Lyapunov Stability Analysis of the Quantization Error for DCS Neural Networks

Size: px
Start display at page:

Download "Lyapunov Stability Analysis of the Quantization Error for DCS Neural Networks"

Transcription

1 Lyapunov Stability Analysis of the Quantization Error for DCS Neural Networks Sampath Yerramalla, Bojan Cukic Lane Department of Computer Science and Electrical Engineering West Virginia University Morgantown WV Edgar Fuller Department of Mathematics West Virginia University Morgantown WV Abstract In this paper we show that the quantization error for Dynamic Cell Structures (DCS) Neural Networks (NN) as defined by Bruske and Sommer provides a measure of the Lyapunov stability of the weight centers of the neural net. We also show, however, that this error is insufficient in itself to verify that DCS neural networks provide stable topological representation of a given fixed input feature manifold. While it is true that DCS generates a topology preserving feature map, it is unclear when and under what circumstances DCS will have achieved an accurate representation. This is especially important in safety critical systems where it is necessary to understand when the topological representation is complete and accurate. The stability analysis here shows that there exists a Lyapunov function for the weight adaptation of the DCS NN system applied to a fixed feature manifold. The Lyapunov function works in parallel during DCS learning, and is able to provide a measure of the effective placement of neural units during the NN s approximation. It does not, however, guarantee the formation of an accurate representation of the feature manifold. Simulation studies from a selected CMU-Benchmark involving the use of the constructed Lyapunov function indicate the existence of a Globally Asymptotically Stable (GAS) state for the placement of neural units, but an example is given where the topology of the constructed network fails to mirror that of the input manifold even though the quantization error continues to decrease monotonically. I. INTRODUCTION The learning process in most common neural nets evolves sequentially over time. This process of learning also being combinatorial and nonlinear in nature, it is a challenging task to evaluate the dynamic properties that govern the performance of a neural net over time. Stability being considered as an important dynamic property, the question then is: Is there an effective tool that can analyze the stability aspects of the dynamic learning process inside the NN? Even after decades of extensive research on neural nets and its wide spread use in many practical applications, the answer to this question still remains uncertain and contradictory. The objective of mathematical theory of nonlinear stability analysis is often misunderstood as the process of finding a solution for the differential equation(s) that govern the dynamics, either analytically or numerically (simulation studies). Theoretically, there is no guarantee for the existence of a solution to a nonlinear differential equation at all times, let alone the complicated task of solving them. In practice however, there exists a variety of techniques and tools that analyze the existence of stable states in nonlinear systems, without actually the need for solving any differential equation [1], [5]. Almost all stability definitions for linear systems lead to a similar stability criterion (R-locus, eigenvalues, etc). Therefore defining a precise stability definition for linear systems can be relaxed. However, eigenvalue methods are not applicable to nonlinear systems and so one needs to be careful while dealing with systems that have non-linear properties. It s often seen that if such systems are stable under one definition of stability they may tend to become unstable under other definitions [5]. This difficulty in imposing strong stability restrictions for nonlinear systems was realized as early as a century ago by a Russian mathematician A.M.Lyapunov. Details on Lyapunov s stability analysis technique for nonlinear discrete systems can be found in [6], [4]. The fact that Lyapunov s direct method or nonlinear theory of stability analysis can be readily applied to validate the existence (or non-existence) of stable states in nonlinear systems, intrigued us in using Lyapunov function in our stability analysis as a means to answer the question posed earlier. Dynamic Cell Structures (DCS) NN represents a family of Topology Representing Networks (TRN), used for both supervised and un-supervised learning. Some applications of DCS include vector quantization, clusetering, dimensional reduction [17], [16]. In this paper we analyze the stability properties of neural unit positioning by DCS NN using Lyapunov theory and we present a qualified answer to the question Is the learning process of DCS NN, stable? Throughout this stability analysis we consider DCS NN as a sequentially evolving discrete logic system, closely related to the family of Logical Discrete Event Systems. For detailed description of a Logical Discrete Event System, refer [2]. We also assume that the feature manifold being mapped by DCS, is fixed or stationary, meaning the manifold doesn t evolve over time. The paper is organized as follows: Section-II, provides a description of the competitive Hebb rule and Kohonen-like rule that govern the learning process of the DCS NN. This description of the building-blocks is later used in Section-III

2 to present a state-space representation of the dynamics of the network. Also in Section-III, the construction process of the Lyapunov function is described in detail along with proving (through proofs) the existence of a Globally Asymptotically Stable learning state in DCS NN. Simulation results are presented in Section-IV using a selected CMU benchmark of artificially generated data. Finally, conclusions are drawn in Section-V based on the proofs and simulation results. II. DCS NN BUILDING BLOCKS In 1994 Jorg Bruske and Gerald Sommers of the University of Kiel, introduced the concept of DCS as a family of Topology Representing, Self-Organizing NN [17], [16]. This Topology Preserving Feature Map (TPFM) generation scheme was motivated by Growing Neural GAS algorithm developed by Fritzke and the former work on TRN by Martinez [13]. DCS uses Kohen-like adaptation to shift the weighted centers of the local neighborhood structure, closer to the feature manifold. This when applied in conjunction with Competitive Hebb Rule (CHR) to update lateral connections of the neighbors, produces a network representation that preserves the features of the input manifold. We believe that these two essential building blocks of DCS algorithm, play a key role in the generation of a TPFM. In an attempt to study these rules in depth, a formal overview is depicted in the following section. A. Competitive Hebb Rule (CHR) DCS NN rests upon a Radial Basis Function (RBF) and an additional layer of lateral connections between the neural units [16]. These lateral connection strengths are symmetric and bounded in nature, C ij = C ji [, 1] I. The goal of CHR is to update the lateral connections by mapping neighborhoods in the input manifold to neighborhoods of the network. Thereby, avoiding any restrictions of the topology of the network [17]. For each input element of the feature manifold, CHR operates by setting the connection strength between the two neural units that are closer to the input than any other neuron pair, to a highest possible connection strength of 1. These two neural units are referred to as the Best Matching Unit (bmu) and Second Best Unit (sbu). CHR then proceeds to decay the strength of all other existing lateral connections emanating from the bmu using a forgetting constant, α. If any of these existing connections drop below a predefined threshold θ, they are set to zero. The set of all neural units that are connected to the bmu is defined as the neighborhood set of the bmu, and represented by nbr. All other connections of the network remain unaltered. In this way CHR induces a Delaunay Triangulation into the network by preserving the neighborhood structure of the feature manifold. 1 (i = bmu) (j = sbu) (i = bmu) (j nbr sbu) (c ij < θ) c ij (t + 1) = (1) αc ij (t) (i = bmu) (j nbr sbu) (c ij θ) c ij (t) i, j bmu It was proved in [13] that algorithms utilizing CHR to update lateral connections between neural units generate a TPFM. B. Kohonen Like Rule Unlike a feedforward Neural Net, the weight center w i associated with a neural unit i of the DCS network represents the location of the neural unit in the output space. It is crucial to realize that these weighted centers be updated in a manner that preserves the geometry of the input manifold. This can be achieved by adjusting the weighted center of the bmu and its surrounding neighborhood structure nbr closer to the input element. Thus, for each element of the feature manifold, DCS adapts the corresponding bmu and its neighborhood set nbr in a Kohonen-like manner, see [17]. Over any training cycle, let w i = w i (t + 1) w i (t) represent the adjustment of the wiehgt center of the neural unit, then the Kohonen-like rule followed in DCS can be represented as follows, ε bmu ( m w i (t)) i = bmu w i = ε nbr ( m w i (t)) i nbr (2) (i bmu) (i nbr) where ε bmu, ε nbr (, 1) are predefined constants known as the learning rates that define the momentum of the update process. For every input element, applying CHR before any other adjustment ensures that sbu is a member of nbr set for all further adjustments within the inner loop of the DCS algorithm. III. STATE-SPACE ANALYSIS The first step of any stability analysis lies in identification of the states involved with the dynamical system. State Space representation is the most common method of representing the identified states of the dynamic system. This representation technique is very effective during the construction of a Lyapunov function for Nonlinear Systems [6]. A. States of DCS Logic The following equations give the representation of the states of DCS logic during CHR and KHR adjustment as discrete differences over a time step. 1) Center State: W 2) Connection State: C = f W ( x W, bmu, nbr, ε bmu, ε nbr ) (3) = f C ( x C, bmu, nbr, α, θ) (4) Here x W and x C are the states of DCS learning that represents the spatial distribution of neural units and the lateral connection structure between the neural units respectively. If M I and A O are the input and output spaces of DCS that contain the feature manifold M and the network representation A, f W : I O and f C : I O are continuous functions that provide the required adjustment to x W and x C respectively.

