Quantum Search Heuristics: Tad Hogg’s Perspective George Viamontes February 4, 2002.

Slides:



Advertisements
Similar presentations
Dana Nau: Lecture slides for Automated Planning Licensed under the Creative Commons Attribution-NonCommercial-ShareAlike License:
Advertisements

CS6800 Advanced Theory of Computation
Counting the bits Analysis of Algorithms Will it run on a larger problem? When will it fail?
Generating Hard Satisfiability Problems1 Bart Selman, David Mitchell, Hector J. Levesque Presented by Xiaoxin Yin.
The Theory of NP-Completeness
Quantum Packet Switching A. Yavuz Oruç Department of Electrical and Computer Engineering University of Maryland, College Park.
1 Introduction to Computability Theory Lecture12: Reductions Prof. Amos Israeli.
Beating Brute Force Search for Formula SAT and QBF SAT Rahul Santhanam University of Edinburgh.
KEG PARTY!!!!!  Keg Party tomorrow night  Prof. Markov will give out extra credit to anyone who attends* *Note: This statement is a lie.
MAE 552 – Heuristic Optimization Lecture 27 April 3, 2002
1 CSE 417: Algorithms and Computational Complexity Winter 2001 Lecture 21 Instructor: Paul Beame.
Ryan Kinworthy 2/26/20031 Chapter 7- Local Search part 1 Ryan Kinworthy CSCE Advanced Constraint Processing.
Grover. Part 2. Components of Grover Loop The Oracle -- O The Hadamard Transforms -- H The Zero State Phase Shift -- Z O is an Oracle H is Hadamards H.
High-Performance Simulation of Quantum Computation using QuIDDs George F. Viamontes, Manoj Rajagopalan, Igor L. Markov, and John P. Hayes Advanced Computer.
MAE 552 – Heuristic Optimization Lecture 26 April 1, 2002 Topic:Branch and Bound.
It’s all about the support: a new perspective on the satisfiability problem Danny Vilenchik.
Analysis of Algorithms CS 477/677
Grover. Part 2 Anuj Dawar. Components of Grover Loop The Oracle -- O The Hadamard Transforms -- H The Zero State Phase Shift -- Z.
2-Layer Crossing Minimisation Johan van Rooij. Overview Problem definitions NP-Hardness proof Heuristics & Performance Practical Computation One layer:
Study Group Randomized Algorithms Jun 7, 2003 Jun 14, 2003.
MAE 552 – Heuristic Optimization Lecture 5 February 1, 2002.
NP-complete and NP-hard problems. Decision problems vs. optimization problems The problems we are trying to solve are basically of two kinds. In decision.
1 CSE 417: Algorithms and Computational Complexity Winter 2001 Lecture 25 Instructor: Paul Beame.
A Fault-tolerant Architecture for Quantum Hamiltonian Simulation Guoming Wang Oleg Khainovski.
Ryan Kinworthy 2/26/20031 Chapter 7- Local Search part 2 Ryan Kinworthy CSCE Advanced Constraint Processing.
1 Message Passing and Local Heuristics as Decimation Strategies for Satisfiability Lukas Kroc, Ashish Sabharwal, Bart Selman (presented by Sebastian Brand)
Introduction to Simulated Annealing 22c:145 Simulated Annealing  Motivated by the physical annealing process  Material is heated and slowly cooled.
1.1 Chapter 1: Introduction What is the course all about? Problems, instances and algorithms Running time v.s. computational complexity General description.
Quantum Error Correction Jian-Wei Pan Lecture Note 9.
Vilalta&Eick: Informed Search Informed Search and Exploration Search Strategies Heuristic Functions Local Search Algorithms Vilalta&Eick: Informed Search.
1 The Theory of NP-Completeness 2012/11/6 P: the class of problems which can be solved by a deterministic polynomial algorithm. NP : the class of decision.
Nattee Niparnan. Easy & Hard Problem What is “difficulty” of problem? Difficult for computer scientist to derive algorithm for the problem? Difficult.
Cove: A Practical Quantum Computer Programming Framework Matt Purkeypile Doctorate of Computer Science Dissertation Defense June 26, 2009.
Tonga Institute of Higher Education Design and Analysis of Algorithms IT 254 Lecture 8: Complexity Theory.
