SlideShare a Scribd company logo
1 of 39
Symmetry Reduction of Information Inequalities
Kai Zhang and Chao Tian
Department of Electrical Engineering and Computer Science
The University of Tennessee Knoxville (UTK)
Sep. 2016
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 1 / 16
Outline
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 2 / 16
Symmetry Reduction
Conclusion
Motivation and Backgrounds
Network Coded Caching
Extremal Pairwise Cyclically Symmetric Entropy Inequalities
Regenerating Code
Outline
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 2 / 16
Symmetry Reduction
Conclusion
Motivation and Backgrounds
Network Coded Caching
Extremal Pairwise Cyclically Symmetric Entropy Inequalities
Regenerating Code
The full message can be recovered by accessing nodes, a failure node can be repaired
by accessing nodes, each transmitting units of data.
nodes each with capacity , the message is stored (coded) across these nodes;
Regenerating Code
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 3 / 16
Proposed by Dimakis, et al (2010) to provide efficient repair:
e.g.
What is the fundamental limit of memory vs. transmission rate ?
Regenerating Code
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 3 / 16
e.g.
What is the fundamental limit of memory vs. transmission rate ?
nodes each with capacity , the message is stored (coded) across these nodes;
The full message can be recovered by accessing nodes, a failure node can be repaired
by accessing nodes, each transmitting units of data.
Proposed by Dimakis, et al (2010) to provide efficient repair:
files, users, each user has a cache of size ;
Network Coded Caching
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 4 / 16
Proposed by Maddah-Ali and Niesen (2014) to reduce network congestion:
e.g.
What is the fundamental limit of memory vs. transmission rate ?
Placement phase: part of each file is pre-fetched to each user’s cache;
Delivery phase: users make requests, server multicasts information to fulfill requests,
the transmission rate is .
files, users, each user has a cache of size ;
Placement phase: part of each file is pre-fetched to each user’s cache;
Network Coded Caching
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 4 / 16
Proposed by Maddah-Ali and Niesen (2014) to reduce network congestion;
e.g.
What is the fundamental limit of memory vs. transmission rate ?
Delivery phase: users make requests, server multicasts information to fulfill requests,
the transmission rate is .
The Fundamental Limits
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 5 / 16
Fundamental limits are derived by its inner-bound and outer-bound:
A recent finding: A Computer-aided approach can be used. [Tian, 2014]
Outer-bound: find new inequalities by manipulating information inequalities (difficult);
Inner-bound: propose new coding schemes;
 Trial and error: almost impossible for long proofs.
Example: A new inequality found for (4, 3, 3) regenerating code problem:
 Tons of them to choose from;
Example: In (4, 3, 3) regenerating code, about 2 million of them.
The Fundamental Limits
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 5 / 16
Outer-bound: find new inequalities by manipulating information inequalities (difficult);
 Trial and error: almost impossible for long proofs.
Fundamental limits are derived by its inner-bound and outer-bound:
A recent finding: A Computer-aided approach can be used. [Tian, 2014]
Example: A new inequality found for (4, 3, 3) regenerating code problem:
Inner-bound: propose new coding schemes;
 Tons of them to choose from;
Example: In (4, 3, 3) regenerating code, about 2 million of them.
A Computer-aided Approach
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 6 / 16
Originally proposed by Yeung (1997), use Linear Program (LP) to prove information inequalities:
Use joint-entropy terms as LP variables, e.g. , etc.;
Use information inequalities as LP constraints, e.g.,
Only Shannon-type inequalities are used, but sufficient for most applications.
However, this method is cursed by dimensionality:
In our problem: many variables and constraints are symmetric Symmetry Reduction
In our problem: Regenerating Code: ; Caching: ;
Shannon-type Inequality (Yeung)
Almost all the information inequalities known to data are implied by the basic
inequalities (non-negativity of two elementary forms):
random variable LP problem: LP variables and LP constraints ;
A Computer-aided Approach
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 6 / 16
Use information inequalities as LP constraints, e.g.,
Only Shannon-type inequalities are used, but sufficient for most applications.
However, this method is cursed by dimensionality:
random variable LP problem: LP variables and LP constraints ;
Shannon-type Inequality (Yeung)
Almost all the information inequalities known to data are implied by the basic
inequalities (non-negativity of two elementary forms):
In our problem: Regenerating Code: ; Caching: ;
Originally proposed by Yeung (1997), use Linear Program (LP) to prove information inequalities:
Use joint-entropy terms as LP variables, e.g. , etc.;
In our problem: many variables and constraints are symmetric Symmetry Reduction
Outline
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16
Symmetry Reduction
Conclusion
Motivation and Backgrounds
Extremal Pairwise Cyclically Symmetric Entropy Inequalities
Regenerating Code
Network Coded Caching
Color set: ,
Background: Pólya’s Counting Theorem
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16
Example: Color a 2x2 chessboard using two colors , how many different ways?
Notice: we do not distinguish a chessboard with its rotations ;
Solved by Pólya (1937): a group of operations permutation group G.
In this example, the chessboard is labeled as ,
Base set: ;
Permutation group: ;
The different ways to color a group of objects can be given by the coefficients of the
cycle index , substituting each by the summation of all color symbol’s taking
power,
where .
Pólya’s Counting Theorem
Critical: “correctly define ” & “find its cycle index ”.
Color set: ,
Background: Pólya’s Counting Theorem
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16
In this example, the chessboard is labeled as ,
Base set: ,
Permutation group: ,
The different ways to color a group of objects can be given by the coefficients of the
