Letters. Comments on Approximation Capability in by Multilayer Feedforward Networks and Related Problems

Size: px
Start display at page:

Download "Letters. Comments on Approximation Capability in by Multilayer Feedforward Networks and Related Problems"

Transcription

1 714 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 9, NO. 4, JULY 1998 Letters Comments on Approxmaton Capablty n by Multlayer Feedforward Networs and Related Problems Guang-Bn Huang and Haroon A. Babr Abstract In the above paper 1 Chen et al. nvestgated the capablty of unformly approxmatng functons n C(R n ) by standard feedforward neural networs. They found that the boundedness condton on the sgmodal functon plays an essental role n the approxmaton, and conjectured that the boundedness of the sgmodal functon s a necessary and suffcent condton for the valdty of the approxmaton theorem. However, we fnd that the conjecture s not correct, that s, the boundedness condton s not suffcent or necessary n C(R n ). Instead, boundedness and unequal lmts at nfntes condtons on the actvaton functons are suffcent, but not necessary n C(R n ). Index Terms Actvaton functons, approxmaton capablty, boundedness, feedforward networs. I. INTRODUCTION In the above paper 1 and n the last several years many researchers (e.g., [1] [9]) have explored the capabltes of standard multlayer feedforward neural networs to approxmate general mappngs from one-dmensonal space to another. Two aspects of the approxmaton problem have been dscussed: unform approxmaton everywhere n C(R n ) 2 [7] and L p approxmaton almost everywhere n compacts of R n [2], [5], [6]. In the above paper Chen et al. clamed that generalzed sgmodal functons 3 can be used as actvaton functons for a sngle hdden layer neural networ to approxmate any arbtrary mappngs n C(R n ). Further, they conjectured that the boundedness of the sgmodal functon s a necessary and suffcent condton for the valdty of the approxmaton theorem (Theorem 1, p. 25 n the above paper). However, we fnd that the conjecture s not correct, that s, the boundedness condton s not suffcent or necessary n C(R n ). Instead, the boundedness and unequal lmts at nfntes condtons on the actvaton functons are suffcent, but not necessary n C(R n ). II. ON SUFFICIENT CONDITIONS Chen et al. proved the followng mportant approxmaton theorem (Theorem 1, p. 25 n the above paper). Lemma 2.1: If s a bounded generalzed sgmodal functon, and f (x) s a contnuous functon n R, for whch lm x!01 f (x) =A and lm x!+1 f (x) =B, where A; B are constants, then for any >0, there exst ; ;w ;b such that j (wx + b) 0 f (x)j < holds for all x 2 R. Chen et al. state that we conjecture the boundedness of (x) s a necessary and suffcent condton for the valdty of (Theorem 1, p. 26). In ths paper, we show that the boundedness of (x) s nether Manuscrpt receved May 13, 1997; revsed March 22, The authors are wth the School of Electrcal and Electronc Engneerng, Nanyang Technologcal Unversty, Sngapore Publsher Item Identfer S (98)04541-X. 1 T. Chen, H. Chen, and R.-W. Lu, IEEE Trans. Neural Networs, vol. 6, pp , Jan C(R n )=ff 2 C(R n ): lmjxj!1 f (x)exstsg. 3 (x) s called a generalzed sgmodal functon f lm x!01 (x) =0 and lm x!+1 (x) =1. a necessary nor a suffcent condton. Instead, the boundedness and unequal lmts at nfntes condtons on the actvaton functons are suffcent, but not necessary. In order to prove our vewpont we only need to extend ther approxmaton theorem (Theorem 1, p. 25) to more general ones. Consder a lnear subspace of C(R), L C(R), where L = f 2 C(R): lm f (x) = lm f (x) : (1) x!01 x!+1 For g 2 C(R), (w; b) 2 R 2, denote g (w; b) 2 C(R) by g (w; b) (x) = g(wx + b), and let F (g) C(R) denote the set of g (w; b), (w; b) 2 R 2. It s obvous that f g 2 L then for all (w; b), g (w; b) 2 L. Thus, we have the followng theorem. Theorem 2.1: For g 2 L, the span of F (g) L s a subset of L. Ths follows from the defnton of the span, and the fact that L s a lnear subspace of C(R). Remar: Because radal bass functons [1] have equal lmts at nfntes, accordng to Theorem 2.1, they cannot be used as actvaton functons by whch a neural networ can approxmate arbtrary contnuous mappngs wth any precson n R although they can be used when consdered n a compact set of R as shown n [1]. Theorem 2.1 shows that gven any (bounded or unbounded) functon g(x) n R, f lm x!01 g(x) = lm x!+1 g(x), then for any arbtrary (contnuous or uncontnuous) mappng f (x) n R wth lm x!01 f (x) 6= lm x!+1 f (x), and for every > 0 there does not exst w, b and 2 R and 2 N such that j g(w x+b )0f(x)j <hold for all x 2 R. Ths means that only boundedness of actvaton functons s not suffcent for standard feedforward neural networs to approxmate any arbtrary mappng n C(R n ). Instead, we fnd that the boundedness and unequal lmts at nfntes condtons on the actvaton functons are suffcent n C(R n ). Theorem 2.2: Gven a bounded functon g(x) n R and there exst lmts lm x!01 g(x), lm x!+1 g(x), and lm x!01 g(x) 6= lm x!+1 g(x), then for any arbtrary mappng f (x) n C(R), for every >0 there exst w, b and, and 2 N such that g(w x + b ) 0 f (x) < holds for all x 2 R. Proof: We can prove the theorem n two steps: 1) lm x!01 g(x) = 0 and lm x!+1 g(x) = M, where M can be any arbtrary nonzero real value and 2) lm x!01 g(x) =N and lm x!+1 g(x) =M, where M and N can be any arbtrary unequal real values. 1) Consder lm x!01 g(x) = 0 and lm x!+1 g(x) = M, where M can be any arbtrary nonzero real value. Let g 1 (x) = g(x)=m. Then lm x!01 g 1(x) =0and lm x!+1 g 1(x) = 1. Accordng to Lemma 2.1, for every >0, there exst 2 N and ;w 0 ;b ; = 1; 111; such that j 0 g 1(w x +

