Architecture and Equilibria 结构和平衡 学生:郑巍 导师:刘三阳

Slides:



Advertisements
Similar presentations
Random Processes Introduction (2)
Advertisements

Multi-Layer Perceptron (MLP)
Chapter3 Pattern Association & Associative Memory
Introduction to Neural Networks Computing
2806 Neural Computation Self-Organizing Maps Lecture Ari Visa.
Performance Optimization
1 L-BFGS and Delayed Dynamical Systems Approach for Unconstrained Optimization Xiaohui XIE Supervisor: Dr. Hon Wah TAM.
Pattern Classification All materials in these slides were taken from Pattern Classification (2nd ed) by R. O. Duda, P. E. Hart and D. G. Stork, John Wiley.
1 Chapter 11 Neural Networks. 2 Chapter 11 Contents (1) l Biological Neurons l Artificial Neurons l Perceptrons l Multilayer Neural Networks l Backpropagation.
Un Supervised Learning & Self Organizing Maps Learning From Examples
Some Fundamentals of Stability Theory
SOMTIME: AN ARTIFICIAL NEURAL NETWORK FOR TOPOLOGICAL AND TEMPORAL CORRELATION FOR SPATIOTEMPORAL PATTERN LEARNING.
Dr. Hala Moushir Ebied Faculty of Computers & Information Sciences
Chapter 6 Associative Models. Introduction Associating patterns which are –similar, –contrary, –in close proximity (spatial), –in close succession (temporal)
Artificial Neural Networks
Presentation on Neural Networks.. Basics Of Neural Networks Neural networks refers to a connectionist model that simulates the biophysical information.
CHAPTER 4 S TOCHASTIC A PPROXIMATION FOR R OOT F INDING IN N ONLINEAR M ODELS Organization of chapter in ISSO –Introduction and potpourri of examples Sample.
Neural NetworksNN 11 Neural netwoks thanks to: Basics of neural network theory and practice for supervised and unsupervised.
Artificial Neural Network Supervised Learning دكترمحسن كاهاني
1 Chapter 11 Neural Networks. 2 Chapter 11 Contents (1) l Biological Neurons l Artificial Neurons l Perceptrons l Multilayer Neural Networks l Backpropagation.
To clarify the statements, we present the following simple, closed-loop system where x(t) is a tracking error signal, is an unknown nonlinear function,
Lecture #11 Stability of switched system: Arbitrary switching João P. Hespanha University of California at Santa Barbara Hybrid Control and Switched Systems.
Ming-Feng Yeh1 CHAPTER 16 AdaptiveResonanceTheory.
Artificial Intelligence Chapter 3 Neural Networks Artificial Intelligence Chapter 3 Neural Networks Biointelligence Lab School of Computer Sci. & Eng.
NEURONAL DYNAMICS 2: ACTIVATION MODELS
ECE-7000: Nonlinear Dynamical Systems Overfitting and model costs Overfitting  The more free parameters a model has, the better it can be adapted.
1 Adaptive Resonance Theory. 2 INTRODUCTION Adaptive resonance theory (ART) was developed by Carpenter and Grossberg[1987a] ART refers to the class of.
Architecture and Equilibra 结构和平衡 Chapter Chapter 6 Architecture and Equilibria Perface lyaoynov stable theorem.
AUTOMATIC CONTROL THEORY II Slovak University of Technology Faculty of Material Science and Technology in Trnava.
Synaptic Dynamics: Unsupervised Learning
1 Lecture 6 Neural Network Training. 2 Neural Network Training Network training is basic to establishing the functional relationship between the inputs.
Chapter 2 Single Layer Feedforward Networks
CHAPTER 10 Widrow-Hoff Learning Ming-Feng Yeh.
Feedback Stabilization of Nonlinear Singularly Perturbed Systems MENG Bo JING Yuanwei SHEN Chao College of Information Science and Engineering, Northeastern.
Neural Networks Presented by M. Abbasi Course lecturer: Dr.Tohidkhah.
Asymptotic behaviour of blinking (stochastically switched) dynamical systems Vladimir Belykh Mathematics Department Volga State Academy Nizhny Novgorod.
Architecture and Equilibria 结构和平衡 Chapter 6 神经网络与模糊系统 学生: 李 琦 导师:高新波.
Additional NN Models Reinforcement learning (RL) Basic ideas: –Supervised learning: (delta rule, BP) Samples (x, f(x)) to learn function f(.) precise error.
Neural Networks 2nd Edition Simon Haykin
Giansalvo EXIN Cirrincione unit #4 Single-layer networks They directly compute linear discriminant functions using the TS without need of determining.
NEURAL NETWORK THEORY NEURONAL DYNAMICS Ⅰ : ACTIVATIONS AND SIGNALS
Computational Intelligence Winter Term 2015/16 Prof. Dr. Günter Rudolph Lehrstuhl für Algorithm Engineering (LS 11) Fakultät für Informatik TU Dortmund.
J. Kubalík, Gerstner Laboratory for Intelligent Decision Making and Control Artificial Neural Networks II - Outline Cascade Nets and Cascade-Correlation.
CSC321: Neural Networks Lecture 9: Speeding up the Learning
Neural networks and support vector machines
Neural Networks Winter-Spring 2014
Chapter 2 Single Layer Feedforward Networks
第 3 章 神经网络.
Real Neurons Cell structures Cell body Dendrites Axon
Outline Introduction Signal, random variable, random process and spectra Analog modulation Analog to digital conversion Digital transmission through baseband.
NEURONAL DYNAMICS 2: ACTIVATION MODELS
NEURONAL DYNAMICS 2: ACTIVATION MODELS
§7-4 Lyapunov Direct Method
Widrow-Hoff Learning (LMS Algorithm).
Synaptic Dynamics: Unsupervised Learning
ECE 471/571 - Lecture 17 Back Propagation.
Synaptic DynamicsII : Supervised Learning
Neuro-Computing Lecture 4 Radial Basis Function Network
Artificial Intelligence Chapter 3 Neural Networks
STOCHASTIC HYDROLOGY Random Processes
Artificial Intelligence Chapter 3 Neural Networks
Stability Analysis of Linear Systems
Biointelligence Laboratory, Seoul National University
Artificial Intelligence Chapter 3 Neural Networks
NEURAL DYNAMIC1: ACTIVATIONHS AND SIGNALS
Adaptive Resonance Theory
Artificial Intelligence Chapter 3 Neural Networks
16. Mean Square Estimation
Chapter 7 Inverse Dynamics Control
Artificial Intelligence Chapter 3 Neural Networks
Presentation transcript:

Architecture and Equilibria 结构和平衡 学生:郑巍 导师:刘三阳 2002.12.4 Chapter 6 Architecture and Equilibria 结构和平衡 学生:郑巍 导师:刘三阳

2002.12.4 Chapter 6 Architecture and Equilibria 6.1 Neutral Network As Stochastic Gradient system Classify Neutral network model By their synaptic connection topologies and by how learning modifies their connection topologies synaptic connection topologies how learning modifies their connection topologies 2006.11.10

2002.12.4 Chapter 6 Architecture and Equilibria 6.1 Neutral Network As Stochastic Gradient system Attention :the taxonomy boundaries are fuzzy because the defining terms are fuzzy. 2006.11.10

Three stochastic gradient systems represent the three main categories Chapter 6 Architecture and Equilibria 6.1 Neutral Network As Stochastic Gradient system Three stochastic gradient systems represent the three main categories Backpropagation (BP) Adaptive vector quantization (AVQ) Random adaptive bidirectional associative memory (RABAM) 2006.11.10

2002.12.4 Chapter 6 Architecture and Equilibria 6.2 Global Equilibria: convergence and stability Neural network :synapses , neurons three dynamical systems synapses dynamical systems neurons dynamical systems joint synapses-neurons dynamical systems Historically,Neural engineers study the first or second neural network independently .They usually study learning in feedforward neural networks and neural stability in nonadaptive feedback neural networks. RABAM and ART network depend on joint equilibration of the synaptic and neuronal dynamical systems. 2006.11.10

