notation 1. Introduction 2. Quantum Statistical Models and Measurements 3. Quantum Statistical Decision Problems 4. Risk Functions and Optimality Criteria 5. Main Result 6. Minimax POVMs and Least Favorable Priors 7. Summary
  9. 9. 0. Notation
  10. 10. Description of Probability in QM Quantum Mechanics (QM): described by operators in a Hilbert space P b bilit Probability: d ib described d b by d it density op. and d measurement t D it Density t operator Probability distributions  ( )   (dx )  Tr ( )M(d x ) Measurement(POVM) M(d x ) M t Measurements  ( ) x a ( x ) (dx )   M(dx ) Statistical processing of data
  11. 11. Density Operators Density operators (density matrices, state) S ( ) : { ll d Hilb t  }  all d.o.s on a Hilbert space  { L() | Tr 1,  0} When dimH  n   n * 0 1 U p U             0 i p   j 0 p  p n Tr 1 1 j  
  12. 12. Measurements in QM (1/2) Mathematical model of measurements = Positive Operator Valued Measure (POVM) POVM (def def. for finite sets) Sample space  = Possible values for a measurement { , , , } 1 2 k  x x  x POVM { POVM{ M j } j  1 1,  , k = a family of linear ops ops. satisfying M M x O j j : ({ })    j j M I Positive operator (positive semidefinite matrix)
  13. 13. Measurements in QM (2/2) How to construct a probability space (Axiom of QM) d d.o o.   S ( ) { , , , } 1 2 k sample space   x x  x j j k M 1, , { }   POVM (Measurement) Then the prob obtaining a measurement outcome x   j prob. j j p : Tr  M * The above p prob. satisfies conditions (p positive and g summing to one M O j  due to both def.s of d.o.s and POVMs.   0 j I M j   j Tr  1 and
  14. 14. 1. Introduction
  15. 15. Quantum Bayesian Hypothesis Testing(1/2) Alice chooses one alphabet (denoted by an integer) and sends one corresponding quantum state (described by a density operator) di d ib d b d i ) 1. { , ,..., } (H ) 1 2 S k 1,2,..., k      2. Bob receives the quantum state and perform a measurement and guesses the alphabet Alice really sent to him. The whole process is described by a POVM j j 1, 2 , ,k {M }   3. The proportion of alphabets is given by a distribution 1, , , 0 1 1     k k       (called a prior distribution) 23214  1  2  4  2  3 
  16. 16. Quantum Bayesian Hypothesis Testing(2/2) * The probability that Bob guesses j when Alice sends i. i j p (B  j | A  i)  Tr  M * Bob’s loss (estimation error) when he guesses j while Alice sends i. w(i, j ) k * Bob’s risk for the i-th alphabet  R i w i j p B j A i M ( )  ( , ) (  |  ) j 1 * Bob’s average risk (his task is to minimize by using a good POVM) k  R ( i)  i i i 1 M )
  17. 17. Summary of Q-BHT Chooses one alphabet and sends one quantum state 1,2,..., { , ,..., } (H) 1 2 S k 1 2 k     Proportion of each alphabet (prior distribution)  1     k  1,  k  0 Guess the alphabet by choosing a measurement (POVM) j j 1, 2 , ,k {M }   k k r :    R ( i ) R ( i) ( i j )T M Average risk  i j i w M ) , Tr  ,M M i  i 1  j 1 Bayes POVM w.r.t. = POVM minimizing the average risk
  18. 18. Minimax POVM 1.Chooses one alphabet and sends one quantum state   { , ,..., } (H) 1 2 S k 1,2,..., k     Guess the alphabet by choosing a measurement (POVM) j j 1, 2 , ,k {M }   Minimize the worst-case risk Bob has no prior information * The worst-case risk : sup ( ) M *M r R i i  Minimax POVM = POVM minimizing the worst-case risk
  19. 19. Quantum Minimax Theorem (Simple Ver.) 1.Chooses one alphabet and sends one quantum state   { , ,..., } (H ) 1 2 S k 1,2,..., k     Guess the alphabet by choosing a measurement (POVM) j j 1, 2 , ,k {M } Theorem ( Hirota and Ikehara, 1982) k   inf sup ( ) sup inf  ( ) M M M M i i i R i R i 1  , ) Minimax POVM agrees with the Bayes POVM w.r.t. the worst case prior. *The worst-case prior to Bob is called a least favorable prior.