2 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 9, NO. 4, JULY b )0f (x)j <. Let = 0 =M; =1; 111;. Then we have g(w x + b ) 0 f (x) < holds for all x 2 R. 2) Consder lm x!01 g(x) =N and lm x!+1 g(x) =M, where M and N can be any unequal real values. Set g 1 (x) = g(x) 0 N. Thus lm x!01 g(x) =0and lm x!+1 g(x) = M 0 N. Accordng to the result obtaned n the prevous step, for every >0, there exst N and ;w ;b ;= 1; 111;0 1 such that 01 g 1 (w x + b ) 0 f (x) < (2) holds for all x 2 R. Suppose x 0 2 R and g(x 0 ) 6= 0. Then we can choose w =0and b = x 0 such that g(w x + b )= g(x 0 ). Let = 0N 01 =g(x 0 ). Hence we have = = g(w x + b ) 0 f (x) g(w x + b ) 0 N 01 =g(x 0 ) 1 g(x 0 ) 0 f (x) g 1 (w x + b ) 0 f (x) <: (3) Ths completes the proof. Lemma 2.2: If g(x) n R satsfes that all lnear combnatons g(w x +b ) are dense n C(R), then all lnear combnatons g(w 1 x + b) are dense n C(Rn ), where w 2 R n and w 1 x s the nner product of w and x. Usng Theorem 2.2 and Lemma 2.2, the followng theorem can be obtaned. Theorem 2.3: Gven any bounded functon g(x) n R whch has lmts lm x!01 g(x), lm x!+1 g(x), and lm x!01 g(x) 6= lm x!+1 g(x), then all lnear combnatons g(w 1 x + b) are dense n C(R n ), where w 2 R n and w 1 x s the nner product of w and x. holds for all x 2 R. In other words, although (x) s unbounded, for every mappng f (x) 2 C(R), for every >0 there exst w, b, and and 2 N such that holds for all x 2 R. [(w x + b ) 0 (0w x 0 b )] 0 f (x) < ACKNOWLEDGMENT The authors would le to than the anonymous revewers for ther valuable suggestons, especally for the proof structure of Theorem 2.1. REFERENCES [1] F. Gros and T. Poggo, Networs and the best approxmaton property, Artfcal Intell. Lab., Massachusetts Inst. Technol., A.I. Memo 1164, [2] K. Horn, M. Stnchcombe, and H. Whte, Multlayer feedforward networs are unversal approxmators, Neural Networs, vol. 2, pp , [3] G. Cybeno, Approxmaton by superpostons of a sgmodal functon, Math. Contr., Sgnals, Syst., vol. 2, no. 4, pp , [4] K. Funahash, On the approxmate realzaton of contnuous mappngs by neural networs, Neural Networs, vol. 2, pp , [5] K. Horn, Approxmaton capabltes of multlayer feedforward networs, Neural Networs, vol. 4, pp , [6] M. Leshno, V. Y. Ln, A. Pnus, and S. Schocen, Multlayer feedforward networs wth a nonpolynomal actvaton functon can approxmate any functon, Neural Networs, vol. 6, pp , [7] Y. Ito, Approxmaton of contnuous functons on R d by lnear combnatons of shfted rotatons of a sgmod functon wth and wthout scalng, Neural Networs, vol. 5, pp , [8], Approxmaton capablty of layered neural networs wth sgmod unts on two layers, Neural Comput., vol. 6, pp , [9] K. Horn, M. Stnchcombe, and H. Whte, Unversal approxmaton of an unnown mappng and ts dervatves usng multlayer feedforward networs, Neural Networs, vol. 3, pp , III. ON NECESSARY CONDITIONS Chen et al. conjectured that the boundedness of was a necessary and suffcent condton for the valdty of Lemma 2.1. From Theorem 2.2 we can easly fnd that the conjecture s not correct. In fact, we can defne (x) = e 0x + 1 ; f x<0 x2 0; f x =0 (4) 1+e 0x + 1 x ; f x>0. 2 Then lm x!0 (x) = +1 and lm x!0 (x) = +1, (x) s unbounded n R. It s obvous that 1(x) = (x) 0 (0x) s bounded and has lmts lm x!01 1 (x), lm x!+1 1 (x), and lm x!01 1 (x) 6= lm x!+1 1 (x). Accordng to Theorem 2.2 for any arbtrary mappng f (x) 2 C(R), for every >0 there exst w, b, and and 2 N such that j 1 (w x+b )0f (x)j <