cycle index , substituting each by the summation of all color symbol’s taking
power,
where .
Pólya’s Counting Theorem
Example: Color a 2x2 chessboard using two colors , how many different ways?
Notice: we do not distinguish a chessboard with its rotations ;
Solved by Pólya (1937): a group of operations permutation group G.
Critical: “correctly define ” & “find its cycle index ”.
Color set: ,
Background: Pólya’s Counting Theorem
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16
In this example, the chessboard is labeled as ,
Base set: ,
Permutation group: ,
The different ways to color a group of objects can be given by the coefficients of the
cycle index , substituting each by the summation of all color symbol’s taking
power,
where .
Pólya’s Counting Theorem
Example: Color a 2x2 chessboard using two colors , how many different ways?
Notice: we do not distinguish a chessboard with its rotations ;
Solved by Pólya (1937): a group of operations permutation group G.
Critical: “correctly define ” & “find its cycle index ”.
Background: Pólya’s Counting Theorem
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16
In this example, the chessboard is labeled as ,
Critical: “correctly define ” & “find its cycle index ”.
In our problem:
Not easy.Color set: ,
Base set: ,
Permutation group: ,
The different ways to color a group of objects can be given by the coefficients of the
cycle index , substituting each by the summation of all color symbol’s taking
power,
where .
Pólya’s Counting Theorem
Example: Color a 2x2 chessboard using two colors , how many different ways?
Notice: we do not distinguish a chessboard with its rotations ;
Solved by Pólya (1937): a group of operations permutation group G.
Second layer: the random variable set , with the induced permutation group ;
A 3-layer General Framework
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 8 / 16
First layer has a trivial symmetry group;
A few notes:
Second layer is problem specific, difficult part is the study of cycle index of ;
Third layer is what we’re interested: LP variables are represented as elements in .
Study of second layer is already difficult, then third layer ??
Get some knowledge of third layer by studying second layer & coloring problem.
First layer: the (maybe multiple) base set(s) , with the symmetry group ;
Third layer: the power set , with the induced permutation group .
3-layer framework:
A 3-layer General Framework
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 8 / 16
Study of second layer is already difficult, then third layer ??
Get some knowledge of third layer by studying second layer & coloring problem.
Second layer: the random variable set , with the induced permutation group ;
First layer has a trivial symmetry group;
A few notes:
Second layer is problem specific, difficult part is the study of cycle index of ;
Third layer is what we’re interested: LP variables are represented as elements in .
First layer: the (maybe multiple) base set(s) , with the symmetry group ;
Third layer: the power set , with the induced permutation group .
3-layer framework:
A 3-layer General Framework
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 8 / 16
Study of second layer is already difficult, then third layer ??
Get some knowledge of third layer by studying second layer & coloring problem.
Second layer: the random variable set , with the induced permutation group ;
First layer has a trivial symmetry group;
A few notes:
Second layer is problem specific, difficult part is the study of cycle index of ;
Third layer is what we’re interested: LP variables are represented as elements in .
First layer: the (maybe multiple) base set(s) , with the symmetry group ;
Third layer: the power set , with the induced permutation group .
3-layer framework:
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 9 / 16
For an regenerating code problem, the random variable set is ,
name change
name change
The “name change” can be represented mathematically by permutations ;
All possible ways of changing names can form a permutation group .
name change
Symmetry Reduction: (4,3,3) Regenerating Code
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 9 / 16
For an regenerating code problem, the random variable set is ,
The “name change” can be represented mathematically by permutations ;
All possible ways of changing names can form a permutation group .
Symmetry Reduction: (4,3,3) Regenerating Code
Induced Permutation
Base set has a permutation group and its cycle index ,
Group:
Cycle index:
induce
Random variable set has a permutation group and its cycle
index ,
Group:
Cycle index: check every permutation in ?
The cycle index of the group on base set
Cycle Index in Regenerating Code Problem
will induce a term cycle index of the group on base set
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16
of them.
Induced Permutation
Base set has a permutation group and its cycle index ,
Group:
Cycle index:
induce
Random variable set has a permutation group and its cycle
index ,
Group:
Cycle index: check every permutation in ?
The cycle index of the group on base set
Cycle Index in Regenerating Code Problem
will induce a term cycle index of the group on base set
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16
of them.
Induced Permutation
Base set has a permutation group and its cycle index ,
Group:
Cycle index:
induce
Random variable set has a permutation group and its cycle
index ,
Group:
Cycle index: check every permutation in ?
The cycle index of the group on base set
Cycle Index in Regenerating Code Problem
will induce a term cycle index of the group on base set
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16
of them.
Induced Permutation
Base set has a permutation group and its cycle index ,
Group:
Cycle index:
induce
Random variable set has a permutation group and its cycle
index ,
Group:
Cycle index: check every permutation in ?
The cycle index of the group on base set
Cycle Index in Regenerating Code Problem
will induce a term cycle index of the group on base set
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16
of them.
Induced Permutation
Base set has a permutation group and its cycle index ,
Group:
Cycle index:
induce
Random variable set has a permutation group and its cycle
index ,
Group:
Cycle index: check every permutation in ?
The cycle index of the group on base set
Cycle Index in Regenerating Code Problem
will induce a term cycle index of the group on base set
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16
of them.
Induced Permutation
The cycle index of the group on base set
Theorem: Cycle Index in Regenerating Code
will induce a term cycle index of the group on base set
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16
Base set has a permutation group and its cycle index ,
Group:
Cycle index:
induce
Random variable set has a permutation group and its cycle
index ,
Group:
Cycle index:
Example: (4,3,3) Regenerating Code
Cycle index:
Number of LP variables ( ):
Intuition: color red denotes , color green denotes ;
The number of variables is the summation of coefficients of minus 1 ;
Note: a simple trick to sum the coefficients is to assign ;
Number of LP constraints:
Type 1: is trivial, use two colors and result is the
coefficient before , result is 2;
Result is , far less than the original .
Type 2: , use three colors
and result is the coefficient before all terms containing , result is 83200;
Result is , far less than the original .
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 11 / 16
Cycle index:
Number of LP variables ( ):
Intuition: color red denotes , color green denotes ;
The number of variables is the summation of coefficients of minus 1 ;
Note: a simple trick to sum the coefficients is to assign ;
Number of LP constraints:
Type 1: is trivial, use two colors and result is the
coefficient before , result is 2;
Result is , far less than the original .
Type 2: , use three colors
and result is the coefficient before all terms containing , result is 83200;
Result is , far less than the original .
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 11 / 16
Example: (4,3,3) Regenerating Code
Symmetry Reduction: Network Coded Caching
Variables: files , users , and transmissions
File index set , with the symmetric group ;
represents different demands.
Two base sets:
User index set , with the symmetric group ;
Induced set:
Example:
File index permutation : ;
with the induced permutation group ;
User index permutation : ;
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 12 / 16
This time, we have to check some permutations to derive .
Fortunately, not all ( ).
Example: N=4, K=3 Cache Network
The file index permutation group and the user index permutation group , each
will have its own cycle index,
The goal: finding the cycle index of the induced set :
, then the permutation induce by and will also have the
If two permutations have the same cycle index, and the same with
Proposition: Cycle Index in Caching
same cycle index.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 13 / 16
Result: LP variables , LP constraints .
The file index permutation group and the user index permutation group , each
will have its own cycle index,
The goal: finding the cycle index of the induced set :
, then the permutation induce by and will also have the
If two permutations have the same cycle index, and the same with
Proposition: Cycle Index in Caching
same cycle index.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 13 / 16
Example: N=4, K=3 Cache Network
This time, we have to check some permutations to derive .
Fortunately, not all ( ).
Result: LP variables , LP constraints .
The file index permutation group and the user index permutation group , each
will have its own cycle index,
The goal: finding the cycle index of the induced set :
, then the permutation induce by and will also have the
If two permutations have the same cycle index, and the same with
Proposition: Cycle Index in Caching
same cycle index.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities
Example: N=4, K=3 Cache Network
Sep. 2016 13 / 16
This time, we have to check some permutations to derive .
Fortunately, not all ( ).
Result: LP variables , LP constraints .
The file index permutation group and the user index permutation group , each
will have its own cycle index,
The goal: finding the cycle index of the induced set :
, then the permutation induce by and will also have the
If two permutations have the same cycle index, and the same with
Proposition: Cycle Index in Caching
same cycle index.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities
Example: N=4, K=3 Cache Network
Sep. 2016 13 / 16
This time, we have to check some permutations to derive .
Fortunately, not all ( ).
Result: LP variables , LP constraints .
, then the permutation induce by and will also have the
The file index permutation group and the user index permutation group , each
will have its own cycle index,
The goal: finding the cycle index of the induced set :
If two permutations have the same cycle index, and the same with
Proposition: Cycle Index in Caching
same cycle index.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities
Example: N=4, K=3 Cache Network
Sep. 2016 13 / 16
This time, we have to check some permutations to derive .
Fortunately, not all ( ).
Result: LP variables , LP constraints .
, then the permutation induce by and will also have the
The file index permutation group and the user index permutation group , each
will have its own cycle index,
The goal: finding the cycle index of the induced set :
If two permutations have the same cycle index, and the same with
Proposition: Cycle Index in Caching
same cycle index.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities
Example: N=4, K=3 Cache Network
Sep. 2016 13 / 16
This time, we have to check some permutations to derive .
Fortunately, not all ( ).
Result: LP variables , LP constraints .
Outline
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 14 / 16
Symmetry Reduction
Conclusion
Motivation and Backgrounds
Extremal Pairwise Cyclically Symmetric Entropy Inequalities
Regenerating Code
Network Coded Caching
Conclusion
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 15 / 16
Contributions:
Symmetry is carefully defined in our problems;
A general framework for different research problems;
Counting the LP variables/constraints coloring problem.
Challenge:
Counting enumerating;
Counting symmetry reduction, there are other problem specific reduction.
K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities
Thank you!
Sep. 2016 16 / 16