3 B. DCS Output/Recall When DCS NN is presented with a test input u R I, the network is searched for the appropriate bmu and the closest of all neural units that is connected to the bmu. Note that this neural unit need not always be the sbu. DCS then applies linear interpolation technique between these two neural units to generate the output vector y(u) R O. The output y(u) generated at time t by the DCS network depends not only on the states x C (t), x W (t), but also the location of the bmu and its connected neighborhood structure. Thus, the output of DCS for a give input stimulus u can be represented as, y(u) = G( x C (t), x W (t), bmu, nbr, u) (5) If bmu is not connected to any other neural unit of the network then the generated output y(u) is based solely on the bmu s location. It is obvious that the output produced by the network with no neurons is zero. C. DCS State Space Representation We consider DCS Neural Net s approximation from any given input manifold M R I. In this neural net, a set of neural units i for i = 1... N with weighted centers w i W R O and lateral connections c ij [, 1] C R O represent the map generated by DCS from input manifold M on to the output space O R O. It was proved by Martinetz and Schulten in [13] that this is a topology preserving operation in the sense that neighborhoods of M are preserved in O. Let G(W, C) be the collection of N neural units of DCS having W NxN as their weighted center matrix and C NxN as their connection strength matrix, that represent the DCS map from a given input manifold M onto an output space O. Then, G(W, C) can be considered as an undirected graph of N multivalent vertices. Definition 1 (DCS Mapping): DCS Mapping G(W, C) of a given manifold M I is the N th order neural unit representation of M in the output space I, gotten by assigning N neural in t n steps of the DCS algorithm. Considering DCS as a discrete time dynamical system, the following state-space representation can be formulated : W = f W ( x W, bmu, nbr, ε bmu, ε nbr ) C = f C ( x C, bmu, nbr, α, θ) y = G(W, C) (6) IV. LYAPUNOV S STABILITY THEORY We want to determine if the DCS algorithm will reliably learn a fixed input manifold M on successive applications. The question is then: How much can the evolving DCS approximation of M, denoted by y f, deviate from a previously learned version of M, denoted by y i? Since the DCS approximation process is a time-varying construction whose state changes according to the difference relations in (6), it is natural to use the Lyapunov formulation of stability for a dynamical system to answer this question. The relevant result of Lyapunov [6] is the following Theorem 1 (Lyapunov Stability): Given a system of differential equations d X dt = f(x 1,..., x n, t) (7) if it is possible to find a positive-definite function V of X for which V () = and dv dt then the trivial solution of (7) is Lyapunov stable or locally stable. The Lyapunov condition in the discrete case can be formulated as follows: Theorem 2 (Discrete Lyapunov Stability): The discrete system of difference equations S = f(s 1,..., s n, t) (8) is stable in the sense of Lyapunov if there is a positive-definite function V for which V () = and V for some time step >. The system besides being stable, if V happens V to be negative-definite, <, then the system is said to be Asymptotically Stable. Moreover if V as t, the system is Globally Asymptotic Stable (GAS). A. Lyapunov function Construction To set up a Lyapunov function for DCS, we measure how accurately a current state of the algorithm models the input manifold in terms of the amount of geometry and topology preserved in the output manifold. To measure the accuracy of the DCS approximation, Bruske and Sommer introduce the Quantization Error as defined in [17]. This measures the effectiveness of the placement of neural units by computing the amount by which a neural unit deviates from an element m M of the input manifold for which it is the Best Matching Unit. We then average this over the total number of neurons, N, during that state of DCS approximation to get Q(t) = 1 m w bmu(m,t) (9) N In the following section we show that Q serves as a Lyapunov function for the neural unit placement of DCS. B. Neural Unit Placement by DCS is Lyapunov stable With the above setup, it follows that Theorem 3: The neural units, w i, assigned by DCS approximation during learning of a fixed input manifold M I are globally, asymptotically stable in the sense of Lyapunov. Proof: First of all we need to show that the Lyapunov function (9), constructed for the time-varying discrete system of DCS is valid. It is obvious that Q since m w bmu(m,t). Also Q() =, since there are no neurons in the DCS network during zero state x C = x W =. To show that Q <, first note that since the time step >, the numerator Q will determine the sign in question. Computing