3 716 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 9, NO. 4, JULY 1998 On Chaotc Smulated Annealng Lpo Wang and Kate Smth Abstract Chen and Ahara recently proposed a chaotc smulated annealng approach to solvng optmzaton problems. By addng a negatve self-couplng to a networ model proposed earler by Ahara et al. and gradually removng ths negatve self-couplng, they used the transent chaos for searchng and self-organzng, thereby achevng remarable mprovement over other neural-networ approaches to optmzaton problems wth or wthout smulated annealng. In ths paper we suggest a new approach to chaotc smulated annealng wth guaranteed convergence and mnmzaton of the energy functon by gradually reducng the tme step n the Euler approxmaton of the dfferental equatons that descrbe the contnuous Hopfeld neural networ. Ths approach elmnates the need to carefully select other system parameters. We also generalze the convergence theorems of Chen and Ahara to arbtrarly ncreasng neuronal nput output functons and to less restrctve and yet more compact forms. Index Terms Annealng, chaos, energy functon, Hopfeld, neural networ, optmzaton. I. INTRODUCTION Combnatoral optmzaton problems are ever present n scence and technology. Snce Hopfeld and Tan s semnal wor [6] on solvng the travellng salesman problem wth a Hopfeld neural networ (HNN) [5], the HNN s [4], [5] have been recognzed as powerful tools for optmzaton (e.g., [10], [13]). The HNN s have an ntrgung property that as each neuron n an HNN updates, an energy functon s monotonously reduced untl the networ stablzes. One can therefore map an optmzaton problem to a HNN such that the cost functon of the problem corresponds to the energy functon of the HNN and the fnal state of the HNN thus suggests a soluton to the optmzaton problem wth a low cost value. Whle some researchers have descrbed HNN s as nothng more than nave gradent descent machnes, the neural framewor does brngs about some mportant advantages over other gradent descent technques: prncpally the nherent parallelsm and hardware mplementaton whch can potentally result n great speed-ups over conventonal technques for combnatoral optmzaton [12]. The employment of HNN s to solve problems of real-world sgnfcance has been hampered, however, by problems over the last decade wth soluton qualty and slow development of sutable hardware to enable large szed problems to be solved. Earler attempts at solvng varous optmzaton problems wth the HNN s suffered from the fact that a HNN can often be trapped at a local mnmum n the complex energy terran, whch gves an optmzaton soluton wth an unacceptably hgh cost [15]. Several methods whch allow for temporary energy ncreases, such as smulated annealng [7], have been proposed. Recent advances have now made modfed HNN s compettve wth the best heurstcs for solvng combnatoral optmzaton problems, and ths has been demonstrated on a varety of real-world problems [10], [11]. The search stll contnues however, for further or alternatve mprovements to the standard Manuscrpt receved November 9, Ths wor was supported n part by the Australan Research Councl and Dean Unversty. L. Wang s wth the School of Computng and Mathematcs, Dean Unversty, Clayton, Vctora 3168, Australa. K. Smth s wth the Department of Busness Systems, Monash Unversty, Clayton, Vctora 3168, Australa. Publsher Item Identfer S (98)04613-X. neural algorthms to address the ssue of soluton qualty: partcularly mprovements whch are easly mplementable n hardware. Chen and Ahara [1] recently proposed a chaotc smulated annealng approach. By addng a negatve self-couplng to a transently chaotc neural networ (TCNN) and gradually removng ths negatve self-couplng, they used the transent chaos generated by the fadng negatve self-couplng for searchng and self-organzng, thereby achevng remarable mprovement over other neural-networ methods, n terms of frequency of fndng near-optmal solutons. However, a number of networ parameters must be carefully chosen so as to guarantee the convergence of the TCNN and ts mnmzaton of the energy upon the removal of the transent chaos [2]. In addton, Chen and Ahara [2] used n provng ther convergence theorems a partcular sgmodal functon for all neurons. Hardware mplementatons may not easly ensure ths form of nput output (I/O) functon and may also need to allow for some varatons n I/O among the neurons. In ths letter, we frst suggest an alternatve approach to chaotc smulated annealng wth guaranteed convergence and mnmzaton of the energy functon, but wthout the need for choosng any other system parameters. We then generalze the convergence theorems to arbtrarly ncreasng I/O and to less restrctve and yet more compact forms. II. AN ALTERNATIVE APPROACH TO CHAOTIC SIMULATED ANNEALING The dynamcs of a n-neuron contnuous Hopfeld neural networ (CHNN) [5] are descrbed by du dt = 0 u + j T j V j + I (1) where = 1; 2; 111 n; I s the external nput to neuron and s sometmes called frng threshold when replaced wth 0I. The nternal state of neuron ; u 2 (01; +1), determnes the output of neuron V (t) =f [u (t)] 2 [0; 1] (2) or u (t) =f 01 [V (t)]. Here f s the neuronal I/O functon for neuron. f may dffer from neuron to neuron and does not need to have any symmetry propertes; however, we assume that f s monotonously ncreasng so that f 01 exsts. For example, the sgmod functon f (x) = 1 [1 + tanh(x)] (3) 2 s often used, wth beng the gan of the I/O functon. Hopfeld ntroduced the followng functon [5] for the CHNN: E(t) = ; j Tj V(t) Vj (t) 0 IV(t) + 1 V (t) f 01 () d: (4) 0 = 0 j T j V j 0 I + f 01 (V )= = 0du =dt, provded that T j = T j (regardless the sgn of the self-couplng T ), then de=dt = (@E=@V )(dv =dt) =0(du =dt)1 f 0 1 (du =dt) < 0, f du =dt 6= 0 for at least one, and de=dt = 0 f and only f du =dt = 0 for all. Hence E s a Lyapunov (energy) functon that monotonously decreases as the networ updates untl the networ stablzes. Based on the exstence and the fnteness of such a Lyapunov functon, Hopfeld concluded [5] that the CHNN must stablze tself: ~u(t)! ~u o,ast! +1. Here ~u o s ndependent of tme t and represents the stable fxed pont of the networ.