More Related Content

Similar to Symmetry Reduction of Information Inequalities

Paper Explained: Understanding the wiring evolution in differentiable neural ...
Paper Explained: Understanding the wiring evolution in differentiable neural ...Paper Explained: Understanding the wiring evolution in differentiable neural ...
Paper Explained: Understanding the wiring evolution in differentiable neural ...
Devansh16
 

Similar to Symmetry Reduction of Information Inequalities (20)

Joco pavone
Joco pavoneJoco pavone
Joco pavone
 
Pula 5 Giugno 2007
Pula 5 Giugno 2007Pula 5 Giugno 2007
Pula 5 Giugno 2007
 
Paper Explained: Understanding the wiring evolution in differentiable neural ...
Paper Explained: Understanding the wiring evolution in differentiable neural ...Paper Explained: Understanding the wiring evolution in differentiable neural ...
Paper Explained: Understanding the wiring evolution in differentiable neural ...
 
CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...
CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...
CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...
 
CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...
CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...
CASCADE BLOCK CIPHER USING BRAIDING/ENTANGLEMENT OF SPIN MATRICES AND BIT ROT...
 
6조
6조6조
6조
 
Modification of some solution techniques of combinatorial
Modification of some solution techniques of combinatorialModification of some solution techniques of combinatorial
Modification of some solution techniques of combinatorial
 
Loss less DNA Solidity Using Huffman and Arithmetic Coding
Loss less DNA Solidity Using Huffman and Arithmetic CodingLoss less DNA Solidity Using Huffman and Arithmetic Coding
Loss less DNA Solidity Using Huffman and Arithmetic Coding
 
clustering.pptx
clustering.pptxclustering.pptx
clustering.pptx
 
Probabilistic Modelling with Information Filtering Networks
Probabilistic Modelling with Information Filtering NetworksProbabilistic Modelling with Information Filtering Networks
Probabilistic Modelling with Information Filtering Networks
 
Simplicial closure and higher-order link prediction LA/OPT
Simplicial closure and higher-order link prediction LA/OPTSimplicial closure and higher-order link prediction LA/OPT
Simplicial closure and higher-order link prediction LA/OPT
 
lecture12-clustering.ppt
lecture12-clustering.pptlecture12-clustering.ppt
lecture12-clustering.ppt
 
lecture12-clustering.ppt
lecture12-clustering.pptlecture12-clustering.ppt
lecture12-clustering.ppt
 
lecture12-clustering.ppt
lecture12-clustering.pptlecture12-clustering.ppt
lecture12-clustering.ppt
 
lecture12-clustering.ppt
lecture12-clustering.pptlecture12-clustering.ppt
lecture12-clustering.ppt
 
Propagation of Policies in Rich Data Flows
Propagation of Policies in Rich Data FlowsPropagation of Policies in Rich Data Flows
Propagation of Policies in Rich Data Flows
 
K mer index of dna sequence based on hash
K mer index of dna sequence based on hashK mer index of dna sequence based on hash
K mer index of dna sequence based on hash
 
Improvement in Traditional Set Partitioning in Hierarchical Trees (SPIHT) Alg...
Improvement in Traditional Set Partitioning in Hierarchical Trees (SPIHT) Alg...Improvement in Traditional Set Partitioning in Hierarchical Trees (SPIHT) Alg...
Improvement in Traditional Set Partitioning in Hierarchical Trees (SPIHT) Alg...
 
Matrix Factorizations at Scale: a Comparison of Scientific Data Analytics on ...
Matrix Factorizations at Scale: a Comparison of Scientific Data Analytics on ...Matrix Factorizations at Scale: a Comparison of Scientific Data Analytics on ...
Matrix Factorizations at Scale: a Comparison of Scientific Data Analytics on ...
 