  20. 20. Example (1/2) states 1 0 0 1/ 2 1/ 2 0           0 0 0     Quantum       , .   0 0 0 1 2 , 1/ 2 1/ 2 0 0 0 0 0 0 0             0 0 0 3 0 0 1         0-1 loss w(( , i, j)  1   ij Minimax POVM 1 1/ 2 1/ 2 0   1 2  2  M  M 1/ 2 1 1/ 2 0 , 0 0 0 . 1  1/ 2  1/ 2 0 2 1/ 2 1 1/ 2 0 , 1 1                  0 0 0 3       M  0 0 0 0 0 0      0 0 1      *This is a counterexample of Theorem 2 in Hirota and Ikehara (1982), where their proof seems not mathematically rigorous.
  21. 21. states Example (2/2) 1 0 0 1/ 2 1/ 2 0         0 0 0     Quantum ,  0 0 0 1 2  , 1/ 2 1/ 2 0     . 0 0 0 0 0 0              0 0 0 3 0 0 1         0-1 loss ij w(i, j)  1   LFP (1)  (2)  1 / 2, (3)  0 LF LF LF    Minimax POVM is constructed as a Bayes POVM w.r.t. LFP. Important li i Implication When completely unknown, it is not necessarily wise to find an optimal POVM with the uniform prior. (Although Statisticians already know this fact long decades ago…)
  22. 22. Rewritten in Technical Terms 11.Nature Parameter space Quantum Statistical Model   { }  S (H)    1,2,..., k  Experimenter and Statistician Decision space POVM on U u uU U  1,2,..., k  {M } Loss function w( , u ) Risk function   u R ( ) w( , u )Tr M M     uU Theorem i f R ( ) i f  ( ) R ( ) inf sup sup inf  M M M M     
  23. 23. Main Result (Brief Summary) Quantum Minimax Theorem       R R inf sup ( )  sup inf ( ) (d ) M M ( ) M M     Po   P Po R (  )   ( )T (  )M(d ) : w( , u Tr du M where  U  R R cf) inf sup ( )  sup inf ( ) (d )     D D      P ( )  Conditions, assumptions, and essence of proof are explained.
  24. 24. Previous Works Wald (1950) Statistical Decision Theory H l Holevo (1973) Q t Quantum St titi Statistical lD ii Decision Th Theory Recent results and applications (many!) Kumagai and Hayashi (2013) Guta and Jencova (2007) Hayashi (1998) Le Cam (1964) ( Classical) Minimax Theorem in statistical decision theory ) y First ver. is given by Wald (1950) we show Quantum Minimax Theorem in quantum statistical decision theory
  25. 25. 2. Quantum Statistical Models and Measurements
  26. 26. Formal Definition of Statistical Models (naïve ) Quantum statistical model = A parametric family of density operators   3 1 2 1 1   i  0  (  ) 2  (  ) 2  (  ) 2  1 Ex.        1 2 3      i 1    1 2 3 2 ( )    , , R 1 2 3      (r , s; )  r  ( ) s  (r  ) ( s  )   R Basic Idea (Holevo, 1976) A quantum statistical model is defined by a map : (H) 1    L    ( ) Next, we see the required conditions for this map
  27. 27. Regular Operator-Valued Function (1/2) Definition  K   Locally compact metric space compact set : {( , ) : ( , )  }  K  K  K d    0 (H) trace-class operators on a Hilbert space 1 L : (H) 1 T   L For a map , we define      K X X T  T  X   K T ( ) : inf :   ( )  ( )  ,( , ) 1 Definition An operator-valued function T  K T K   lim ( ) 0 0     is regular
  28. 28. Regular Operator-Valued Function (2/2) Remark 1 lim ( ) 0  K T K      0 Uniformly continuous w.r.t. trace norm on every compact set as sup T (  ) T (  )  0  0 Converse does not hold if 1 ( , ) K dim H       Remark 2 The regularity assures that the following operator-valued integral is well-defined   f ( )T ( ) (d ) for every ( )   P () 0 f  C  and      K X X T  T  X   K T ( ) : inf :   ( )  ( )  ,( , ) 1 y
  29. 29. Quantum Statistical Models Definition S (H) : {X  L(H) : Tr X  1, X  0}  :   S (H) is called a quantum statistical model if it satisfies the conditions below. Conditions 1. Identifiability (one-to-one map) 1 2 1 2  ( )   ( ),   2. Regularity lim ( ) 0 0 T  K  K      ↑Necessary for our arguments
  30. 30. Measurements (POVMs) U Decision space (locally compact space)   AU Borel sets (the smallest sigma algebra containing all open sets) D fi i i Definition M Positive-operator valued function is called a POVM if it satisfies B  A(U )  MB   0        U B B A j i B B     M  B M( B )     j j 1 1 j j , , 1 2  i j   ,  MU   I Po (U ) All POVMs over U