4 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 9, NO. 4, JULY Now let us consder a dfferent networ whch conssts of n neurons wth the same I/O functons f, but obeys the followng dynamc equatons: u (t +1t) = 1 0 1t u (t)+1t T j V j (t)+i : (5) j We note that n the lmt 1t! 0, ths equaton s the same as the CHNN (1). In fact, (5) s the Euler approxmaton of the CHNN (1). The Euler approxmaton of the CHNN (5) s dentcal to the TCNN of Chen and Ahara [1], [2], [9]: f 1t = 1 and u (t +1)=u (t)+ j T 0 j V j (t)+i 0 (6) =10 1t ; T0 j =1tT j ; I 0 =1tI : (7) By addng a large negatve self-couplng T 0 and gradually removng t n the TCNN (6), Chen and Ahara [1] maredly enhanced the probablty of reachng an optmal or near-optmal soluton. In ths approach, other system parameters, such as and [the gan n the sgmodal functon equaton (3)], must be carefully selected wth respect to the synaptc weght matrx accordng to ther stablty theorems [2], n order to assure networ convergence and energy mnmzaton. We suggest an alternatve approach to chaotc smulated annealng: startng from the Euler approxmaton of the CHNN (5) wth a large tme-step 1t, where the dynamcs are chaotc [2], we gradually reduce the tme-step 1t. The system s guaranteed to converge and to mnmze the CHNN energy functon (4), snce n the lmt of 1t! 0, the system approaches the CHNN whch s stable and mnmzes the CHNN energy functon. Ths approach does not requre dffcult choces of any system parameters to assure networ convergence and energy converson. III. GENERALIZED STABILITY THEOREMS Chen and Ahara [2] derved stablty theorems for both the Euler approxmaton of the CHNN (5) and the TCNN (6) for both synchronous and asychronous updatng, usng a partcular form of the neuronal I/O (3). We now generalze ther theorems to arbtrarly ncreasng I/O functons (2) and to less restrctve but more compact forms. Let us calculate the change n energy between two tme steps when the TCNN (6) s updated synchronously, accordng to (4) 1E(t) E(t +1)0E(t) = ; j 1t 1t T j 0 1V (t)1v j (t) + 1V (t) f 01 [V (t)] 0 1V (t) f 01 [V (t + 1)] +(10 ) fg [V (t + 1)] 0 G [V (t)]g: (8) where G (V ) V f 01 0 () d and 1V (t) V (t +1)0V (t). Expandng G at V (t +1)and usng the fact that G s concave-up [8], we obtan G [V (t + 1)] 0 G [V (t)] G 0 [V (t + 1)] 1V (t) mn d G [1V d V (t)] 2 (9) where mnfd 2 G =d 2 V g = mnf1=fg1= 0 max s the mnmum curvature of G ; max beng the maxmum slope of the I/O functons. Furthmore, f 0 0ff 01 [V (t + 1)] 0 f 01 [V (t)]g 1V (t) 0 [1V (t)] 2 : max (10) Combnng (8) (10), n the case where 1 0, we obtan 1E(t) 1t ; j T 0 j + (1+) j 1V (t)1v j (t) (11) where j =0f 6= j and j =1f = j. Hence 1E(t) 0, or the networ s stable, f matrx ft 0 j +[(1+)= max ] j g s postvedefnte. Therefore a suffcent stablty condton for a synchronous TCNN (6) s (1 + ) 1 0; and < 0Tmn 0 (12) max where Tmn 0 s the mnmum egenvalue of matrx Tj. 0 For >1, we expand G at V (t) and obtan, nstead of (9) G [V (t + 1)] 0 G [V (t)] G 0 [V (t)]1v (t) mn d 2 G d 2 [1V (t)] 2 : (13) V We thus have an alternatve stablty condton for the TCNN n synchronous mode, smlar to (12) 2 >1; and > 0Tmn: 0 (14) max The stablty condton for the Euler approxmaton of the CHNN (5) can be derved easly from (7) and (12), wth the usual assumpton that > 0 2 1t ; 1t < ; and T mn 0: (15) 1 0 maxt mn Both the TCNN and the Euler approxmaton of the CHNN are stable f Tmn 0 = 1tT mn > 0, snce the matrx n (11) s automatcally postve-defnte. It s straghtforward to derve that the stablty condtons for both the TCNN and the Euler approxmaton of the CHNN for asynchronous updatng are obtaned from the condtons for synchronous updatng (12), (14), and (15) wth Tmn 0 (or T mn) replaced by mnftg 0 (or T ). Our stablty condtons consst of two parameter regons for the TCNN (12) and (14) and one parameter regon for the Euler approxmaton of the CHNN (15), and are therefore more compact compared to those of Chen and Ahara [2] (three regons for each type of networ). Furthermore, our condtons are less restrctve on the parameters nvolved. For nstance, n the one-neuron TCNN consdered by Chen and Ahara [2], the I/O functon s gven by (3) wth a gan =125. Hence max = =2 =62:5. Wth =0:9, Chen and Ahara s [2] theorems ndcate that the networ s stable f 0T11 0 < 0:0288, whereas our condton (12) gves 0T11 0 < 0:0304 and s closer to the bfurcaton pont 0T11 0 =0:0331. IV. CONCLUSIONS In summary, we have proposed an alternatve approach to chaotc smulated annealng, n whch the tme-step 1t n the Euler approxmaton of the CHNN starts from a large value, where chaos exsts, and reduces to a small value so that the networ stablzes. Ths approach guarantees convergence and mnmzaton of the energy functon, and elmnates the need of choosng other system parameters. It should prove to be a powerful tool for effcently obtan optmal or near-optmal solutons to a varety of optmzaton problems, whch s currently under nvestgaton. We have also generalzed the Chen Ahara convergence theorems for the transently chaotc neural networ and the Euler approxmaton of the CHNN to arbtrarly ncreasng neuronal I/O functons and to less restrctve but more compact forms. Ths should be useful for both hardware mplementatons and software smulatons, as well as further theoretc analyss of the systems.