Convergence undermines stability Chapter 6 Architecture and Equilibria 6.2 Global Equilibria: convergence and stability Equilibrium is steady state (for fixed-point attractors) Convergence is synaptic equilibrium. Stability is neuronal equilibrium. We denote steady state in the neuronal field Another forms with noise Stability - Equilibrium dilemma : Neuron fluctuate faster than synapses fluctuate. Convergence undermines stability 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Competitive learning adaptively quantizes the input pattern space characterizes the continuous distributions of pattern. We shall prove that: Competitive AVQ synaptic vector converge exponentially to pattern-class centroid. They vibrate about the centroid in a Brownian motion 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Competitive AVQ Stochastic Differential Equations The Random Indicator function Supervised learning algorithms depend explicitly on the indicator functions.Unsupervised learning algorithms don’t require this pattern-class information. Centriod 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms The Stochastic unsupervised competitive learning law: We assume The equilibrium and convergence depend on approximation (6-11) ,so 6-10 reduces : 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Competitive AVQ Algorithms 1. Initialize synaptic vectors: 2.For random sample ,find the closet(“winning”)synaptic vector 3.Update the wining synaptic vectors by the UCL ,SCL,or DCL learning algorithm. 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Unsupervised Competitive Learning (UCL) defines a slowly decreasing sequence of learning coefficient Supervised Competitive Learning (SCL) 2006.11.10

2002.12.4 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Differential Competitive Learning (DCL) denotes the time change of the jth neuron’s competitive signal . In practice we only use the sign of (6-20) Stochastic Equilibrium and Convergence Competitive synaptic vector converge to decision-class centroids. May converge to locally maxima. 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms AVQ centroid theorem: if a competitive AVQ system converges,it converge to the centroid of the sampled decision class. Proof. Suppose the jth neuron in Fy wins the activity competition. Suppose the jth synaptic vector codes for decision class Suppose the synaptic vector has reached equilibrium 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Arguments: The spatial and temporal integrals are approximate equal. The AVQ centriod theorem assumes that convergence occurs. The AVQ centroid convergence theorem ensure : exponential convergence 2006.11.10

Chapter 6 Architecture and Equilibria 6.4 AVQ Convergence Theorem Competitive synaptic vectors converge exponentially quickly to pattern-class centroids. Proof.Consider the random quadratic form L The pattern vectors x do not change in time. (still valid if the pattern vector x change slowly relative to synaptic changes.) 2006.11.10

Chapter 6 Architecture and Equilibria 6.4 AVQ Convergence Theorem The average E[L] as Lyapunov function for the stochastic competitive dynamical system. Assume: Noise process is zero-mean and independence of the noise process with “signal”process 2006.11.10

Chapter 6 Architecture and Equilibria 6.4 AVQ Convergence Theorem So ,on average by the learning law 6-12, iff any synaptic vector move along its trajectory. So, the competitive AVQ system is asymptotically stable and in general converges exponentially quickly to a locally equilibrium. Suppose If Then every synaptic vector has Reached equilibrium and is constant . 2006.11.10

Chapter 6 Architecture and Equilibria 6.4 AVQ Convergence Theorem Since p(x) is a nonnegative weight function. The weighted integral of the learning difference must equal zero : So equilibrium synaptic vector equal centroids. Q.E.D 2006.11.10

Chapter 6 Architecture and Equilibra 6.4 AVQ Convergence Theorem Argument Total mean-squared error of vector quantization for the partition So the AVQ convergence theorem implies that the class centroid, and asymptotically ,competitive synaptic vector-total mean-squared error. By The Synaptic vectors perform stochastic gradient descent on the mean-squared-error surface in pattern-plus-error In the sense :competitive learning reduces to stochastic gradient descent 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.5 Global stability of feedback neural networks Global stability is jointly neuronal-synaptics steady state. Global stability theorems are powerful but limited. Their power: their dimension independence nonlinear generality their exponentially fast convergence to fixed points. Their limitation: do not tell us where the equilibria occur in the state space. 2006.11.10