4 using (9) we see that over any time interval = t i+1 t i >, adjustments of the weighted centers w i are determined by the Kohonen Rule. Over any learning cycle, DCS adjusts the bmu and its neighbors for every m M according to the rule in (2). Let m w bmu(m,t) be represented by, then we see that over a single training cycle Q = 1 N + 1 d( m, t + 1) 1 N = N d( m, t + 1) (N + 1) N(N + 1) = N (d( m, t + 1) ) N(N + 1) (1) For any m M, we need to show that either the corresponding portion of the numerator of (1) is negative or that some other portion compensates when it is positive. There are three ways in which a neural unit s weighted center may get adjusted. It may get updated 1) as the bmu for m, 2) as the bmu of some other m M, 3) or as one of the neighbors of the bmu of some other m M. In the first case, to show that (1) < it is sufficient to show that d( m, t + 1) < (11) Computing using (2) gives d( m, t + 1) = m w BMU( m,t+1) = m (w bmu( m,t) + ε bmu ( m w bmu( m,t) )) = ( m w bmu(t) ) ε bmu ( m w bmu(t) ) = m w bmu(t) ε bmu m + ε bmu w bmu(t) ) = (1 ε bmu ) m w bmu(t) = (1 ε bmu ) Since ε BMU (, 1), d( m, t + 1) < which implies (11). In the second case, the best matching unit for m, bmu( m), may get updated as the best matching unit for some other m M. This update will be either preceded or followed by and update of the neuron as bmu for m. If preceded and the neuron is closer to m than m, the resulting decrease in d( m, t) will be larger that the increase in. If preceded and the neuron is farther from m than m, the increase in due to this adjustment will be smaller than the decrease from the neuron s adjustment as bmu. If followed by the first case adjustment, the previous two relationships are reversed but the net effect on Q is still a decrease. All of these follow from the triangle inequality for the euclidean metric. The third case will have Q < since ε bmu >> ε sbu in general. In case the two values are comparable, the result follows from the same argument as in the second case above. As a result, the function Q is a Lyapunov function for the weight system of DCS and furthermore since Q < and Q as t, the stability is global and asymptotic. This result reflects the fact that DCS is topology preserving as a Kohonen based algorithm as shown by Martinetz and Schulten [13]. Unfortunately, this topology preservation does not correlate with the stability of the neural unit placement as is demonstrated in the next section. V. SIMULATION RESULTS The fixed feature manifold M I R 2 is the twospirals as provided by CMU s archive for artificial neural net benchmark data [21], [2]. In order to provide a better understanding to this stability analysis, the parameters (density,radius) of the training data are set in a manner similar to that described by Bruske and Sommer in [17]. Thus, the density of the twin-spiral training data is set to 5 and the spiral radius to 6.5. The program two-spirals.c of [21], generates 962 input elements in the x-y plane. Fig.1(a) is a snapshot of a 51st Fig. 1. DCS mapping and Lyapunov function at 19 and its corresponding quantization error order DCS NN representation, A (51) O, of the fixed twinspiral feature manifold M I. The relatively fewer number of neural units of the network fail to produce an accurate representation of M in A. This is also indicated by a corresponding large value in the quantization error in Fig.1(b). As the neural units are incremented in number to 251, the network gradually evolved into a topology preserved feature map, as in Fig.2(a). Meaning that the neighborhoods of M are preserved in 251st order DCS neural network representation, A (251). The relatively small quantization error in the network is indicated in Fig.2(b). However, if the simulation is continued for another Fig. 2. DCS mapping and Lyapunov function at 19 and its corresponding quantization error

5 iteration as shown in Fig.3(a), a fault or a defective connection is developed in the network. In order for a better understanding of the concept of defective network representation, let us give a precise definition for a fault in DCS learning. Definition 2 (Fault): We say that a fault has occurred in DCS learning, if two neural units, {i, j} of the network, A O, that are mapped to relatively distant input elements {m i, m j } M I, develop a a positive connection strength between them, c ij >. The faults however, are not detected by the quantization error term of Fig.3(b), that continues to decrease monotonically. Fig Faults in the DCS network near 25 epochs VI. CONCLUSION In this paper we proved through Lyapunov stability analysis, the quantization error of the DCS NN mapping of a fixed input manifold of achieving a globally asymptotic stable learning state. The credit for this stabile behavior is given to the design of the DCS approximating algorithm using Kohonenlike adaptation that adjusts the positions of weighted centers in conjunction with a Competitive Hebbian Rule that updates the lateral connection structure to form a topology preserving feature map. It was proved through simulation study that the formation of an accurate representation of M I in A O, depends greatly on a precise moment where the the learning process needs to be stopped. The simulation study also proved that the quantization error, by itself does not indicate the formation of an accurate topology preserved feature map, and thus cannot be used directly as a criteria for stopping the learning. Finally we propose the need for an efficient measure of the amount of topology preservation in the network in terms of a topology error, operating in conjunction with the quantization error to provide a performance measure to the accuracy of DCS NN s approximation. [7] Bernd Fritzke. Growing Cell Structures - A Self-organizing Network for Unsupervised and Supervised Learning, ICSI, Vol. 7, No. 9, Pages , May [8] Bernd Fritzke. Growing Self-organizing Networks - why?, European Symposium on Artificial Neural Network, Pages 61-72, [9] Bernd Fritzke. A Growing Neural Gas Network Learns Topologies, Advances in Neural Information Processing Systems, Vol. 7, pages , MIT Press, [1] Bernd Fritzke. Unsupervised Clustering With Growing Cell Structurea, Proc. of the IJCNN, Pages , [11] Bernd Fritzke. Growing Grid - A Self-Organizing Network With Constant Neighborhood Range and Adaptation Strength, Neural Processing Letters, Vol. 2, No. 5, Pages 9-13, [12] Teuvo Kohonen. The Self-Organizing Map, Proc. of the IEEE, Vol. 78, No. 9, Pages , September, 199. [12] Jurgen Rahmel. On The Role of Topology for Neural Network Interpretation, Proc. of European Conference on Artificial Intelligence, [13] Thomas Martinez, and Klaus Schulten. Topology Representing Networks, Neural Networks, Vol. 7, No. 3, Pages , [14] Lucius Chudy, and Igor Farkas. Prediction of Chaotic Time-Series Using Dynamic Cell Structures and Local Linear Models, Institute of Measurement Science, Slovak Academy of Sciences, Slovakia. [15] Lucius Chudy, and Igor Farkas. Modified Dynamic Cell Structures as a Thinning Algorithm, Institute of Measurement Science, Slovak Academy of Sciences, Slovakia. [16] Ingo Ahrns, Jorg Bruske, and Gerald Sommer. On-line Learning with Dynamic Cell Structures, Proc. of ICANN, Vol. 2, Pages , [17] Jorg Bruske, and Gerald Sommer. Dynamic Cell Structures, NIPS, Vol. 7, Pages , [18] Jorg Bruske, and Gerald Sommer. Dynamic Cell Structure Learns Perfectly Topology Preserving Map, Neural Computation, Vol. 7, No. 4, , [19] Charles C. Jorgensen. Feedback Linearized Aircraft Control Using Dynamic Cell Structures, NASA Ames Research Center, ISSCI [2] Fahlman S.E. CMU Benchmark Collection for Neural Net Learning Algorithms, Carnegie Mellon Univ., School of Computer Science, machinereadable data repository, Pittsburgh, [21] CMU Benchmark Archive. bench/cmu/, 22. REFERENCES [1] N.Rouche, P.Habets, and M.Laloy. Stability Theory by Liapunov s Direct Method, Springer-Verlag, New York Inc., [2] Kevin M.Passino, N.Michel, and Panos J. Antsaklis. Lyapunov Stability of a Class of Discrete Event Systems, IEEE Tran. on Automatic Control, Vol. 39, No. 2, February [3] Xinghuo Yu, M. Onder Efe,and Okyay Kaynak. A Backpropagation Learning Framework For Feedforward Neural Networks, IEEE transactions, No /1, 21. [4] W. L. Brogan. Modern Control Theory, II Edition, Prentice Hall Inc., [5] Bernard Friedland. Advanced Control System, Prentice Hall Inc., [6] V. I. Zubov. Methods of A. M. Lyapunov and Their Applications, U.S. Atomic Energy Commission, 1957.