5 718 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 9, NO. 4, JULY 1998 ACKNOWLEDGMENT Many helpful comments and suggestons from the revewers are gratefully apprecated. REFERENCES [1] L. Chen and K. Ahara, Chaotc smulated annealng by a neuralnetwor model wth transent chaos, Neural Networs, vol. 8, no. 6, pp , [2], Chaos and asymptotcal stablty n dscrete-tme neural networs, Physca, vol. 104, pp , June, [3] F. Fogelman-Soulé, C. Meja, E. Goles, and S. Martnez, Energy functon n neural networs wth contnuous local functons, Complex Syst., vol. 3, pp , [4] J. J. Hopfeld, Neural networs and physcal systems wth emergent collectve computatonal abltes, n Proc. Nat. Academy Sc. USA, Apr. 1982, vol. 79, pp [5], Neurons wth graded response have collectve computatonal propertes le those of two-state neurons, n Proc. Nat. Academy Sc. USA, May 1984, vol. 81, pp [6] J. J. Hopfeld and D. W. Tan, Neural computaton of decsons n optmzaton problems, Bol. Cybern., vol. 52, pp , [7] S. Krpatrc, C. D. Gelatt, Jr., and M. P. Vecch, Optmzaton by smulated annealng, Scence, vol. 220, pp , [8] C. M. Marcus and R. M. Westervelt, Dynamcs of terated-map neural networs, Phys. Rev. A, vol. 40, pp , July [9] H. Nozawa, A neural-networ model as a globally coupled map and applcatons based on chaos, Chaos, vol. 2, no. 3, p , [10] K. Smth and M. Palanswam, Statc and dynamc channel assgnment usng neural networs, IEEE J. Select. Areas Commun., vol. 15, no. 2, pp , [11] K. Smth, M. Palanswam, and M. Krshnamoorthy, Neural technques for combnatoral optmzaton wth applcatons, IEEE Trans. Neural Networs, accepted. [12] M. Verleysen and P. Jespers, An analog VLSI mplementaton of Hopfeld s neural networ, IEEE Mcro, pp , Dec [13] L. Wang, Dscrete-tme convergence theory and updatng rules for neural networs wth energy functons, IEEE Trans. Neural Networs, vol. 8, pp , Mar [14], On the dynamcs of dscrete-tme, contnuous-state Hopfeld neural networs, IEEE Trans. Crcuts Syst. II, to be publshed. [15] G. V. Wlson and G. S. Pawley, On the stablty of the travelng salesman problem algorthm of Hopfeld and Tan, Bol. Cybern., vol. 58, pp. 63, Comments on The Effects of Quantzaton on Multlayer Neural Networs Oh-Jun Kwon and Sung-Yang Bang In ths letter we pont out and correct the errors n the above paper 1 n the dervatons of the followng equatons: 1) n the left column of p y = (10 1 2N ) 2 1 K1 (1) 144 Manuscrpt receved November 9, The authors are wth the Department of Computer Scence and Engneerng, Pohang Unversty of Scence and Technology, Pohang, , Korea. Publsher Item Identfer S (98)04613-X. 1 G. Dündar and K. Rose, IEEE Trans. Neural Networs, vol. 6, pp , Nov ) n the rght column of p y = (11 1 2N ) 2 A A 2 0 tanh A 2 : (2) We thn the dervatons should be as follows. a) Snce the products of the weght and the nput are ndependent, we can rewrte the varance of y 0 as follows: 2 K 01 y =var wx 0 0 =0 K 01 = var[wx 0 0 ]: (3) =0 Also w 0 and x 0 are ndependent and E[w 0 ] = 0 by the assumpton of the unform dstrbuton of the weghts gven by [1]. And, by the assumpton that the nputs are unformly dstrbuted between zero and one, the expectaton of an nput s Therefore E[x 0 ]= 1 2 : (4) var[wx 0 0 ]=E[(wx 0 0 ) 2 ] 0 E 2 [wx 0 0 ] = E[(wx 0 0 ) 2 ] = E[(w) 0 2 ] 1 E[(x 0 ) 2 ] = var[w] 0 1 (var[x]+e 0 2 [x 0 ]) = (10 1 2N ) = (10 1 2N ) 2 : (5) 36 Fnally, substtutng (5) nto (3), we obtan 2 y = (10 1 2N ) 2 1 K1: (6) 36 b) The output of the hdden neurons s between zero and one snce the followng sgmodal functon s used n neurons: f (u) = (1 + e 0u ) 01 : (7) By the assumpton gven by [1] that the outputs of the hdden neurons are unformly dstrbuted, the expectaton of the output of a hdden neuron s E[x 1 l ]= 1 2 : (8) Therefore, as n (a), we can derve the varance of y 1 as follows: 2 K 01 y =var wlx 1 1 l l=0 K 01 = var[wlx 0 1 l ] l=0 K 01 = var[w l ] 1 (var[x 1 l ]+E 2 [x 1 l ]) l=0 = K2 ( N ) A A 2 0 tanh A = (11 1 2N ) A A 0 tanh A 2 : (9)

