| Physical Optimization applies a set of Optimization (minimization) methods motivated by physical processes to general optimization problems |
| These include simulated annealing, neural networks, deterministic annealing, simulated tempering and genetic algorithms |
| We look at general TSP, clustering in physical spaces, track finding, navigation, school class scheduling, Random field Ising Models and data decomposition and other computing optimization problems |
| We discuss when methods such as neural networks are effective |
001 Physical Optimization and Computation
002 Abstract of Physical Computation/Optimization Presentation
003 Physical Optimization and Computation Approaches and their Field
of Origin
004 Optimization as used by Mother Nature and Physics
005 Some Overall Questions Relevant In Classisfying Optimization
Problems and Methods
006 Two Types of Global Mininum and their relation to Local Minima
007 Characteristics of Some Basic Optimization Methods
008 Basic Philosophy of Physical Computation
009 Typical Formalism for Physical Optimization
010 Global and Local Minima in Temperature Dependent Free Energy
011 Comparison of Physical Optimization Methods
012 Sample Problem Illustrating Deterministic Annealing (Gurewitz and
Rose)
013 A deterministic annealing approach to clustering (Gurewitz and
Rose)
014 Details of Clustering Algorithm
015 Comparison of Isodata and Deterministic Annealing
016 Temperature Dependence of Deterministic Annealing
017 Temperature Lowered "below" cluster size
018 Phase Transitions in Physical Optimization Approach
019 TSP or Travelling Salesperson Problem
Classic NP-complete discrete optimization problem
020 Neural Net Compared to Elastic Net
021 Generalized Elastic Network
(Simic's derivation of Durbin and Willshaw's Elastic Net for TSP)
022 Terms in Neural and Elastic Net Energy Functions
023 General Structure of Physical Optimization
024 Comparison of Strategy in Elastic and Strategy
025 Physical Model Underlying Elastic Net
026 Typical TSP Solution with Elastic Net
027 Deterministic Annealing versus Multistate Neurons
028 Elastic Net for Navigation
029 Physical Optimization Formulation of Navigation Problems
030 Results of a Simple Two Vehicle Navigation Problem
031 Results of a Simple Four Vehicle Navigation Problem
032 Deterministic Annealing for Navigation
033 General Comments on Physical Optimization for Navigation
034 Physical Optimization in Computational Chemistry
035 Some Applications of Deterministic Annealing
036 Simulated Tempering -- a New Approach to Monte Carlo
Optimization/Simulated Annealing
037 The Conventional Simulated Annealing and its Problems for Random
Field Ising Models
038 Key Idea in The Tempering Approach
039 RFIM with Simulated Tempering
040 RFIM with Simulated Tempering
041 Some Scheduling Problems in NASA
042 Physical Computation Formulation of University Class Scheduling
Problems
043 Hard Constraints in University Class Scheduling
044 Soft(er) Constraints
045 Soft(er) Constraints -- Continued
046 Approaches to Complexity
047 Computing as a set of Maps
048 Computing is "just" an optimization problem but what
should we optimize?
049 General Issues for Physical Optimization in Computing
050 Physical Optimization in the Execution of Programs
051 Use of Physical Optimization in High Performance Fortran
052 Typical Example of Data Mapping Problem
053 Next slide is also page 26 of aus talk a
Features of Data to Processor Space Mapping:
054 Data Allocation Approaches
055 Computing as a Physics Problem
056 Mapping Problem: Criteria
057 Decomposition of an Arch onto 16 Processors in a Hypercube
058 Comparison of Parallel Data Decomposition Algorithms
059 Comparison of Parallel Data Decomposition Algorithms
060 MultiScale Methods in Parallel Data Decomposition
061 Mapping Times for Multiscale Algorithms
062 One can get Different Answers from Heuristics depending on Initial
Labelling
063 Note: Lesson from 1990 CRPC workshop on TSP at Rice
064 An Irregular Decomposition for Fluid Flow
065 Comparison of Neural Networks for TSP and Data Decomposition
066 NP Completeness and Neural Networks In Summary
067 Optimization in Program Preparation / Code Generation
068 Track Finding Posed as a Problem
069 Track Finding when there are a lot of tracks
070 Neural Networks for Track Finding
071 Track Finding in Intermediate Cases
072 Original Data Set Used by Gurewitz and Rose
073 Results of Deterministic Annealing applied to Dirty Dataset
074 Conclusions on Physical Optimization for Track Finding
075 Conclusions in Physical Optimization
076 Goodbye! Many Choices - Which is best When?