Lyapunov Analysis of Neural Network Stability in an Adaptive Flight Control System

Lyapunov Analysis of Neural Network Stability in an Adaptive Flight Control System Lyapunov Analysis of Neural Network Stability in an Adaptive Flight Control System Sampath Yerramalla 1, Edgar Fuller 2, Martin Mladenovski 1, and Bojan Cukic 1 1 Lane Department of Computer Science and

More information

Ingo Ahrns, Jorg Bruske, Gerald Sommer. Christian Albrechts University of Kiel - Cognitive Systems Group. Preusserstr Kiel - Germany

Ingo Ahrns, Jorg Bruske, Gerald Sommer. Christian Albrechts University of Kiel - Cognitive Systems Group. Preusserstr Kiel - Germany On-line Learning with Dynamic Cell Structures Ingo Ahrns, Jorg Bruske, Gerald Sommer Christian Albrechts University of Kiel - Cognitive Systems Group Preusserstr. 1-9 - 24118 Kiel - Germany Phone: ++49

More information

ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92

ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92 ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92 BIOLOGICAL INSPIRATIONS Some numbers The human brain contains about 10 billion nerve cells (neurons) Each neuron is connected to the others through 10000

More information

Artificial Neural Networks. Edward Gatt

Artificial Neural Networks. Edward Gatt Artificial Neural Networks Edward Gatt What are Neural Networks? Models of the brain and nervous system Highly parallel Process information much more like the brain than a serial computer Learning Very

More information

Neural Network to Control Output of Hidden Node According to Input Patterns

Neural Network to Control Output of Hidden Node According to Input Patterns American Journal of Intelligent Systems 24, 4(5): 96-23 DOI:.5923/j.ajis.2445.2 Neural Network to Control Output of Hidden Node According to Input Patterns Takafumi Sasakawa, Jun Sawamoto 2,*, Hidekazu

More information

Artificial Neural Networks Examination, March 2004

Artificial Neural Networks Examination, March 2004 Artificial Neural Networks Examination, March 2004 Instructions There are SIXTY questions (worth up to 60 marks). The exam mark (maximum 60) will be added to the mark obtained in the laborations (maximum

More information

Introduction to Neural Networks

Introduction to Neural Networks Introduction to Neural Networks What are (Artificial) Neural Networks? Models of the brain and nervous system Highly parallel Process information much more like the brain than a serial computer Learning

More information

Short Communication EVOLUTIONARY NEURAL GAS: A SCALE-FREE SELF- ORGANIZING NEURAL NET WITHOUT TOPOLOGICAL CONSTRAINT

Short Communication EVOLUTIONARY NEURAL GAS: A SCALE-FREE SELF- ORGANIZING NEURAL NET WITHOUT TOPOLOGICAL CONSTRAINT Short Communication EVOLUTIONARY NEURAL GAS: A SCALE-FREE SELF- ORGANIZING NEURAL NET WITHOUT TOPOLOGICAL CONSTRAINT Luigi Lella 1, Ignazio Licata 1) Università Politecnica delle Marche, D.E.I.T., via

More information

Supervised (BPL) verses Hybrid (RBF) Learning. By: Shahed Shahir

Supervised (BPL) verses Hybrid (RBF) Learning. By: Shahed Shahir Supervised (BPL) verses Hybrid (RBF) Learning By: Shahed Shahir 1 Outline I. Introduction II. Supervised Learning III. Hybrid Learning IV. BPL Verses RBF V. Supervised verses Hybrid learning VI. Conclusion

More information

Artificial Neural Network

Artificial Neural Network Artificial Neural Network Contents 2 What is ANN? Biological Neuron Structure of Neuron Types of Neuron Models of Neuron Analogy with human NN Perceptron OCR Multilayer Neural Network Back propagation

More information

y(n) Time Series Data

y(n) Time Series Data Recurrent SOM with Local Linear Models in Time Series Prediction Timo Koskela, Markus Varsta, Jukka Heikkonen, and Kimmo Kaski Helsinki University of Technology Laboratory of Computational Engineering

More information

Artificial Neural Networks Examination, June 2005

Artificial Neural Networks Examination, June 2005 Artificial Neural Networks Examination, June 2005 Instructions There are SIXTY questions. (The pass mark is 30 out of 60). For each question, please select a maximum of ONE of the given answers (either

More information

Artificial Neural Networks Examination, June 2004

Artificial Neural Networks Examination, June 2004 Artificial Neural Networks Examination, June 2004 Instructions There are SIXTY questions (worth up to 60 marks). The exam mark (maximum 60) will be added to the mark obtained in the laborations (maximum

More information

Extending the Associative Rule Chaining Architecture for Multiple Arity Rules

Extending the Associative Rule Chaining Architecture for Multiple Arity Rules Extending the Associative Rule Chaining Architecture for Multiple Arity Rules Nathan Burles, James Austin, and Simon O Keefe Advanced Computer Architectures Group Department of Computer Science University

More information

Neural Networks Introduction

Neural Networks Introduction Neural Networks Introduction H.A Talebi Farzaneh Abdollahi Department of Electrical Engineering Amirkabir University of Technology Winter 2011 H. A. Talebi, Farzaneh Abdollahi Neural Networks 1/22 Biological

More information

( t) Identification and Control of a Nonlinear Bioreactor Plant Using Classical and Dynamical Neural Networks

( t) Identification and Control of a Nonlinear Bioreactor Plant Using Classical and Dynamical Neural Networks Identification and Control of a Nonlinear Bioreactor Plant Using Classical and Dynamical Neural Networks Mehmet Önder Efe Electrical and Electronics Engineering Boðaziçi University, Bebek 80815, Istanbul,

More information

Generalization and Function Approximation

Generalization and Function Approximation Generalization and Function Approximation 0 Generalization and Function Approximation Suggested reading: Chapter 8 in R. S. Sutton, A. G. Barto: Reinforcement Learning: An Introduction MIT Press, 1998.

