DocumentCode :
1420063
Title :
An annealing framework with learning memory
Author :
Lo, Chun-Chi ; Hsu, Ching-Chi
Author_Institution :
Dept. of Comput. Sci. & Inf. Eng., Nat. Taiwan Univ., Taipei, Taiwan
Volume :
28
Issue :
5
fYear :
1998
fDate :
9/1/1998 12:00:00 AM
Firstpage :
648
Lastpage :
661
Abstract :
Simulated annealing can be viewed as a process that generates a sequence of Markov chains, i.e., it keeps no memory about the states visited in the past of the process. This property makes simulated annealing time-consuming in exploring needless states and difficult in controlling the temperature and transition number. In this paper, we propose a new annealing model with memory that records important information about the states visited in the past. After mapping applications onto a physical system containing particles with discrete states, the new annealing method systematically explores the configuration space, learns the energy information of it, and converges to a well-optimized state. Such energy information is encoded in a learning scheme. The scheme generates states distributed in Boltzmann-style probability according to the energy information recorded in it. Moreover, with the assistance of the learning scheme, controlling over the annealing process become simple and deterministic. From qualitative and quantitative analyses in this paper, we can see that this convenient framework provides an efficient technique for combinatorial optimization problems and good confidence in the solution quality
Keywords :
Boltzmann equation; Markov processes; learning (artificial intelligence); simulated annealing; Boltzmann-style probability; Markov chain sequence; combinatorial optimization; configuration space; learning memory; qualitative analyses; quantitative analyses; simulated annealing; Convergence; Mathematical model; Microscopy; Probability distribution; Sampling methods; Simulated annealing; Space exploration; Space technology; Temperature control; Temperature distribution;
fLanguage :
English
Journal_Title :
Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on
Publisher :
ieee
ISSN :
1083-4427
Type :
jour
DOI :
10.1109/3468.709611
Filename :
709611
Link To Document :
بازگشت