Interpretability of machine learning
Interpretability of machine learningInterpretability of machine learning
Interpretability of machine learning
 

Recently uploaded

Maher Othman Interior Design Portfolio..
Maher Othman Interior Design Portfolio..Maher Othman Interior Design Portfolio..
Maher Othman Interior Design Portfolio..
MaherOthman7
 

Recently uploaded (20)

Diploma Engineering Drawing Qp-2024 Ece .pdf
Diploma Engineering Drawing Qp-2024 Ece .pdfDiploma Engineering Drawing Qp-2024 Ece .pdf
Diploma Engineering Drawing Qp-2024 Ece .pdf
 
analog-vs-digital-communication (concept of analog and digital).pptx
analog-vs-digital-communication (concept of analog and digital).pptxanalog-vs-digital-communication (concept of analog and digital).pptx
analog-vs-digital-communication (concept of analog and digital).pptx
 
15-Minute City: A Completely New Horizon
15-Minute City: A Completely New Horizon15-Minute City: A Completely New Horizon
15-Minute City: A Completely New Horizon
 
engineering chemistry power point presentation
engineering chemistry  power point presentationengineering chemistry  power point presentation
engineering chemistry power point presentation
 
Basics of Relay for Engineering Students
Basics of Relay for Engineering StudentsBasics of Relay for Engineering Students
Basics of Relay for Engineering Students
 
21scheme vtu syllabus of visveraya technological university
21scheme vtu syllabus of visveraya technological university21scheme vtu syllabus of visveraya technological university
21scheme vtu syllabus of visveraya technological university
 
UNIT-2 image enhancement.pdf Image Processing Unit 2 AKTU
UNIT-2 image enhancement.pdf Image Processing Unit 2 AKTUUNIT-2 image enhancement.pdf Image Processing Unit 2 AKTU
UNIT-2 image enhancement.pdf Image Processing Unit 2 AKTU
 
5G and 6G refer to generations of mobile network technology, each representin...
5G and 6G refer to generations of mobile network technology, each representin...5G and 6G refer to generations of mobile network technology, each representin...
5G and 6G refer to generations of mobile network technology, each representin...
 
UNIT 4 PTRP final Convergence in probability.pptx
UNIT 4 PTRP final Convergence in probability.pptxUNIT 4 PTRP final Convergence in probability.pptx
UNIT 4 PTRP final Convergence in probability.pptx
 
Research Methodolgy & Intellectual Property Rights Series 1
Research Methodolgy & Intellectual Property Rights Series 1Research Methodolgy & Intellectual Property Rights Series 1
Research Methodolgy & Intellectual Property Rights Series 1
 
Passive Air Cooling System and Solar Water Heater.ppt
Passive Air Cooling System and Solar Water Heater.pptPassive Air Cooling System and Solar Water Heater.ppt
Passive Air Cooling System and Solar Water Heater.ppt
 
Independent Solar-Powered Electric Vehicle Charging Station
Independent Solar-Powered Electric Vehicle Charging StationIndependent Solar-Powered Electric Vehicle Charging Station
Independent Solar-Powered Electric Vehicle Charging Station
 
Maher Othman Interior Design Portfolio..
Maher Othman Interior Design Portfolio..Maher Othman Interior Design Portfolio..
Maher Othman Interior Design Portfolio..
 
Interfacing Analog to Digital Data Converters ee3404.pdf
Interfacing Analog to Digital Data Converters ee3404.pdfInterfacing Analog to Digital Data Converters ee3404.pdf
Interfacing Analog to Digital Data Converters ee3404.pdf
 
NO1 Best Powerful Vashikaran Specialist Baba Vashikaran Specialist For Love V...
NO1 Best Powerful Vashikaran Specialist Baba Vashikaran Specialist For Love V...NO1 Best Powerful Vashikaran Specialist Baba Vashikaran Specialist For Love V...
NO1 Best Powerful Vashikaran Specialist Baba Vashikaran Specialist For Love V...
 
handbook on reinforce concrete and detailing
handbook on reinforce concrete and detailinghandbook on reinforce concrete and detailing
handbook on reinforce concrete and detailing
 
Seismic Hazard Assessment Software in Python by Prof. Dr. Costas Sachpazis
Seismic Hazard Assessment Software in Python by Prof. Dr. Costas SachpazisSeismic Hazard Assessment Software in Python by Prof. Dr. Costas Sachpazis
Seismic Hazard Assessment Software in Python by Prof. Dr. Costas Sachpazis
 
Filters for Electromagnetic Compatibility Applications
Filters for Electromagnetic Compatibility ApplicationsFilters for Electromagnetic Compatibility Applications
Filters for Electromagnetic Compatibility Applications
 
CLOUD COMPUTING SERVICES - Cloud Reference Modal
CLOUD COMPUTING SERVICES - Cloud Reference ModalCLOUD COMPUTING SERVICES - Cloud Reference Modal
CLOUD COMPUTING SERVICES - Cloud Reference Modal
 
History of Indian Railways - the story of Growth & Modernization
History of Indian Railways - the story of Growth & ModernizationHistory of Indian Railways - the story of Growth & Modernization
History of Indian Railways - the story of Growth & Modernization
 