More information

Artificial Neural Network and Fuzzy Logic

Artificial Neural Network and Fuzzy Logic Artificial Neural Network and Fuzzy Logic 1 Syllabus 2 Syllabus 3 Books 1. Artificial Neural Networks by B. Yagnanarayan, PHI - (Cover Topologies part of unit 1 and All part of Unit 2) 2. Neural Networks

More information

Convergence of Hybrid Algorithm with Adaptive Learning Parameter for Multilayer Neural Network

Convergence of Hybrid Algorithm with Adaptive Learning Parameter for Multilayer Neural Network Convergence of Hybrid Algorithm with Adaptive Learning Parameter for Multilayer Neural Network Fadwa DAMAK, Mounir BEN NASR, Mohamed CHTOUROU Department of Electrical Engineering ENIS Sfax, Tunisia {fadwa_damak,

More information

Multilayer Neural Networks

Multilayer Neural Networks Multilayer Neural Networks Multilayer Neural Networks Discriminant function flexibility NON-Linear But with sets of linear parameters at each layer Provably general function approximators for sufficient

More information

Effects of Interactive Function Forms in a Self-Organized Critical Model Based on Neural Networks

Effects of Interactive Function Forms in a Self-Organized Critical Model Based on Neural Networks Commun. Theor. Phys. (Beijing, China) 40 (2003) pp. 607 613 c International Academic Publishers Vol. 40, No. 5, November 15, 2003 Effects of Interactive Function Forms in a Self-Organized Critical Model

More information

4. Multilayer Perceptrons

4. Multilayer Perceptrons 4. Multilayer Perceptrons This is a supervised error-correction learning algorithm. 1 4.1 Introduction A multilayer feedforward network consists of an input layer, one or more hidden layers, and an output

More information

Keywords- Source coding, Huffman encoding, Artificial neural network, Multilayer perceptron, Backpropagation algorithm

Keywords- Source coding, Huffman encoding, Artificial neural network, Multilayer perceptron, Backpropagation algorithm Volume 4, Issue 5, May 2014 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Huffman Encoding

More information

ANN Control of Non-Linear and Unstable System and its Implementation on Inverted Pendulum

ANN Control of Non-Linear and Unstable System and its Implementation on Inverted Pendulum Research Article International Journal of Current Engineering and Technology E-ISSN 2277 4106, P-ISSN 2347-5161 2014 INPRESSCO, All Rights Reserved Available at http://inpressco.com/category/ijcet ANN

More information

2- AUTOASSOCIATIVE NET - The feedforward autoassociative net considered in this section is a special case of the heteroassociative net.

2- AUTOASSOCIATIVE NET - The feedforward autoassociative net considered in this section is a special case of the heteroassociative net. 2- AUTOASSOCIATIVE NET - The feedforward autoassociative net considered in this section is a special case of the heteroassociative net. - For an autoassociative net, the training input and target output

More information

AI Programming CS F-20 Neural Networks

AI Programming CS F-20 Neural Networks AI Programming CS662-2008F-20 Neural Networks David Galles Department of Computer Science University of San Francisco 20-0: Symbolic AI Most of this class has been focused on Symbolic AI Focus or symbols

More information

Integer weight training by differential evolution algorithms

Integer weight training by differential evolution algorithms Integer weight training by differential evolution algorithms V.P. Plagianakos, D.G. Sotiropoulos, and M.N. Vrahatis University of Patras, Department of Mathematics, GR-265 00, Patras, Greece. e-mail: vpp

More information

ESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN ,

ESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN , Relevance determination in learning vector quantization Thorsten Bojer, Barbara Hammer, Daniel Schunk, and Katharina Tluk von Toschanowitz University of Osnabrück, Department of Mathematics/ Computer Science,

More information

CS 4700: Foundations of Artificial Intelligence Ungraded Homework Solutions

CS 4700: Foundations of Artificial Intelligence Ungraded Homework Solutions CS 4700: Foundations of Artificial Intelligence Ungraded Homework Solutions 1. Neural Networks: a. There are 2 2n distinct Boolean functions over n inputs. Thus there are 16 distinct Boolean functions

More information

Christian Mohr

Christian Mohr Christian Mohr 20.12.2011 Recurrent Networks Networks in which units may have connections to units in the same or preceding layers Also connections to the unit itself possible Already covered: Hopfield

More information

The Minimum Number of Hidden Neurons Does Not Necessarily Provide the Best Generalization

The Minimum Number of Hidden Neurons Does Not Necessarily Provide the Best Generalization The Minimum Number of Hidden Neurons Does Not Necessarily Provide the Best Generalization Jason M. Kinser The Institute for Biosciences, Bioinformatics, and Biotechnology George Mason University, MSN 4E3

More information

Multilayer Neural Networks

Multilayer Neural Networks Multilayer Neural Networks Introduction Goal: Classify objects by learning nonlinearity There are many problems for which linear discriminants are insufficient for minimum error In previous methods, the

More information

Feedforward Neural Nets and Backpropagation

Feedforward Neural Nets and Backpropagation Feedforward Neural Nets and Backpropagation Julie Nutini University of British Columbia MLRG September 28 th, 2016 1 / 23 Supervised Learning Roadmap Supervised Learning: Assume that we are given the features

More information

ADALINE for Pattern Classification

ADALINE for Pattern Classification POLYTECHNIC UNIVERSITY Department of Computer and Information Science ADALINE for Pattern Classification K. Ming Leung Abstract: A supervised learning algorithm known as the Widrow-Hoff rule, or the Delta

More information

An artificial neural networks (ANNs) model is a functional abstraction of the

An artificial neural networks (ANNs) model is a functional abstraction of the CHAPER 3 3. Introduction An artificial neural networs (ANNs) model is a functional abstraction of the biological neural structures of the central nervous system. hey are composed of many simple and highly

More information

Artificial Neural Networks The Introduction

Artificial Neural Networks The Introduction Artificial Neural Networks The Introduction 01001110 01100101 01110101 01110010 01101111 01101110 01101111 01110110 01100001 00100000 01110011 01101011 01110101 01110000 01101001 01101110 01100001 00100000

More information

Neural Networks and Fuzzy Logic Rajendra Dept.of CSE ASCET

Neural Networks and Fuzzy Logic Rajendra Dept.of CSE ASCET Unit-. Definition Neural network is a massively parallel distributed processing system, made of highly inter-connected neural computing elements that have the ability to learn and thereby acquire knowledge

More information

Pattern Classification

Pattern Classification Pattern Classification All materials in these slides were taen from Pattern Classification (2nd ed) by R. O. Duda,, P. E. Hart and D. G. Stor, John Wiley & Sons, 2000 with the permission of the authors

More information

An Adaptive LQG Combined With the MRAS Based LFFC for Motion Control Systems

An Adaptive LQG Combined With the MRAS Based LFFC for Motion Control Systems Journal of Automation Control Engineering Vol 3 No 2 April 2015 An Adaptive LQG Combined With the MRAS Based LFFC for Motion Control Systems Nguyen Duy Cuong Nguyen Van Lanh Gia Thi Dinh Electronics Faculty

More information

On Node-Fault-Injection Training of an RBF Network

On Node-Fault-Injection Training of an RBF Network On Node-Fault-Injection Training of an RBF Network John Sum 1, Chi-sing Leung 2, and Kevin Ho 3 1 Institute of E-Commerce, National Chung Hsing University Taichung 402, Taiwan pfsum@nchu.edu.tw 2 Department

More information

Several ways to solve the MSO problem

Several ways to solve the MSO problem Several ways to solve the MSO problem J. J. Steil - Bielefeld University - Neuroinformatics Group P.O.-Box 0 0 3, D-3350 Bielefeld - Germany Abstract. The so called MSO-problem, a simple superposition