NP Complexity By Mussie Araya. What is NP Complexity? Formal Definition: NP is the set of decision problems solvable in polynomial time by a non- deterministic.
A DNA-Based Random Walk Method for Solving k-SAT Sergio Diaz, Juan Luis Esteban and Mitsunori Ogihara.
Major objective of this course is: Design and analysis of modern algorithms Different variants Accuracy Efficiency Comparing efficiencies Motivation thinking.
Week 10Complexity of Algorithms1 Hard Computational Problems Some computational problems are hard Despite a numerous attempts we do not know any efficient.
P, NP, and Exponential Problems Should have had all this in CS 252 – Quick review Many problems have an exponential number of possibilities and we can.
Boolean Minimizer FC-Min: Coverage Finding Process Petr Fišer, Hana Kubátová Czech Technical University Department of Computer Science and Engineering.
Monte-Carlo methods for Computation and Optimization Spring 2015 Based on “N-Grams and the Last-Good-Reply Policy Applied in General Game Playing” (Mandy.
1 The Theory of NP-Completeness 2 Cook ’ s Theorem (1971) Prof. Cook Toronto U. Receiving Turing Award (1982) Discussing difficult problems: worst case.
Investigating Adaptive Compilation using the MIPSpro Compiler Keith D. Cooper Todd Waterman Department of Computer Science Rice University Houston, TX.
1 Markov Decision Processes Infinite Horizon Problems Alan Fern * * Based in part on slides by Craig Boutilier and Daniel Weld.
Decision Trees Binary output – easily extendible to multiple output classes. Takes a set of attributes for a given situation or object and outputs a yes/no.
NP-complete Problems Prof. Sin-Min Lee Department of Computer Science.
Review of Propositional Logic Syntax
Quantum Computing MAS 725 Hartmut Klauck NTU
Heuristics for Efficient SAT Solving As implemented in GRASP, Chaff and GSAT.
As if computers weren’t fast enough already…
Custom Computing Machines for the Set Covering Problem Paper Written By: Christian Plessl and Marco Platzner Swiss Federal Institute of Technology, 2002.
SAT Solving As implemented in - DPLL solvers: GRASP, Chaff and
Adiabatic Quantum Computing Josh Ball with advisor Professor Harsh Mathur Problems which are classically difficult to solve may be solved much more quickly.
Dana Nau: Lecture slides for Automated Planning Licensed under the Creative Commons Attribution-NonCommercial-ShareAlike License:
The NP class. NP-completeness Lecture2. The NP-class The NP class is a class that contains all the problems that can be decided by a Non-Deterministic.
A greedy algorithm is an algorithm that follows the problem solving heuristic of making the locally optimal choice at each stage with the hope of finding.
On the Relation Between Simulation-based and SAT-based Diagnosis CMPE 58Q Giray Kömürcü Boğaziçi University.
Conceptual Foundations © 2008 Pearson Education Australia Lecture slides for this course are based on teaching materials provided/referred by: (1) Statistics.
1 Intro to AI Local Search. 2 Intro to AI Local search and optimization Local search: –use single current state & move to neighboring states Idea: –start.
Advanced Algorithms Analysis and Design
The NP class. NP-completeness
Inference and search for the propositional satisfiability problem
Comparing Genetic Algorithm and Guided Local Search Methods
CS21 Decidability and Tractability
Example: Applying EC to the TSP Problem
Objective of This Course
A Ridiculously Brief Overview
Resolution Proofs for Combinational Equivalence
CSC 380: Design and Analysis of Algorithms
Presentation transcript:

Quantum Search Heuristics: Tad Hogg’s Perspective George Viamontes February 4, 2002

Outline General Structure General Structure k-SAT Example k-SAT Example Comparisons to Trugenberger Comparisons to Trugenberger Conclusions Conclusions