6 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 9, NO. 4, JULY The new result of a) ndcate that 2 y s four tmes that ndcated by [1] and hence A wll be twce as large. Therefore, accordng to the new results of a) and b), 2 y wll approxmately four tmes as large as that ndcated by [1]. larger for the same reasons. Thus, var(1y 0 )=K =36: (We have used the smpler var() notaton rather than 2 to correct typographcal errors n our manuscrpt caused by confuson of superscrpts.) Thus, the later equatons should be Authors Reply G. Dündar and K. Rose and var(w 1 ) =( N ) 2 =12 var(1 1)=1 2 1=12: We would le to than Kwon and Bang for ther correctons to the equatons n Secton II of our paper. We agree wth ther correctons. They fnd that var(y 0 )=K 1(1 N 0 ) 2 =36 whch s four tmes larger than our expresson. In addton, the expresson whch follows ths n our paper should also be four tmes Manuscrpt receved March 22, G. Dündar was wth the Department of Electrcal and Computer Scence Engneerng, Rensselaer Polytechnc Insttute, Troy, NY USA. He s now wth the Department of Electrcal and Electronc Engneerng, Boğazç Unversty, Bebe 80815, Istanbul, Turey. K. Rose s wth the Department of Electrcal and Computer Scence Engneerng, Rensselaer Polytechnc Insttute, Troy, NY USA. Publsher Item Identfer S (98) Kwon and Bang correct our expresson for var(y 1 ) and we agree wth ther results. However, these changes wll also affect the fnal equaton n Secton II of our paper whch should be var(1y 1 )=K 2 [(( N ) 2 =12)var(1x 1 )+E[(x 1 ) 2 jvar(1 1 )]: The error s relatvely small snce we are replacng var(x 1 ) n the orgnal equaton by E[(x 1 ) 2 ]=var(x 1 )+1=4: Snce smlar errors were made n the dervatons of var(y 0 ) and var(1y 0 ), etc., these have lttle effect on the results of Fg. 3 where we dvde the rms error for the output by the rms value for the output sne wave. Thus these errors tend to cancel and our conclusons are not affected.

Wavelet chaotic neural networks and their application to continuous function optimization

Wavelet chaotic neural networks and their application to continuous function optimization Vol., No.3, 04-09 (009) do:0.436/ns.009.307 Natural Scence Wavelet chaotc neural networks and ther applcaton to contnuous functon optmzaton Ja-Ha Zhang, Yao-Qun Xu College of Electrcal and Automatc Engneerng,

More information

EEE 241: Linear Systems

EEE 241: Linear Systems EEE : Lnear Systems Summary #: Backpropagaton BACKPROPAGATION The perceptron rule as well as the Wdrow Hoff learnng were desgned to tran sngle layer networks. They suffer from the same dsadvantage: they

More information

Multilayer Perceptron (MLP)

Multilayer Perceptron (MLP) Multlayer Perceptron (MLP) Seungjn Cho Department of Computer Scence and Engneerng Pohang Unversty of Scence and Technology 77 Cheongam-ro, Nam-gu, Pohang 37673, Korea seungjn@postech.ac.kr 1 / 20 Outlne

More information

Linear Approximation with Regularization and Moving Least Squares

Linear Approximation with Regularization and Moving Least Squares Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...

More information

Numerical Heat and Mass Transfer

Numerical Heat and Mass Transfer Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and

More information

Feature Selection: Part 1

Feature Selection: Part 1 CSE 546: Machne Learnng Lecture 5 Feature Selecton: Part 1 Instructor: Sham Kakade 1 Regresson n the hgh dmensonal settng How do we learn when the number of features d s greater than the sample sze n?

More information

Chapter Newton s Method

Chapter Newton s Method Chapter 9. Newton s Method After readng ths chapter, you should be able to:. Understand how Newton s method s dfferent from the Golden Secton Search method. Understand how Newton s method works 3. Solve

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

U.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016

U.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016 U.C. Berkeley CS94: Spectral Methods and Expanders Handout 8 Luca Trevsan February 7, 06 Lecture 8: Spectral Algorthms Wrap-up In whch we talk about even more generalzatons of Cheeger s nequaltes, and

More information

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons

More information

Errors for Linear Systems

Errors for Linear Systems Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch

More information

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011 Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected

More information

Lecture Notes on Linear Regression

Lecture Notes on Linear Regression Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume

More information

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,

More information

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:

More information

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng

More information

NUMERICAL DIFFERENTIATION

NUMERICAL DIFFERENTIATION NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the

More information

Generalized Linear Methods

Generalized Linear Methods Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set

More information

9 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations

9 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations Physcs 171/271 - Chapter 9R -Davd Klenfeld - Fall 2005 9 Dervaton of Rate Equatons from Sngle-Cell Conductance (Hodgkn-Huxley-lke) Equatons We consder a network of many neurons, each of whch obeys a set

More information

CHALMERS, GÖTEBORGS UNIVERSITET. SOLUTIONS to RE-EXAM for ARTIFICIAL NEURAL NETWORKS. COURSE CODES: FFR 135, FIM 720 GU, PhD

CHALMERS, GÖTEBORGS UNIVERSITET. SOLUTIONS to RE-EXAM for ARTIFICIAL NEURAL NETWORKS. COURSE CODES: FFR 135, FIM 720 GU, PhD CHALMERS, GÖTEBORGS UNIVERSITET SOLUTIONS to RE-EXAM for ARTIFICIAL NEURAL NETWORKS COURSE CODES: FFR 35, FIM 72 GU, PhD Tme: Place: Teachers: Allowed materal: Not allowed: January 2, 28, at 8 3 2 3 SB

More information

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could

More information

MMA and GCMMA two methods for nonlinear optimization

MMA and GCMMA two methods for nonlinear optimization MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons

More information

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number

More information

APPENDIX A Some Linear Algebra

APPENDIX A Some Linear Algebra APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,

More information

Multilayer Perceptrons and Backpropagation. Perceptrons. Recap: Perceptrons. Informatics 1 CG: Lecture 6. Mirella Lapata

Multilayer Perceptrons and Backpropagation. Perceptrons. Recap: Perceptrons. Informatics 1 CG: Lecture 6. Mirella Lapata Multlayer Perceptrons and Informatcs CG: Lecture 6 Mrella Lapata School of Informatcs Unversty of Ednburgh mlap@nf.ed.ac.uk Readng: Kevn Gurney s Introducton to Neural Networks, Chapters 5 6.5 January,

More information

8 Derivation of Network Rate Equations from Single- Cell Conductance Equations

8 Derivation of Network Rate Equations from Single- Cell Conductance Equations Physcs 178/278 - Davd Klenfeld - Wnter 2015 8 Dervaton of Network Rate Equatons from Sngle- Cell Conductance Equatons We consder a network of many neurons, each of whch obeys a set of conductancebased,

More information

1 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations

1 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations Physcs 171/271 -Davd Klenfeld - Fall 2005 (revsed Wnter 2011) 1 Dervaton of Rate Equatons from Sngle-Cell Conductance (Hodgkn-Huxley-lke) Equatons We consder a network of many neurons, each of whch obeys

More information

Assortment Optimization under MNL

Assortment Optimization under MNL Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.