More information

GLOBAL ANALYSIS OF PIECEWISE LINEAR SYSTEMS USING IMPACT MAPS AND QUADRATIC SURFACE LYAPUNOV FUNCTIONS

GLOBAL ANALYSIS OF PIECEWISE LINEAR SYSTEMS USING IMPACT MAPS AND QUADRATIC SURFACE LYAPUNOV FUNCTIONS GLOBAL ANALYSIS OF PIECEWISE LINEAR SYSTEMS USING IMPACT MAPS AND QUADRATIC SURFACE LYAPUNOV FUNCTIONS Jorge M. Gonçalves, Alexandre Megretski y, Munther A. Dahleh y California Institute of Technology

More information

In the Name of God. Lectures 15&16: Radial Basis Function Networks

In the Name of God. Lectures 15&16: Radial Basis Function Networks 1 In the Name of God Lectures 15&16: Radial Basis Function Networks Some Historical Notes Learning is equivalent to finding a surface in a multidimensional space that provides a best fit to the training

More information

A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier

A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier Seiichi Ozawa, Shaoning Pang, and Nikola Kasabov Graduate School of Science and Technology, Kobe

More information

Immediate Reward Reinforcement Learning for Projective Kernel Methods

Immediate Reward Reinforcement Learning for Projective Kernel Methods ESANN'27 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), 25-27 April 27, d-side publi., ISBN 2-9337-7-2. Immediate Reward Reinforcement Learning for Projective Kernel Methods

More information

Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption

Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption ANDRÉ NUNES DE SOUZA, JOSÉ ALFREDO C. ULSON, IVAN NUNES

More information

PDF hosted at the Radboud Repository of the Radboud University Nijmegen

PDF hosted at the Radboud Repository of the Radboud University Nijmegen PDF hosted at the Radboud Repository of the Radboud University Nijmegen The following full text is a preprint version which may differ from the publisher's version. For additional information about this

More information

Matching the dimensionality of maps with that of the data

Matching the dimensionality of maps with that of the data Matching the dimensionality of maps with that of the data COLIN FYFE Applied Computational Intelligence Research Unit, The University of Paisley, Paisley, PA 2BE SCOTLAND. Abstract Topographic maps are

More information

Using a Hopfield Network: A Nuts and Bolts Approach

Using a Hopfield Network: A Nuts and Bolts Approach Using a Hopfield Network: A Nuts and Bolts Approach November 4, 2013 Gershon Wolfe, Ph.D. Hopfield Model as Applied to Classification Hopfield network Training the network Updating nodes Sequencing of

More information

Data Mining Part 5. Prediction

Data Mining Part 5. Prediction Data Mining Part 5. Prediction 5.5. Spring 2010 Instructor: Dr. Masoud Yaghini Outline How the Brain Works Artificial Neural Networks Simple Computing Elements Feed-Forward Networks Perceptrons (Single-layer,

More information

Blind Equalization Formulated as a Self-organized Learning Process

Blind Equalization Formulated as a Self-organized Learning Process Blind Equalization Formulated as a Self-organized Learning Process Simon Haykin Communications Research Laboratory McMaster University 1280 Main Street West Hamilton, Ontario, Canada L8S 4K1 Abstract In

More information

Learning Vector Quantization

Learning Vector Quantization Learning Vector Quantization Neural Computation : Lecture 18 John A. Bullinaria, 2015 1. SOM Architecture and Algorithm 2. Vector Quantization 3. The Encoder-Decoder Model 4. Generalized Lloyd Algorithms

More information

POWER SYSTEM DYNAMIC SECURITY ASSESSMENT CLASSICAL TO MODERN APPROACH

POWER SYSTEM DYNAMIC SECURITY ASSESSMENT CLASSICAL TO MODERN APPROACH Abstract POWER SYSTEM DYNAMIC SECURITY ASSESSMENT CLASSICAL TO MODERN APPROACH A.H.M.A.Rahim S.K.Chakravarthy Department of Electrical Engineering K.F. University of Petroleum and Minerals Dhahran. Dynamic

More information

Lecture 7 Artificial neural networks: Supervised learning

Lecture 7 Artificial neural networks: Supervised learning Lecture 7 Artificial neural networks: Supervised learning Introduction, or how the brain works The neuron as a simple computing element The perceptron Multilayer neural networks Accelerated learning in

More information

Neural Networks. Hopfield Nets and Auto Associators Fall 2017

Neural Networks. Hopfield Nets and Auto Associators Fall 2017 Neural Networks Hopfield Nets and Auto Associators Fall 2017 1 Story so far Neural networks for computation All feedforward structures But what about.. 2 Loopy network Θ z = ቊ +1 if z > 0 1 if z 0 y i

More information

EEE 241: Linear Systems

EEE 241: Linear Systems EEE 4: Linear Systems Summary # 3: Introduction to artificial neural networks DISTRIBUTED REPRESENTATION An ANN consists of simple processing units communicating with each other. The basic elements of

More information

Introduction to Neural Networks: Structure and Training

Introduction to Neural Networks: Structure and Training Introduction to Neural Networks: Structure and Training Professor Q.J. Zhang Department of Electronics Carleton University, Ottawa, Canada www.doe.carleton.ca/~qjz, qjz@doe.carleton.ca A Quick Illustration

More information

COMS 4771 Introduction to Machine Learning. Nakul Verma

COMS 4771 Introduction to Machine Learning. Nakul Verma COMS 4771 Introduction to Machine Learning Nakul Verma Announcements HW1 due next lecture Project details are available decide on the group and topic by Thursday Last time Generative vs. Discriminative

More information

Artificial Neural Networks Examination, March 2002

Artificial Neural Networks Examination, March 2002 Artificial Neural Networks Examination, March 2002 Instructions There are SIXTY questions (worth up to 60 marks). The exam mark (maximum 60) will be added to the mark obtained in the laborations (maximum

More information

HOPFIELD neural networks (HNNs) are a class of nonlinear

HOPFIELD neural networks (HNNs) are a class of nonlinear IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, VOL. 52, NO. 4, APRIL 2005 213 Stochastic Noise Process Enhancement of Hopfield Neural Networks Vladimir Pavlović, Member, IEEE, Dan Schonfeld,

More information

CMSC 421: Neural Computation. Applications of Neural Networks

CMSC 421: Neural Computation. Applications of Neural Networks CMSC 42: Neural Computation definition synonyms neural networks artificial neural networks neural modeling connectionist models parallel distributed processing AI perspective Applications of Neural Networks

More information

Neural Networks Lecture 2:Single Layer Classifiers

Neural Networks Lecture 2:Single Layer Classifiers Neural Networks Lecture 2:Single Layer Classifiers H.A Talebi Farzaneh Abdollahi Department of Electrical Engineering Amirkabir University of Technology Winter 2011. A. Talebi, Farzaneh Abdollahi Neural

More information

Curvilinear Components Analysis and Bregman Divergences

Curvilinear Components Analysis and Bregman Divergences and Machine Learning. Bruges (Belgium), 8-3 April, d-side publi., ISBN -9337--. Curvilinear Components Analysis and Bregman Divergences Jigang Sun, Malcolm Crowe and Colin Fyfe Applied Computational Intelligence