What are we trying to solve? Quantum Heuristics may be most useful for NP problems Quantum Heuristics may be most useful for NP problems NP problem structure: NP problem structure: Exponential number of candidate solutions as problem size increases Exponential number of candidate solutions as problem size increases Quick test for any given candidate solution to see if it is indeed a correct solution Quick test for any given candidate solution to see if it is indeed a correct solution

Quantum Heuristic vs. NP Quantum algorithms can represent all candidate solutions simultaneously in a superposition Quantum algorithms can represent all candidate solutions simultaneously in a superposition Tests of candidate solutions can be done on all candidates at once with a single operation Tests of candidate solutions can be done on all candidates at once with a single operation Test is often in the form of a cost function Test is often in the form of a cost function

Generic Quantum Heuristic HUHP … Implementation-defined interatcion with Psi …

Generic Quantum Heuristic Hadamards put Psi into a superposition of candidate solutions Hadamards put Psi into a superposition of candidate solutions U modifies the probability amplitudes of Psi to favor better candidate solutions U modifies the probability amplitudes of Psi to favor better candidate solutions P does phase adjustments on Psi P does phase adjustments on Psi C is a control or work qubit C is a control or work qubit Quantum Heuristics vary a lot Quantum Heuristics vary a lot P is optional P is optional C can have different roles C can have different roles

High-Level Breakdown Put data qubits (Psi) into a superposition of all possible solutions Put data qubits (Psi) into a superposition of all possible solutions Do stuff to the probability amplitudes in order to increase the chance of measuring a good solution and decrease the chance of measuring a bad one Do stuff to the probability amplitudes in order to increase the chance of measuring a good solution and decrease the chance of measuring a bad one “Un-superposition” the data qubits “Un-superposition” the data qubits Do optional other stuff to the data (like changing phases) Do optional other stuff to the data (like changing phases) Use extra control/work qubit(s) as necessary Use extra control/work qubit(s) as necessary

The Goal Iterate the previous circuit until there is a good probability of measuring good candidate solutions Iterate the previous circuit until there is a good probability of measuring good candidate solutions Hopefully the number of iterations will be kept to a minimum Hopefully the number of iterations will be kept to a minimum This is the arena of competition with classical heuristics This is the arena of competition with classical heuristics

Generic Quantum Heuristic HUHP … Implementation-defined interatcion with Psi … G G B

Outline General Structure General Structure k-SAT Example k-SAT Example Comparisons to Trugenberger Comparisons to Trugenberger Conclusions Conclusions

What is k-SAT? k-SAT is the problem of finding a satisfying truth assignment for a boolean function in CNF (i.e. an assignment that causes the whole function to be a 1) k-SAT is the problem of finding a satisfying truth assignment for a boolean function in CNF (i.e. an assignment that causes the whole function to be a 1) The “k” represents the number of variables per clause The “k” represents the number of variables per clause E.G. A 3-SAT instance: E.G. A 3-SAT instance:

One Way to Solve k-SAT The GSAT (“Greedy SAT”) algorithm: The GSAT (“Greedy SAT”) algorithm: First produce a random set of variable assignments (select a random set of variables and negate each one with probability ½) First produce a random set of variable assignments (select a random set of variables and negate each one with probability ½) Flip (negate) variables whose new value will result in the satisfying of more clauses Flip (negate) variables whose new value will result in the satisfying of more clauses The flipping is essentially a cost function in which unsatisfied clauses result in a higher cost The flipping is essentially a cost function in which unsatisfied clauses result in a higher cost GSAT runs until an overall minimum cost is reached or it has run for a prespecified number of steps GSAT runs until an overall minimum cost is reached or it has run for a prespecified number of steps

Not the Best Solution It turns out that GSAT isn’t the best heuristic for solving k-SAT It turns out that GSAT isn’t the best heuristic for solving k-SAT Walk-SAT on average performs better Walk-SAT on average performs better Difference is that Walk-SAT doesn’t always rely on the cost function Difference is that Walk-SAT doesn’t always rely on the cost function It will randomly choose between minimizing cost and flipping a random variable in an unsatisfied clause It will randomly choose between minimizing cost and flipping a random variable in an unsatisfied clause