More information

8 Derivation of Network Rate Equations from Single- Cell Conductance Equations

8 Derivation of Network Rate Equations from Single- Cell Conductance Equations Physcs 178/278 - Davd Klenfeld - Wnter 2019 8 Dervaton of Network Rate Equatons from Sngle- Cell Conductance Equatons Our goal to derve the form of the abstract quanttes n rate equatons, such as synaptc

More information

The Minimum Universal Cost Flow in an Infeasible Flow Network

The Minimum Universal Cost Flow in an Infeasible Flow Network Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran

More information

Lecture 10 Support Vector Machines II

Lecture 10 Support Vector Machines II Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed

More information

A Hybrid Variational Iteration Method for Blasius Equation

A Hybrid Variational Iteration Method for Blasius Equation Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 1932-9466 Vol. 10, Issue 1 (June 2015), pp. 223-229 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) A Hybrd Varatonal Iteraton Method

More information

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin Proceedngs of the 007 Wnter Smulaton Conference S G Henderson, B Bller, M-H Hseh, J Shortle, J D Tew, and R R Barton, eds LOW BIAS INTEGRATED PATH ESTIMATORS James M Calvn Department of Computer Scence

More information

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results. Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson

More information

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0 MODULE 2 Topcs: Lnear ndependence, bass and dmenson We have seen that f n a set of vectors one vector s a lnear combnaton of the remanng vectors n the set then the span of the set s unchanged f that vector

More information

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U) Econ 413 Exam 13 H ANSWERS Settet er nndelt 9 deloppgaver, A,B,C, som alle anbefales å telle lkt for å gøre det ltt lettere å stå. Svar er gtt . Unfortunately, there s a prntng error n the hnt of

More information

Grover s Algorithm + Quantum Zeno Effect + Vaidman

Grover s Algorithm + Quantum Zeno Effect + Vaidman Grover s Algorthm + Quantum Zeno Effect + Vadman CS 294-2 Bomb 10/12/04 Fall 2004 Lecture 11 Grover s algorthm Recall that Grover s algorthm for searchng over a space of sze wors as follows: consder the

More information

The Order Relation and Trace Inequalities for. Hermitian Operators

The Order Relation and Trace Inequalities for. Hermitian Operators Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence

More information

PARTICIPATION FACTOR IN MODAL ANALYSIS OF POWER SYSTEMS STABILITY

PARTICIPATION FACTOR IN MODAL ANALYSIS OF POWER SYSTEMS STABILITY POZNAN UNIVE RSITY OF TE CHNOLOGY ACADE MIC JOURNALS No 86 Electrcal Engneerng 6 Volodymyr KONOVAL* Roman PRYTULA** PARTICIPATION FACTOR IN MODAL ANALYSIS OF POWER SYSTEMS STABILITY Ths paper provdes a

More information

Canonical transformations

Canonical transformations Canoncal transformatons November 23, 2014 Recall that we have defned a symplectc transformaton to be any lnear transformaton M A B leavng the symplectc form nvarant, Ω AB M A CM B DΩ CD Coordnate transformatons,

More information

Case Study of Markov Chains Ray-Knight Compactification

Case Study of Markov Chains Ray-Knight Compactification Internatonal Journal of Contemporary Mathematcal Scences Vol. 9, 24, no. 6, 753-76 HIKAI Ltd, www.m-har.com http://dx.do.org/.2988/cms.24.46 Case Study of Marov Chans ay-knght Compactfcaton HaXa Du and

More information

Neural networks. Nuno Vasconcelos ECE Department, UCSD

Neural networks. Nuno Vasconcelos ECE Department, UCSD Neural networs Nuno Vasconcelos ECE Department, UCSD Classfcaton a classfcaton problem has two types of varables e.g. X - vector of observatons (features) n the world Y - state (class) of the world x X

More information

Markov Chain Monte Carlo Lecture 6

Markov Chain Monte Carlo Lecture 6 where (x 1,..., x N ) X N, N s called the populaton sze, f(x) f (x) for at least one {1, 2,..., N}, and those dfferent from f(x) are called the tral dstrbutons n terms of mportance samplng. Dfferent ways

More information

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009 College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:

More information

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

More information

CHAPTER III Neural Networks as Associative Memory

CHAPTER III Neural Networks as Associative Memory CHAPTER III Neural Networs as Assocatve Memory Introducton One of the prmary functons of the bran s assocatve memory. We assocate the faces wth names, letters wth sounds, or we can recognze the people

More information

Research Article Green s Theorem for Sign Data

Research Article Green s Theorem for Sign Data Internatonal Scholarly Research Network ISRN Appled Mathematcs Volume 2012, Artcle ID 539359, 10 pages do:10.5402/2012/539359 Research Artcle Green s Theorem for Sgn Data Lous M. Houston The Unversty of

More information

Section 8.3 Polar Form of Complex Numbers

Section 8.3 Polar Form of Complex Numbers 80 Chapter 8 Secton 8 Polar Form of Complex Numbers From prevous classes, you may have encountered magnary numbers the square roots of negatve numbers and, more generally, complex numbers whch are the

More information

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS HCMC Unversty of Pedagogy Thong Nguyen Huu et al. A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS Thong Nguyen Huu and Hao Tran Van Department of mathematcs-nformaton,