  31. 31. Born Rule Axiom of Quantum Mechanics For the quantum system described by  and a measurement described by a POVM M(dx) t measurement t outcome i is di tib t distributed d di according t to x ~ Tr M(dx ) (dx )      POVM x M(dx ) (dx )  
  32. 32. 3. Quantum Statistical Decision Problems
  33. 33. Basic Setting Situation For the quantum system specified with the unknown parameter experimenters extract information for some purposes.  ( ) POVM x a ( x )  (dx ) M(dx )  Typical sequence of task 1 1. To choose a measurement (POVM) 2. To perform the measurement over the quantum system 3. To take an action a(x) among choices based on the measurement outcome x formally called a decision function
  34. 34. Decision Functions Example - Estimate the unknown parameter x 1    a ( x )  xn n x) - Estimate the unknown d.o. rather than parameter     a ( x ) a ( x ) 0         1 2 ( ) ( ) 0    0 0 1 ( ) ( ) 1 3 3 * 2 a x a x a x a x  - Validate the entanglement/separability a ( x )  0,1 - Constr ct Construct confidence region (credible region) [a L (x), a R (x)]
  35. 35. Remarks for Non-statisticians Remarks 1. If the quantum state is completely known, then the distribution of measurement outcome is also known and the best action is chosen. 2. Action has to be made from finite data. 3. Precise estimation of the parameter is only a typical example.  ( ) POVM x a ( x ) (dx )   M(dx )
  36. 36. Loss Functions Performance of decision functions is compared by adopting the loss functions (smaller is better). Ex: Parameter Estimation Quantum Statistical Model {  ( ) :    R m }  Action space Loss function (squared error) 2 w( , a )    a Later we see the formal definition of the loss function.
  37. 37. Measurements Over Decisions Basic Idea From the g g, beginning, we y only consider POVMs over the action p space. Put together  ( ) POVM x a ( x )  (dx ) M(dx )  ( da )   POVM a N(da )  ( )
  38. 38. Quantum Statistical Decision Problems Formal Definitions  Parameter space (locally compact space)  ( ) Quantum statistical model U *Decision p space ( y locally p compact p ) space) w :  U  R  { } Loss function (lower semi continuous**; bounded below) The triplet ( ( ),U , w) l d bl is called a quantum statistical decision problem. cf) statistical decision ( p U w) problem ,,  p (dx ) Statistical model  U Decision space w :  U  R  { } Loss function *we follow Holevo’s terminology instead of saying “action space”. ** we impose a slightly stronger condition on the loss than LeCam, Holevo.
  39. 39. 4. Risk Functions and Optimality Criteria
  40. 40. Comparison of POVMs The loss (e.g. squared error) depends on both unknown parameter  and our decision u u, which is a random variable variable.( (u ~   (du )  Tr ( )M(du ) ) In order to compare two POVMs, we focus on the average of the loss w w.r r.t t. the distribution of u u. ( ) POVM E[w( , u )]  u (du )   M(du ) N(du ' ) Compared at the same  POVM E'[w( ,u')]  ( ) u ' (du ' )  
  41. 41. D fi iti Risk Functions Definition The risk function of M  Po (U ) R ( ) : w( , u )Tr ( )M(d u ) M     E w( , u )     U u ~ Tr ( )M(d u ) R kf N i i i 1. Smaller risk is better. Remarks for Non-statisticians 2. Generally there exists no POVM achieving the uniformly smallest risk among POVMs. Since R M (  ) depends on the unknown parameter, we need additional optimality criteria.
  42. 42. Optimality Criteria (1/2) Definition A POVM is said to be minimax if it minimizes the supremum of the risk function, i.e., sup ( ) inf sup ( ) M M R   R  M* M ( )  Po U    Historical Remark Bogomolov showed the existence of a minimax POVM (Bogomolov 1981) in a more general framework. (Description of quantum states and measurements is different from recent one.)