However… Hogg introduces a quantum heuristic for solving k-SAT and chooses to compare it with GSAT rather than Walk-SAT Hogg introduces a quantum heuristic for solving k-SAT and chooses to compare it with GSAT rather than Walk-SAT Though not very useful, it makes sense to compare with GSAT since quantum heuristics, like GSAT, generally rely exclusively on a cost function Though not very useful, it makes sense to compare with GSAT since quantum heuristics, like GSAT, generally rely exclusively on a cost function

Limitations of Hogg’s Decision Overlooks an unexplored avenue of research which involves introducing random walks into quantum heuristics Overlooks an unexplored avenue of research which involves introducing random walks into quantum heuristics Hogg’s heuristic on average has about the same performance as GSAT Hogg’s heuristic on average has about the same performance as GSAT Evidence that quantum heuristics may not be better than classical heuristics since Walk- SAT is better than GSAT Evidence that quantum heuristics may not be better than classical heuristics since Walk- SAT is better than GSAT

One Possible Benefit Portfolios involve running different heuristics concurrently on the same problem instances Portfolios involve running different heuristics concurrently on the same problem instances Halt when one of the heuristics has a solution Halt when one of the heuristics has a solution The problem instances that GSAT performs well on are different than the instances Hogg’s quantum heuristic performs well on The problem instances that GSAT performs well on are different than the instances Hogg’s quantum heuristic performs well on Perhaps quantum heuristics could be used to create more powerful heuristic portfolios Perhaps quantum heuristics could be used to create more powerful heuristic portfolios

Mathematical View Hogg’s implementation of the U operator: Hogg’s implementation of the U operator: Diagonal matrix with as the elements Diagonal matrix with as the elements s is the number of 1-bits in the overall superposition (tau is explained in the next slide) s is the number of 1-bits in the overall superposition (tau is explained in the next slide) And the P operator And the P operator Diagonal matrix with as the elements Diagonal matrix with as the elements c(s) is the number of unsatisfied clauses introduced by a particular solution in the superposition s (rho is explained in the next slide) c(s) is the number of unsatisfied clauses introduced by a particular solution in the superposition s (rho is explained in the next slide)

Mathematical View

Other Details Hogg’s heuristic uses only a single work qubit in addition to the data qubits (Psi) Hogg’s heuristic uses only a single work qubit in addition to the data qubits (Psi) As the h term indicates, the heuristic is applied iteratively As the h term indicates, the heuristic is applied iteratively

More Limitations Phase Parameters seem to be determined experimentally (Hogg does not indicate where he gets particular values from) Phase Parameters seem to be determined experimentally (Hogg does not indicate where he gets particular values from) Since an iteration counter is used directly, the quantum circuit requires a counter of some sort (Hogg does not mention this at all) Since an iteration counter is used directly, the quantum circuit requires a counter of some sort (Hogg does not mention this at all)

Recap of Hogg’s Heuristic On average, performs as well as GSAT but has different behavior for different problem instances On average, performs as well as GSAT but has different behavior for different problem instances Not as good as the best classical heuristic Not as good as the best classical heuristic Has certain non-trivial implementation details that aren’t discussed Has certain non-trivial implementation details that aren’t discussed

Outline General Structure General Structure k-SAT Example k-SAT Example Comparisons to Trugenberger Comparisons to Trugenberger Conclusions Conclusions

Recall… Carlo Trugenberger has also presented a quantum heuristic Carlo Trugenberger has also presented a quantum heuristic Bears some similarities to Hogg’s heuristic but also has fundamental differences Bears some similarities to Hogg’s heuristic but also has fundamental differences

Similarities Trugenberger uses a U operator that is also a diagonal matrix with terms Trugenberger uses a U operator that is also a diagonal matrix with terms Seems to indicate that such terms would be prevalent in any quantum heuristic due to their property of using phase to cancel out bad solutions Seems to indicate that such terms would be prevalent in any quantum heuristic due to their property of using phase to cancel out bad solutions