Symmetry Reduction of Information Inequalities

  • 1. Symmetry Reduction of Information Inequalities Kai Zhang and Chao Tian Department of Electrical Engineering and Computer Science The University of Tennessee Knoxville (UTK) Sep. 2016 K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 1 / 16
  • 2. Outline K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 2 / 16 Symmetry Reduction Conclusion Motivation and Backgrounds Network Coded Caching Extremal Pairwise Cyclically Symmetric Entropy Inequalities Regenerating Code
  • 3. Outline K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 2 / 16 Symmetry Reduction Conclusion Motivation and Backgrounds Network Coded Caching Extremal Pairwise Cyclically Symmetric Entropy Inequalities Regenerating Code
  • 4. The full message can be recovered by accessing nodes, a failure node can be repaired by accessing nodes, each transmitting units of data. nodes each with capacity , the message is stored (coded) across these nodes; Regenerating Code K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 3 / 16 Proposed by Dimakis, et al (2010) to provide efficient repair: e.g. What is the fundamental limit of memory vs. transmission rate ?
  • 5. Regenerating Code K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 3 / 16 e.g. What is the fundamental limit of memory vs. transmission rate ? nodes each with capacity , the message is stored (coded) across these nodes; The full message can be recovered by accessing nodes, a failure node can be repaired by accessing nodes, each transmitting units of data. Proposed by Dimakis, et al (2010) to provide efficient repair:
  • 6. files, users, each user has a cache of size ; Network Coded Caching K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 4 / 16 Proposed by Maddah-Ali and Niesen (2014) to reduce network congestion: e.g. What is the fundamental limit of memory vs. transmission rate ? Placement phase: part of each file is pre-fetched to each user’s cache; Delivery phase: users make requests, server multicasts information to fulfill requests, the transmission rate is .
  • 7. files, users, each user has a cache of size ; Placement phase: part of each file is pre-fetched to each user’s cache; Network Coded Caching K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 4 / 16 Proposed by Maddah-Ali and Niesen (2014) to reduce network congestion; e.g. What is the fundamental limit of memory vs. transmission rate ? Delivery phase: users make requests, server multicasts information to fulfill requests, the transmission rate is .
  • 8. The Fundamental Limits K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 5 / 16 Fundamental limits are derived by its inner-bound and outer-bound: A recent finding: A Computer-aided approach can be used. [Tian, 2014] Outer-bound: find new inequalities by manipulating information inequalities (difficult); Inner-bound: propose new coding schemes;  Trial and error: almost impossible for long proofs. Example: A new inequality found for (4, 3, 3) regenerating code problem:  Tons of them to choose from; Example: In (4, 3, 3) regenerating code, about 2 million of them.
  • 9. The Fundamental Limits K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 5 / 16 Outer-bound: find new inequalities by manipulating information inequalities (difficult);  Trial and error: almost impossible for long proofs. Fundamental limits are derived by its inner-bound and outer-bound: A recent finding: A Computer-aided approach can be used. [Tian, 2014] Example: A new inequality found for (4, 3, 3) regenerating code problem: Inner-bound: propose new coding schemes;  Tons of them to choose from; Example: In (4, 3, 3) regenerating code, about 2 million of them.
  • 10. A Computer-aided Approach K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 6 / 16 Originally proposed by Yeung (1997), use Linear Program (LP) to prove information inequalities: Use joint-entropy terms as LP variables, e.g. , etc.; Use information inequalities as LP constraints, e.g., Only Shannon-type inequalities are used, but sufficient for most applications. However, this method is cursed by dimensionality: In our problem: many variables and constraints are symmetric Symmetry Reduction In our problem: Regenerating Code: ; Caching: ; Shannon-type Inequality (Yeung) Almost all the information inequalities known to data are implied by the basic inequalities (non-negativity of two elementary forms): random variable LP problem: LP variables and LP constraints ;
  • 11. A Computer-aided Approach K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 6 / 16 Use information inequalities as LP constraints, e.g., Only Shannon-type inequalities are used, but sufficient for most applications. However, this method is cursed by dimensionality: random variable LP problem: LP variables and LP constraints ; Shannon-type Inequality (Yeung) Almost all the information inequalities known to data are implied by the basic inequalities (non-negativity of two elementary forms): In our problem: Regenerating Code: ; Caching: ; Originally proposed by Yeung (1997), use Linear Program (LP) to prove information inequalities: Use joint-entropy terms as LP variables, e.g. , etc.; In our problem: many variables and constraints are symmetric Symmetry Reduction
  • 12. Outline K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16 Symmetry Reduction Conclusion Motivation and Backgrounds Extremal Pairwise Cyclically Symmetric Entropy Inequalities Regenerating Code Network Coded Caching
  • 13. Color set: , Background: Pólya’s Counting Theorem K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16 Example: Color a 2x2 chessboard using two colors , how many different ways? Notice: we do not distinguish a chessboard with its rotations ; Solved by Pólya (1937): a group of operations permutation group G. In this example, the chessboard is labeled as , Base set: ; Permutation group: ; The different ways to color a group of objects can be given by the coefficients of the cycle index , substituting each by the summation of all color symbol’s taking power, where . Pólya’s Counting Theorem Critical: “correctly define ” & “find its cycle index ”.