More information

488 Solutions to the XOR Problem

488 Solutions to the XOR Problem 488 Solutions to the XOR Problem Frans M. Coetzee * eoetzee@eee.emu.edu Department of Electrical Engineering Carnegie Mellon University Pittsburgh, PA 15213 Virginia L. Stonick ginny@eee.emu.edu Department

More information

FEEDBACK GMDH-TYPE NEURAL NETWORK AND ITS APPLICATION TO MEDICAL IMAGE ANALYSIS OF LIVER CANCER. Tadashi Kondo and Junji Ueno

FEEDBACK GMDH-TYPE NEURAL NETWORK AND ITS APPLICATION TO MEDICAL IMAGE ANALYSIS OF LIVER CANCER. Tadashi Kondo and Junji Ueno International Journal of Innovative Computing, Information and Control ICIC International c 2012 ISSN 1349-4198 Volume 8, Number 3(B), March 2012 pp. 2285 2300 FEEDBACK GMDH-TYPE NEURAL NETWORK AND ITS

More information

Computational Intelligence Lecture 6: Associative Memory

Computational Intelligence Lecture 6: Associative Memory Computational Intelligence Lecture 6: Associative Memory Farzaneh Abdollahi Department of Electrical Engineering Amirkabir University of Technology Fall 2011 Farzaneh Abdollahi Computational Intelligence

More information

Implicitely and Densely Discrete Black-Box Optimization Problems

Implicitely and Densely Discrete Black-Box Optimization Problems Implicitely and Densely Discrete Black-Box Optimization Problems L. N. Vicente September 26, 2008 Abstract This paper addresses derivative-free optimization problems where the variables lie implicitly

More information

Dynamic Classification of Power Systems via Kohonen Neural Network

Dynamic Classification of Power Systems via Kohonen Neural Network Dynamic Classification of Power Systems via Kohonen Neural Network S.F Mahdavizadeh Iran University of Science & Technology, Tehran, Iran M.A Sandidzadeh Iran University of Science & Technology, Tehran,

More information

Stability of Amygdala Learning System Using Cell-To-Cell Mapping Algorithm

Stability of Amygdala Learning System Using Cell-To-Cell Mapping Algorithm Stability of Amygdala Learning System Using Cell-To-Cell Mapping Algorithm Danial Shahmirzadi, Reza Langari Department of Mechanical Engineering Texas A&M University College Station, TX 77843 USA danial_shahmirzadi@yahoo.com,

More information

Introduction to Artificial Neural Networks

Introduction to Artificial Neural Networks Facultés Universitaires Notre-Dame de la Paix 27 March 2007 Outline 1 Introduction 2 Fundamentals Biological neuron Artificial neuron Artificial Neural Network Outline 3 Single-layer ANN Perceptron Adaline

More information

EM-algorithm for Training of State-space Models with Application to Time Series Prediction

EM-algorithm for Training of State-space Models with Application to Time Series Prediction EM-algorithm for Training of State-space Models with Application to Time Series Prediction Elia Liitiäinen, Nima Reyhani and Amaury Lendasse Helsinki University of Technology - Neural Networks Research

More information

This is a repository copy of Improving the associative rule chaining architecture.

This is a repository copy of Improving the associative rule chaining architecture. This is a repository copy of Improving the associative rule chaining architecture. White Rose Research Online URL for this paper: http://eprints.whiterose.ac.uk/75674/ Version: Accepted Version Book Section:

More information

9 Competitive Neural Networks

9 Competitive Neural Networks 9 Competitive Neural Networks the basic competitive neural network consists of two layers of neurons: The distance-measure layer, The competitive layer, also known as a Winner-Takes-All (WTA) layer The

More information

On the convergence speed of artificial neural networks in the solving of linear systems

On the convergence speed of artificial neural networks in the solving of linear systems Available online at http://ijimsrbiauacir/ Int J Industrial Mathematics (ISSN 8-56) Vol 7, No, 5 Article ID IJIM-479, 9 pages Research Article On the convergence speed of artificial neural networks in

More information

A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier

A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier Seiichi Ozawa 1, Shaoning Pang 2, and Nikola Kasabov 2 1 Graduate School of Science and Technology,

More information

18.6 Regression and Classification with Linear Models

18.6 Regression and Classification with Linear Models 18.6 Regression and Classification with Linear Models 352 The hypothesis space of linear functions of continuous-valued inputs has been used for hundreds of years A univariate linear function (a straight

More information

Hover Control for Helicopter Using Neural Network-Based Model Reference Adaptive Controller

Hover Control for Helicopter Using Neural Network-Based Model Reference Adaptive Controller Vol.13 No.1, 217 مجلد 13 العدد 217 1 Hover Control for Helicopter Using Neural Network-Based Model Reference Adaptive Controller Abdul-Basset A. Al-Hussein Electrical Engineering Department Basrah University

More information

Frequency Selective Surface Design Based on Iterative Inversion of Neural Networks

Frequency Selective Surface Design Based on Iterative Inversion of Neural Networks J.N. Hwang, J.J. Choi, S. Oh, R.J. Marks II, "Query learning based on boundary search and gradient computation of trained multilayer perceptrons", Proceedings of the International Joint Conference on Neural

More information

Learning Vector Quantization (LVQ)

Learning Vector Quantization (LVQ) Learning Vector Quantization (LVQ) Introduction to Neural Computation : Guest Lecture 2 John A. Bullinaria, 2007 1. The SOM Architecture and Algorithm 2. What is Vector Quantization? 3. The Encoder-Decoder

More information

A Globally Stabilizing Receding Horizon Controller for Neutrally Stable Linear Systems with Input Constraints 1

A Globally Stabilizing Receding Horizon Controller for Neutrally Stable Linear Systems with Input Constraints 1 A Globally Stabilizing Receding Horizon Controller for Neutrally Stable Linear Systems with Input Constraints 1 Ali Jadbabaie, Claudio De Persis, and Tae-Woong Yoon 2 Department of Electrical Engineering

More information

Artificial Neural Networks

Artificial Neural Networks Artificial Neural Networks 鮑興國 Ph.D. National Taiwan University of Science and Technology Outline Perceptrons Gradient descent Multi-layer networks Backpropagation Hidden layer representations Examples

More information

Neural Network Based Response Surface Methods a Comparative Study

Neural Network Based Response Surface Methods a Comparative Study . LS-DYNA Anwenderforum, Ulm Robustheit / Optimierung II Neural Network Based Response Surface Methods a Comparative Study Wolfram Beyer, Martin Liebscher, Michael Beer, Wolfgang Graf TU Dresden, Germany

More information

Unit 8: Introduction to neural networks. Perceptrons

Unit 8: Introduction to neural networks. Perceptrons Unit 8: Introduction to neural networks. Perceptrons D. Balbontín Noval F. J. Martín Mateos J. L. Ruiz Reina A. Riscos Núñez Departamento de Ciencias de la Computación e Inteligencia Artificial Universidad

More information

CSE 352 (AI) LECTURE NOTES Professor Anita Wasilewska. NEURAL NETWORKS Learning

CSE 352 (AI) LECTURE NOTES Professor Anita Wasilewska. NEURAL NETWORKS Learning CSE 352 (AI) LECTURE NOTES Professor Anita Wasilewska NEURAL NETWORKS Learning Neural Networks Classifier Short Presentation INPUT: classification data, i.e. it contains an classification (class) attribute.