Similarities Trugenberger’s heuristic also follows the Hadamard – U – Hadamard pattern Trugenberger’s heuristic also follows the Hadamard – U – Hadamard pattern A cost function is also used A cost function is also used

Differences Trugenberger’s heuristic is far more general and robust (possible advantage) Trugenberger’s heuristic is far more general and robust (possible advantage) The cost function is user-defined The cost function is user-defined Multiple control qubits are used rather than the single work qubit used by Hogg (possible drawback) Multiple control qubits are used rather than the single work qubit used by Hogg (possible drawback) No dependence on iterations is explicitly defined (possible advantage) No dependence on iterations is explicitly defined (possible advantage)

Differences Trugenberger does not utilize the extra P operator to modify phases Trugenberger does not utilize the extra P operator to modify phases Instead, Trugenberger’s U gate is enhanced to take care of the cost function and phase modification in a single operator Instead, Trugenberger’s U gate is enhanced to take care of the cost function and phase modification in a single operator He does this by expanding the U gate to also include U inverse He does this by expanding the U gate to also include U inverse By controlling this beefed up U gate with a control bit, the phase modifications can be combined with cost By controlling this beefed up U gate with a control bit, the phase modifications can be combined with cost The U inverse functionality helps to cancel out bad solutions and beef up good solutions The U inverse functionality helps to cancel out bad solutions and beef up good solutions

The Winner? Hard to say without simulation Hard to say without simulation Probably boils down to three factors: Probably boils down to three factors: Will quantum counting be worse than using multiple control qubits? Will quantum counting be worse than using multiple control qubits? Is it harder to implement the beefed up U gate or the “simpler” U gate/P gate combination Is it harder to implement the beefed up U gate or the “simpler” U gate/P gate combination Will Hogg’s heuristic suffer significantly from the delay of transforming any NP problem to SAT (Trugenberger is not bound to SAT) Will Hogg’s heuristic suffer significantly from the delay of transforming any NP problem to SAT (Trugenberger is not bound to SAT)

Outline General Structure General Structure k-SAT Example k-SAT Example Comparisons to Trugenberger Comparisons to Trugenberger Conclusions Conclusions

Hope for Quantum Heuristics? Hogg’s heuristic doesn’t show a benefit in doing things “quantumly” rather than classically Hogg’s heuristic doesn’t show a benefit in doing things “quantumly” rather than classically However, from the theory of portfolios, we can already see that there is some benefit to combining the quantum and the classical However, from the theory of portfolios, we can already see that there is some benefit to combining the quantum and the classical Perhaps a good cost function definition in Trugenberger’s heuristic would save the day Perhaps a good cost function definition in Trugenberger’s heuristic would save the day

Smoke and Mirrors There seems to be a communication gap between quantum heuristic researchers There seems to be a communication gap between quantum heuristic researchers Despite the striking similarities, Hogg does not cite Trugenberger and Trugenberger only cites one of Hogg’s earlier works Despite the striking similarities, Hogg does not cite Trugenberger and Trugenberger only cites one of Hogg’s earlier works Hogg’s experimental results are not encouraging, and Trugenberger presents no experimental results Hogg’s experimental results are not encouraging, and Trugenberger presents no experimental results

Future Avenues On the bright side, since quantum heuristics have not been widely explored or applied, there is still hope On the bright side, since quantum heuristics have not been widely explored or applied, there is still hope Introduction of randomness into quantum heuristics may allow them to surpass classical heuristics which exploit randomness Introduction of randomness into quantum heuristics may allow them to surpass classical heuristics which exploit randomness Problems whose cost functions are more expensive to compute would give quantum heuristics the edge Problems whose cost functions are more expensive to compute would give quantum heuristics the edge Exploration of quantum-classical portfolios Exploration of quantum-classical portfolios Perhaps restructuring of the major gates would lead to further improvement Perhaps restructuring of the major gates would lead to further improvement