Chapter 6 Architecture and Equilibra 6 Chapter 6 Architecture and Equilibra 6.5 Global stability of feedback neural networks Stability-Convergence Dilemma Stability-Convergence Dilemma arise from the asymmetry in neuronal and synaptic fluctuation rates. Neurons change faster than synapses change. Neurons fluctuate at the millisecond level. Synapses fluctuate at the second or even minute level. The fast-changing neurons must balance the slow-changing synapses. 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.5 Global stability of feedback neural networks Stability-Convergence Dilemma 1.Asymmetry:Neurons in and fluctuate faster than the synapses in M. 2.stability: (pattern formation). 3.Learning: 4.Undoing: the ABAM theorem offers a general solution to stability-convergence dilemma. 2006.11.10

Chapter 6 Architecture and Equilibria 6.6 The ABAM Theorem The ABAM Theorem(证明的关键是找到一个合适的Lyapunov函数) The Hebbian ABAM and competitive ABAM models are globally stable. Hebbian ABAM model: Competitive ABAM model , replacing 6-35 with 6-36 2006.11.10

Chapter 6 Architecture and Equilibria 6.6 The ABAM Theorem If the positivity assumptions Then, the models are asymptotically stable, and the squared activation and synaptic velocities decrease exponentially quickly to their equilibrium values: Proof. the proof uses the bounded lyapunov function L 2006.11.10

Chapter 6 Architecture and Equilibria 6.6 The ABAM Theorem Make the difference to 6-37: 2006.11.10

Chapter 6 Architecture and Equilibria 6.6 The ABAM Theorem To prove global stability for the competitive learning law 6-36 We prove the stronger asymptotic stable of the ABAM models with the positivity assumptions. 2006.11.10

Chapter 6 Architecture and Equilibria 6.6 The ABAM Theorem Along trajectories for any nonzero change in any neuronal activation or any synapse. Trajectories end in equilibrium points. Indeed 6-43 implies: The squared velocities decease exponentially quickly because of the strict negativity of (6-43) and ,to rule out pathologies . Q.E.D because of the second-order assumption of nondegenerate Hessian matrix. 2006.11.10

2002.12.4 Chapter 6 Architecture and Equilibria 6.7 structural stability of unsupervised learning and RABAM Is unsupervised learning structural stability? Structural stability is insensitivity to small perturbations Structural stability ignores many small perturbations. Such perturbations preserve qualitative properties. Basins of attractions maintain their basic shape. 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.7 Structural stability of unsupervised learning and RABAM Random Adaptive Bidirectional Associative Memories RABAM Brownian diffusions perturb RABAM model. (也就是加进一种噪声) The differential equations in 6-33 through 6-35 now become stochastic differential equations, with random processes as solutions. The diffusion signal hebbian law RABAM model: 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.7 Structural stability of unsupervised learning and RABAM With the stochastic competitive law: If is sufficiently steep 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.7 Structural stability of unsupervised learning and RABAM With noise (independent zero-mean Gaussian white-noise process). the signal hebbian noise RABAM model: 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.7 Structural stability of unsupervised learning and RABAM RABAM Theorem. The RABAM model (6-46)-(6-48) or (6-50)-(6-54), is global stable.if signal functions are strictly increasing and amplification functions and are strictly positive, the RABAM model is asymptotically stable. Proof. The ABAM lyapunov function L in (6-37) now defines a random process. At each time t,L(t) is a random variable. The expected ABAM lyapunov function E(L) is a lyapunov function for the RABAM. 2006.11.10

Chapter 6 Architecture and Equilibria 6 Chapter 6 Architecture and Equilibria 6.7 Structural stability of unsupervised learning and RABAM 2006.11.10

2002.12.4 Chapter 6 Architecture and Equilibria 6.7 Structural stability of unsupervised learning and RABAM 【Reference】 [1] “Neural Networks and Fuzzy Systems -Chapter 6” P.221-261 Bart kosko University of Southern California. 2006.11.10