More information

Administration. Registration Hw3 is out. Lecture Captioning (Extra-Credit) Scribing lectures. Questions. Due on Thursday 10/6

Administration. Registration Hw3 is out. Lecture Captioning (Extra-Credit) Scribing lectures. Questions. Due on Thursday 10/6 Administration Registration Hw3 is out Due on Thursday 10/6 Questions Lecture Captioning (Extra-Credit) Look at Piazza for details Scribing lectures With pay; come talk to me/send email. 1 Projects Projects

More information

Non-linear Measure Based Process Monitoring and Fault Diagnosis

Non-linear Measure Based Process Monitoring and Fault Diagnosis Non-linear Measure Based Process Monitoring and Fault Diagnosis 12 th Annual AIChE Meeting, Reno, NV [275] Data Driven Approaches to Process Control 4:40 PM, Nov. 6, 2001 Sandeep Rajput Duane D. Bruns

More information

Artificial Neural Networks. Historical description

Artificial Neural Networks. Historical description Artificial Neural Networks Historical description Victor G. Lopez 1 / 23 Artificial Neural Networks (ANN) An artificial neural network is a computational model that attempts to emulate the functions of

More information

C4 Phenomenological Modeling - Regression & Neural Networks : Computational Modeling and Simulation Instructor: Linwei Wang

C4 Phenomenological Modeling - Regression & Neural Networks : Computational Modeling and Simulation Instructor: Linwei Wang C4 Phenomenological Modeling - Regression & Neural Networks 4040-849-03: Computational Modeling and Simulation Instructor: Linwei Wang Recall.. The simple, multiple linear regression function ŷ(x) = a

More information

DEEP LEARNING AND NEURAL NETWORKS: BACKGROUND AND HISTORY

DEEP LEARNING AND NEURAL NETWORKS: BACKGROUND AND HISTORY DEEP LEARNING AND NEURAL NETWORKS: BACKGROUND AND HISTORY 1 On-line Resources http://neuralnetworksanddeeplearning.com/index.html Online book by Michael Nielsen http://matlabtricks.com/post-5/3x3-convolution-kernelswith-online-demo

More information

Neural Networks. Prof. Dr. Rudolf Kruse. Computational Intelligence Group Faculty for Computer Science

Neural Networks. Prof. Dr. Rudolf Kruse. Computational Intelligence Group Faculty for Computer Science Neural Networks Prof. Dr. Rudolf Kruse Computational Intelligence Group Faculty for Computer Science kruse@iws.cs.uni-magdeburg.de Rudolf Kruse Neural Networks 1 Hopfield Networks Rudolf Kruse Neural Networks

More information

INTRODUCTION TO NEURAL NETWORKS

INTRODUCTION TO NEURAL NETWORKS INTRODUCTION TO NEURAL NETWORKS R. Beale & T.Jackson: Neural Computing, an Introduction. Adam Hilger Ed., Bristol, Philadelphia and New York, 990. THE STRUCTURE OF THE BRAIN The brain consists of about

More information

MCE/EEC 647/747: Robot Dynamics and Control. Lecture 8: Basic Lyapunov Stability Theory

MCE/EEC 647/747: Robot Dynamics and Control. Lecture 8: Basic Lyapunov Stability Theory MCE/EEC 647/747: Robot Dynamics and Control Lecture 8: Basic Lyapunov Stability Theory Reading: SHV Appendix Mechanical Engineering Hanz Richter, PhD MCE503 p.1/17 Stability in the sense of Lyapunov A

More information

Neural Networks. Fundamentals Framework for distributed processing Network topologies Training of ANN s Notation Perceptron Back Propagation

Neural Networks. Fundamentals Framework for distributed processing Network topologies Training of ANN s Notation Perceptron Back Propagation Neural Networks Fundamentals Framework for distributed processing Network topologies Training of ANN s Notation Perceptron Back Propagation Neural Networks Historical Perspective A first wave of interest

More information

Power Grid State Estimation after a Cyber-Physical Attack under the AC Power Flow Model

Power Grid State Estimation after a Cyber-Physical Attack under the AC Power Flow Model Power Grid State Estimation after a Cyber-Physical Attack under the AC Power Flow Model Saleh Soltan, Gil Zussman Department of Electrical Engineering Columbia University, New York, NY Email: {saleh,gil}@ee.columbia.edu

More information

Lecture 4: Feed Forward Neural Networks

Lecture 4: Feed Forward Neural Networks Lecture 4: Feed Forward Neural Networks Dr. Roman V Belavkin Middlesex University BIS4435 Biological neurons and the brain A Model of A Single Neuron Neurons as data-driven models Neural Networks Training

More information

COGS Q250 Fall Homework 7: Learning in Neural Networks Due: 9:00am, Friday 2nd November.

COGS Q250 Fall Homework 7: Learning in Neural Networks Due: 9:00am, Friday 2nd November. COGS Q250 Fall 2012 Homework 7: Learning in Neural Networks Due: 9:00am, Friday 2nd November. For the first two questions of the homework you will need to understand the learning algorithm using the delta

More information

On the SIR s ( Signal -to- Interference -Ratio) in. Discrete-Time Autonomous Linear Networks

On the SIR s ( Signal -to- Interference -Ratio) in. Discrete-Time Autonomous Linear Networks On the SIR s ( Signal -to- Interference -Ratio) in arxiv:93.9v [physics.data-an] 9 Mar 9 Discrete-Time Autonomous Linear Networks Zekeriya Uykan Abstract In this letter, we improve the results in [5] by

More information

Review on Aircraft Gain Scheduling

Review on Aircraft Gain Scheduling Review on Aircraft Gain Scheduling Z. Y. Kung * and I. F. Nusyirwan a Department of Aeronautical Engineering, Faculty of Mechanical Engineering, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia.

More information

NONLINEAR CLASSIFICATION AND REGRESSION. J. Elder CSE 4404/5327 Introduction to Machine Learning and Pattern Recognition

NONLINEAR CLASSIFICATION AND REGRESSION. J. Elder CSE 4404/5327 Introduction to Machine Learning and Pattern Recognition NONLINEAR CLASSIFICATION AND REGRESSION Nonlinear Classification and Regression: Outline 2 Multi-Layer Perceptrons The Back-Propagation Learning Algorithm Generalized Linear Models Radial Basis Function

More information

Iterative Laplacian Score for Feature Selection

Iterative Laplacian Score for Feature Selection Iterative Laplacian Score for Feature Selection Linling Zhu, Linsong Miao, and Daoqiang Zhang College of Computer Science and echnology, Nanjing University of Aeronautics and Astronautics, Nanjing 2006,

More information