  • 14. Color set: , Background: Pólya’s Counting Theorem K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16 In this example, the chessboard is labeled as , Base set: , Permutation group: , The different ways to color a group of objects can be given by the coefficients of the cycle index , substituting each by the summation of all color symbol’s taking power, where . Pólya’s Counting Theorem Example: Color a 2x2 chessboard using two colors , how many different ways? Notice: we do not distinguish a chessboard with its rotations ; Solved by Pólya (1937): a group of operations permutation group G. Critical: “correctly define ” & “find its cycle index ”.
  • 15. Color set: , Background: Pólya’s Counting Theorem K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16 In this example, the chessboard is labeled as , Base set: , Permutation group: , The different ways to color a group of objects can be given by the coefficients of the cycle index , substituting each by the summation of all color symbol’s taking power, where . Pólya’s Counting Theorem Example: Color a 2x2 chessboard using two colors , how many different ways? Notice: we do not distinguish a chessboard with its rotations ; Solved by Pólya (1937): a group of operations permutation group G. Critical: “correctly define ” & “find its cycle index ”.
  • 16. Background: Pólya’s Counting Theorem K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 7 / 16 In this example, the chessboard is labeled as , Critical: “correctly define ” & “find its cycle index ”. In our problem: Not easy.Color set: , Base set: , Permutation group: , The different ways to color a group of objects can be given by the coefficients of the cycle index , substituting each by the summation of all color symbol’s taking power, where . Pólya’s Counting Theorem Example: Color a 2x2 chessboard using two colors , how many different ways? Notice: we do not distinguish a chessboard with its rotations ; Solved by Pólya (1937): a group of operations permutation group G.
  • 17. Second layer: the random variable set , with the induced permutation group ; A 3-layer General Framework K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 8 / 16 First layer has a trivial symmetry group; A few notes: Second layer is problem specific, difficult part is the study of cycle index of ; Third layer is what we’re interested: LP variables are represented as elements in . Study of second layer is already difficult, then third layer ?? Get some knowledge of third layer by studying second layer & coloring problem. First layer: the (maybe multiple) base set(s) , with the symmetry group ; Third layer: the power set , with the induced permutation group . 3-layer framework:
  • 18. A 3-layer General Framework K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 8 / 16 Study of second layer is already difficult, then third layer ?? Get some knowledge of third layer by studying second layer & coloring problem. Second layer: the random variable set , with the induced permutation group ; First layer has a trivial symmetry group; A few notes: Second layer is problem specific, difficult part is the study of cycle index of ; Third layer is what we’re interested: LP variables are represented as elements in . First layer: the (maybe multiple) base set(s) , with the symmetry group ; Third layer: the power set , with the induced permutation group . 3-layer framework:
  • 19. A 3-layer General Framework K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 8 / 16 Study of second layer is already difficult, then third layer ?? Get some knowledge of third layer by studying second layer & coloring problem. Second layer: the random variable set , with the induced permutation group ; First layer has a trivial symmetry group; A few notes: Second layer is problem specific, difficult part is the study of cycle index of ; Third layer is what we’re interested: LP variables are represented as elements in . First layer: the (maybe multiple) base set(s) , with the symmetry group ; Third layer: the power set , with the induced permutation group . 3-layer framework:
  • 20. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 9 / 16 For an regenerating code problem, the random variable set is , name change name change The “name change” can be represented mathematically by permutations ; All possible ways of changing names can form a permutation group . name change Symmetry Reduction: (4,3,3) Regenerating Code
  • 21. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 9 / 16 For an regenerating code problem, the random variable set is , The “name change” can be represented mathematically by permutations ; All possible ways of changing names can form a permutation group . Symmetry Reduction: (4,3,3) Regenerating Code
  • 22. Induced Permutation Base set has a permutation group and its cycle index , Group: Cycle index: induce Random variable set has a permutation group and its cycle index , Group: Cycle index: check every permutation in ? The cycle index of the group on base set Cycle Index in Regenerating Code Problem will induce a term cycle index of the group on base set K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16 of them.
  • 23. Induced Permutation Base set has a permutation group and its cycle index , Group: Cycle index: induce Random variable set has a permutation group and its cycle index , Group: Cycle index: check every permutation in ? The cycle index of the group on base set Cycle Index in Regenerating Code Problem will induce a term cycle index of the group on base set K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16 of them.
  • 24. Induced Permutation Base set has a permutation group and its cycle index , Group: Cycle index: induce Random variable set has a permutation group and its cycle index , Group: Cycle index: check every permutation in ? The cycle index of the group on base set Cycle Index in Regenerating Code Problem will induce a term cycle index of the group on base set K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16 of them.
  • 25. Induced Permutation Base set has a permutation group and its cycle index , Group: Cycle index: induce Random variable set has a permutation group and its cycle index , Group: Cycle index: check every permutation in ? The cycle index of the group on base set Cycle Index in Regenerating Code Problem will induce a term cycle index of the group on base set K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16 of them.
