Modeling and qualitative analysis of continuous-time neural networks under pure structural variations

被引：2

作者：

Grujic, LT ^{[1
]}

Michel, AN ^{[1
]}

机构：

[1] UNIV NOTRE DAME, DEPT ELECT ENGN, NOTRE DAME, IN 46556 USA

来源：

MATHEMATICS AND COMPUTERS IN SIMULATION | 1996年 / 40卷 / 5-6期

基金：

美国国家科学基金会;

关键词：

Hopfield neural networks [1-4]w ere mainly modeled by assuming a time-invariant structure and used to study their qualitative dynamical properties [5; 6]. During the learning process a neural nework is subjected to (arbitrary or designed) structural variation [4; 7]. Dynamics of neural networks under designed structural variations was investigated in [8; a nd of those under unpredictable structural and/or parameter variations in [10].H owever; modeling of Hopfield neural networks under pure structural variations has not been carded out; which is done in what follows. Stability analysis of neural networks initiated by Hopfield [1 ] for a special class of neural networks was developed for any type of Hopfield neural network and extended to the qualitative analysis of their motions in forced regimes in [5; 11 ]. It was emphasized in [1; 9] that in neurobiology the structure of the circuit changes every time learning occurs and that a memory stored in an artificial neural network is achieved by an appropriate choice of conductances. Stability analysis of neural networks under structural variations was considered in [5; 7; and under both parameter and structural variations in [8]. In what follows; new results are obtained for exponential stability of x -----0 of Hopfield neural networks and for estimates of the domain of the exponential stability; which is also defined herein; under arbitrary t Supported in part by NSF Grant ECS 91-07728. * Corresponding author;

D O I：

10.1016/0378-4754(95)00004-6

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

A qualitative analysis is developed for continuous-time neural networks subjected to random pure structural variations. Simple algebraic conditions are established for both structural exponential stability of x = 0 of the neural network and for estimates of its domain of attraction. Bounds on motions of the neural network in a forced regime are provided. They do not require any information about its actual structure, which can be completely unknown and may vary unpredictably.

引用

页码：523 / 533

页数：11

共 50 条

[21] Closed-form continuous-time neural networks
Ramin Hasani
Mathias Lechner
Alexander Amini
Lucas Liebenwein
Aaron Ray
Max Tschaikowski
Gerald Teschl
Daniela Rus
Nature Machine Intelligence, 2022, 4 : 992 - 1003
[22] A learning result for continuous-time recurrent neural networks
Sontag, Eduardo D.
Systems and Control Letters, 1998, 34 (03): : 151 - 158
[23] CONTINUOUS-TIME DYNAMICS OF ASYMMETRICALLY DILUTED NEURAL NETWORKS
KREE, R
ZIPPELIUS, A
PHYSICAL REVIEW A, 1987, 36 (09): : 4421 - 4427
[24] DISCRETE-TIME VERSUS CONTINUOUS-TIME MODELS OF NEURAL NETWORKS
WANG, X
BLUM, EK
JOURNAL OF COMPUTER AND SYSTEM SCIENCES, 1992, 45 (01) : 1 - 19
[25] Analysis of Cluster Consensus in Continuous-Time Networks
Develer, Umit
Akar, Mehmet
2018 ANNUAL AMERICAN CONTROL CONFERENCE (ACC), 2018, : 448 - 453
[26] Global dissipativity of continuous-time recurrent neural networks with time delay
Liao, XX
Wang, J
PHYSICAL REVIEW E, 2003, 68 (01):
[27] Transient regime duration in continuous-time neural networks with delay
Pakdaman, K
Grotta-Ragazzo, C
Malta, CP
PHYSICAL REVIEW E, 1998, 58 (03): : 3623 - 3627
[28] Transient regime duration in continuous-time neural networks with delay
Phys Rev E., 3-B (3623):
[29] State Derivative Normalization for Continuous-Time Deep Neural Networks
Weigand, Jonas
Beintema, Gerben I.
Ulmen, Jonas
Goerges, Daniel
Toth, Roland
Schoukens, Maarten
Ruskowski, Martin
IFAC PAPERSONLINE, 2024, 58 (15): : 253 - 258
[30] Integrated Neural Networks for Nonlinear Continuous-Time System Identification
Mavkov, Bojan
Forgione, Marco
Piga, Dario
IEEE CONTROL SYSTEMS LETTERS, 2020, 4 (04): : 851 - 856

← 1 2 3 4 5 →