  43. 43. Optimality Criteria (2/2) P() All probability distributions (defined on Borel sets) In y Bayesian , statistics,   ( P() is called a p prior distribution. Definition The average risk of M  P o ( U ) w.r.t.   R (  ) (d  ) : ,M M   r  A POVM is said to be Bayes if it minimizes the average risk, i.e., ,M M ( ) ,M inf    r r Po U  Historical Remark Holevo showed the existence of a Bayes POVM (Holevo 1976; see also Ozawa, 1980) in a more general framework.
  44. 44. Ex. of Loss Functions and Risk Functions Parameter Estimation { ( ) :    R m } U   2 w( , u )    u R ( ) u Tr ( )M(d u ) 2 M     E [ ] 2    u     U Construction of Predictive Density Operator { ( ) n :    R p } U  S (H  m ) w( , u )  D ( ( )  m || u )     R ( ) D( ( ) m || u )Tr ( ) m M(d u ) M      U
  45. 45. 5. Main Result
  46. 46. Main Result (Quantum Minimax Theorem) Theorem (*) Let  and U be a compact metric space. Then the following equality holds for every quantum statistical decision problem.       R R inf sup ( )  sup inf ( ) (d ) M ( ) M ( ) M ( ) M     Po U   P Po U   R ( ) : w( , u )Tr ( )M(d u ) M where the risk function is given by     U Corollary y For every closed convex subset Q  Po (U ) the above assertion holds. *see quant-ph 1410.3639 for proof
  47. 47. Key Lemmas For Theorem Compactness of the POVMs Holevo (1976) If the decision space U is compact, then P Po (U ) is also compact. Equicontinuity of risk functions Lemma by FT w :   U  R Loss function If w is bounded and continuous, then : { : M ( )} ( ) The equicontinuity implies M F  R  Po U  C  is (uniformly) equicontinuous. F  C () the compactness of under the uniform convergence topology We show main theorem by using Le Cam’s result with both lemmas. However, not a consequence of their previous old results.
  48. 48. 6. Minimax POVMs and Least Favorable Priors
  49. 49. Previous Works (LFPs) Wald (1950) Statistical Decision Theory Jeffreys (1961) Jeffreys prior Bernardo ( , 1979, ) 2005) Reference prior; reference analysis Komaki (2011) Latent information prior Tanaka (2012) MDP p prior ( Reference p prior for p pure-states model) ) Objective priors in Bayesian analysis are indeed least favorable priors.
  50. 50. Main Result (Existence Theorem of LFP) Definition If a prior supremum, i.e., achieves the supremum i e the following holds  R    R         inf ( ) (d )  sup inf ( ) (d ) M ( ) M ( ) M ( ) M P Po U Po U LF  The prior L F is called a least favorable prior(LFP).    R ( ) : w( , u )Tr ( )M(du ) M where     U Theorem Let  and U be a p compact metric p space. w: continuous loss function. Then, for every decision problem, there exists a LFP.
  51. 51. Pathological Example Remark Even on a compact space, a bounded lower semicontinuous (and not continuous) loss function does not necessarily admit a LFP. Example   [0,1] R( ) ( )  ( , )  M R  L  u 1  ( 1 0  0 1    M M 1 sup R ( ) sup R ( ) (d )  [ 0 ,1] [ 0 ,1]   But for every prior, 1   R ( ) (d ) [ 0 ,1] M
  52. 52. Minimax POVM is constructed using LFP Corollary Let  and U be a compact space metric space. If a LFP exists for a quantum statistical decision problem, Every minimax POVM is a Bayes POVM with LFP respect to the LFP. In particular, if the Bayes POVM w.r.t. the LFP is unique, then it is minimax. y Corollary2 For every closed convex subset Q  Po (U ) the above assertion holds. Much of theoretical results are derived from quantum minimax theorem.
  53. 53. 7. Summary
  54. 54. Discussion 1. We show quantum minimax theorem, which gives theoretical basis in quantum statistical decision theory, in particular in objective Bayesian analysis. 2. For every closed convex subset of POVMs, all of assertions still hold. Our result has also practical meanings for experimenters. Future Works FutureWorks - Show other decision-theoretic results and previous known results by using quantum minimax theorem - Propose a practical algorithm of finding a least favorable prior - Define geometrical structure in the quantum statistical decision problem If you are interested in details, please see quant-ph 1410.3639 ; citations are also welcome!

×