  • 26. Induced Permutation Base set has a permutation group and its cycle index , Group: Cycle index: induce Random variable set has a permutation group and its cycle index , Group: Cycle index: check every permutation in ? The cycle index of the group on base set Cycle Index in Regenerating Code Problem will induce a term cycle index of the group on base set K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16 of them.
  • 27. Induced Permutation The cycle index of the group on base set Theorem: Cycle Index in Regenerating Code will induce a term cycle index of the group on base set K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 10 / 16 Base set has a permutation group and its cycle index , Group: Cycle index: induce Random variable set has a permutation group and its cycle index , Group: Cycle index:
  • 28. Example: (4,3,3) Regenerating Code Cycle index: Number of LP variables ( ): Intuition: color red denotes , color green denotes ; The number of variables is the summation of coefficients of minus 1 ; Note: a simple trick to sum the coefficients is to assign ; Number of LP constraints: Type 1: is trivial, use two colors and result is the coefficient before , result is 2; Result is , far less than the original . Type 2: , use three colors and result is the coefficient before all terms containing , result is 83200; Result is , far less than the original . K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 11 / 16
  • 29. Cycle index: Number of LP variables ( ): Intuition: color red denotes , color green denotes ; The number of variables is the summation of coefficients of minus 1 ; Note: a simple trick to sum the coefficients is to assign ; Number of LP constraints: Type 1: is trivial, use two colors and result is the coefficient before , result is 2; Result is , far less than the original . Type 2: , use three colors and result is the coefficient before all terms containing , result is 83200; Result is , far less than the original . K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 11 / 16 Example: (4,3,3) Regenerating Code
  • 30. Symmetry Reduction: Network Coded Caching Variables: files , users , and transmissions File index set , with the symmetric group ; represents different demands. Two base sets: User index set , with the symmetric group ; Induced set: Example: File index permutation : ; with the induced permutation group ; User index permutation : ; K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 12 / 16
  • 31. This time, we have to check some permutations to derive . Fortunately, not all ( ). Example: N=4, K=3 Cache Network The file index permutation group and the user index permutation group , each will have its own cycle index, The goal: finding the cycle index of the induced set : , then the permutation induce by and will also have the If two permutations have the same cycle index, and the same with Proposition: Cycle Index in Caching same cycle index. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 13 / 16 Result: LP variables , LP constraints .
  • 32. The file index permutation group and the user index permutation group , each will have its own cycle index, The goal: finding the cycle index of the induced set : , then the permutation induce by and will also have the If two permutations have the same cycle index, and the same with Proposition: Cycle Index in Caching same cycle index. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 13 / 16 Example: N=4, K=3 Cache Network This time, we have to check some permutations to derive . Fortunately, not all ( ). Result: LP variables , LP constraints .
  • 33. The file index permutation group and the user index permutation group , each will have its own cycle index, The goal: finding the cycle index of the induced set : , then the permutation induce by and will also have the If two permutations have the same cycle index, and the same with Proposition: Cycle Index in Caching same cycle index. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Example: N=4, K=3 Cache Network Sep. 2016 13 / 16 This time, we have to check some permutations to derive . Fortunately, not all ( ). Result: LP variables , LP constraints .
  • 34. The file index permutation group and the user index permutation group , each will have its own cycle index, The goal: finding the cycle index of the induced set : , then the permutation induce by and will also have the If two permutations have the same cycle index, and the same with Proposition: Cycle Index in Caching same cycle index. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Example: N=4, K=3 Cache Network Sep. 2016 13 / 16 This time, we have to check some permutations to derive . Fortunately, not all ( ). Result: LP variables , LP constraints .
  • 35. , then the permutation induce by and will also have the The file index permutation group and the user index permutation group , each will have its own cycle index, The goal: finding the cycle index of the induced set : If two permutations have the same cycle index, and the same with Proposition: Cycle Index in Caching same cycle index. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Example: N=4, K=3 Cache Network Sep. 2016 13 / 16 This time, we have to check some permutations to derive . Fortunately, not all ( ). Result: LP variables , LP constraints .
  • 36. , then the permutation induce by and will also have the The file index permutation group and the user index permutation group , each will have its own cycle index, The goal: finding the cycle index of the induced set : If two permutations have the same cycle index, and the same with Proposition: Cycle Index in Caching same cycle index. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Example: N=4, K=3 Cache Network Sep. 2016 13 / 16 This time, we have to check some permutations to derive . Fortunately, not all ( ). Result: LP variables , LP constraints .
  • 37. Outline K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 14 / 16 Symmetry Reduction Conclusion Motivation and Backgrounds Extremal Pairwise Cyclically Symmetric Entropy Inequalities Regenerating Code Network Coded Caching
  • 38. Conclusion K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Sep. 2016 15 / 16 Contributions: Symmetry is carefully defined in our problems; A general framework for different research problems; Counting the LP variables/constraints coloring problem. Challenge: Counting enumerating; Counting symmetry reduction, there are other problem specific reduction.
  • 39. K. Zhang and C. Tian (UTK) Symmetry Reduction of Information Inequalities Thank you! Sep. 2016 16 / 16