Entropy landscape of solutions in the binary perceptron problem

被引:35
作者
Huang, Haiping [1 ,2 ]
Wong, K. Y. Michael [2 ]
Kabashima, Yoshiyuki [1 ]
机构
[1] Tokyo Inst Technol, Dept Computat Intelligence & Syst Sci, Yokohama, Kanagawa 2268502, Japan
[2] Hong Kong Univ Sci & Technol, Dept Phys, Hong Kong, Hong Kong, Peoples R China
关键词
NEURAL-NETWORKS; ALGORITHM; CAPACITY; DYNAMICS; GRAPHS; SPACE;
D O I
10.1088/1751-8113/46/37/375002
中图分类号
O4 [物理学];
学科分类号
0702 ;
摘要
The statistical picture of the solution space for a binary perceptron is studied. The binary perceptron learns a random classification of input random patterns by a set of binary synaptic weights. The learning of this network is difficult especially when the pattern (constraint) density is close to the capacity, which is supposed to be intimately related to the structure of the solution space. The geometrical organization is elucidated by the entropy landscape from a reference configuration and of solution-pairs separated by a given Hamming distance in the solution space. We evaluate the entropy at the annealed level as well as replica symmetric level and the mean field result is confirmed by the numerical simulations on single instances using the proposed message passing algorithms. From the first landscape (a random configuration as a reference), we see clearly how the solution space shrinks as more constraints are added. From the second landscape of solution-pairs, we deduce the coexistence of clustering and freezing in the solution space.
引用
收藏
页数:18
相关论文
共 44 条
[1]   Efficient supervised learning in networks with binary synapses [J].
Baldassi, Carlo ;
Braunstein, Alfredo ;
Brunel, Nicolas ;
Zecchina, Riccardo .
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2007, 104 (26) :11079-11084
[2]   Random codes: Minimum distances and error exponents [J].
Barg, A ;
Forney, GD .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2002, 48 (09) :2568-2573
[3]   BROKEN SYMMETRIES IN MULTILAYERED PERCEPTRONS [J].
BARKAI, E ;
HANSEL, D ;
SOMPOLINSKY, H .
PHYSICAL REVIEW A, 1992, 45 (06) :4146-4161
[4]   Irreducible free energy expansion and overlaps locking in mean field spin glasses [J].
Barra, Adriano .
JOURNAL OF STATISTICAL PHYSICS, 2006, 123 (03) :601-614
[5]   The Dynamics of Message Passing on Dense Graphs, with Applications to Compressed Sensing [J].
Bayati, Mohsen ;
Montanari, Andrea .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2011, 57 (02) :764-785
[6]   A variational description of the ground state structure in random satisfiability problems [J].
Biroli, G ;
Monasson, R ;
Weigt, M .
EUROPEAN PHYSICAL JOURNAL B, 2000, 14 (03) :551-568
[7]   Learning by message passing in networks of discrete synapses [J].
Braunstein, A ;
Zecchina, R .
PHYSICAL REVIEW LETTERS, 2006, 96 (03)
[8]  
Cover T.M., 2006, ELEMENTS INFORM THEO, V2nd ed
[9]   Entropy landscape and non-Gibbs solutions in constraint satisfaction problems [J].
Dall'Asta, L. ;
Ramezanpour, A. ;
Zecchina, R. .
PHYSICAL REVIEW E, 2008, 77 (03)
[10]   Pairs of SAT-assignments in random boolean formulae [J].
Daude, Herve ;
Mezard, Marc ;
Mora, Thierry ;
Zecchina, Riccardo .
THEORETICAL COMPUTER SCIENCE, 2008, 393 (1-3) :260-279