More information

Lecture 12: Discrete Laplacian

Lecture 12: Discrete Laplacian Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly

More information

Week 5: Neural Networks

Week 5: Neural Networks Week 5: Neural Networks Instructor: Sergey Levne Neural Networks Summary In the prevous lecture, we saw how we can construct neural networks by extendng logstc regresson. Neural networks consst of multple

More information

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family IOSR Journal of Mathematcs IOSR-JM) ISSN: 2278-5728. Volume 3, Issue 3 Sep-Oct. 202), PP 44-48 www.osrjournals.org Usng T.O.M to Estmate Parameter of dstrbutons that have not Sngle Exponental Famly Jubran

More information

Linear Feature Engineering 11

Linear Feature Engineering 11 Lnear Feature Engneerng 11 2 Least-Squares 2.1 Smple least-squares Consder the followng dataset. We have a bunch of nputs x and correspondng outputs y. The partcular values n ths dataset are x y 0.23 0.19

More information

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14 APPROXIMAE PRICES OF BASKE AND ASIAN OPIONS DUPON OLIVIER Prema 14 Contents Introducton 1 1. Framewor 1 1.1. Baset optons 1.. Asan optons. Computng the prce 3. Lower bound 3.1. Closed formula for the prce

More information

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 31 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 6. Rdge regresson The OLSE s the best lnear unbased

More information

NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS

NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS IJRRAS 8 (3 September 011 www.arpapress.com/volumes/vol8issue3/ijrras_8_3_08.pdf NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS H.O. Bakodah Dept. of Mathematc

More information

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system Transfer Functons Convenent representaton of a lnear, dynamc model. A transfer functon (TF) relates one nput and one output: x t X s y t system Y s The followng termnology s used: x y nput output forcng

More information

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017 U.C. Berkeley CS94: Beyond Worst-Case Analyss Handout 4s Luca Trevsan September 5, 07 Summary of Lecture 4 In whch we ntroduce semdefnte programmng and apply t to Max Cut. Semdefnte Programmng Recall that

More information

A Robust Method for Calculating the Correlation Coefficient

A Robust Method for Calculating the Correlation Coefficient A Robust Method for Calculatng the Correlaton Coeffcent E.B. Nven and C. V. Deutsch Relatonshps between prmary and secondary data are frequently quantfed usng the correlaton coeffcent; however, the tradtonal

More information

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1]

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1] DYNAMIC SHORTEST PATH SEARCH AND SYNCHRONIZED TASK SWITCHING Jay Wagenpfel, Adran Trachte 2 Outlne Shortest Communcaton Path Searchng Bellmann Ford algorthm Algorthm for dynamc case Modfcatons to our algorthm

More information

Foundations of Arithmetic

Foundations of Arithmetic Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an

More information

Design and Optimization of Fuzzy Controller for Inverse Pendulum System Using Genetic Algorithm

Design and Optimization of Fuzzy Controller for Inverse Pendulum System Using Genetic Algorithm Desgn and Optmzaton of Fuzzy Controller for Inverse Pendulum System Usng Genetc Algorthm H. Mehraban A. Ashoor Unversty of Tehran Unversty of Tehran h.mehraban@ece.ut.ac.r a.ashoor@ece.ut.ac.r Abstract:

More information

Kernel Methods and SVMs Extension

Kernel Methods and SVMs Extension Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general

More information

COEFFICIENT DIAGRAM: A NOVEL TOOL IN POLYNOMIAL CONTROLLER DESIGN

COEFFICIENT DIAGRAM: A NOVEL TOOL IN POLYNOMIAL CONTROLLER DESIGN Int. J. Chem. Sc.: (4), 04, 645654 ISSN 097768X www.sadgurupublcatons.com COEFFICIENT DIAGRAM: A NOVEL TOOL IN POLYNOMIAL CONTROLLER DESIGN R. GOVINDARASU a, R. PARTHIBAN a and P. K. BHABA b* a Department

More information

Structure and Drive Paul A. Jensen Copyright July 20, 2003

Structure and Drive Paul A. Jensen Copyright July 20, 2003 Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.

More information

Expected Value and Variance

Expected Value and Variance MATH 38 Expected Value and Varance Dr. Neal, WKU We now shall dscuss how to fnd the average and standard devaton of a random varable X. Expected Value Defnton. The expected value (or average value, or

More information

IV. Performance Optimization

IV. Performance Optimization IV. Performance Optmzaton A. Steepest descent algorthm defnton how to set up bounds on learnng rate mnmzaton n a lne (varyng learnng rate) momentum learnng examples B. Newton s method defnton Gauss-Newton

More information

PHYS 705: Classical Mechanics. Calculus of Variations II

PHYS 705: Classical Mechanics. Calculus of Variations II 1 PHYS 705: Classcal Mechancs Calculus of Varatons II 2 Calculus of Varatons: Generalzaton (no constrant yet) Suppose now that F depends on several dependent varables : We need to fnd such that has a statonary

More information

One-sided finite-difference approximations suitable for use with Richardson extrapolation

One-sided finite-difference approximations suitable for use with Richardson extrapolation Journal of Computatonal Physcs 219 (2006) 13 20 Short note One-sded fnte-dfference approxmatons sutable for use wth Rchardson extrapolaton Kumar Rahul, S.N. Bhattacharyya * Department of Mechancal Engneerng,

More information

Erratum: A Generalized Path Integral Control Approach to Reinforcement Learning

Erratum: A Generalized Path Integral Control Approach to Reinforcement Learning Journal of Machne Learnng Research 00-9 Submtted /0; Publshed 7/ Erratum: A Generalzed Path Integral Control Approach to Renforcement Learnng Evangelos ATheodorou Jonas Buchl Stefan Schaal Department of

More information

Finding Dense Subgraphs in G(n, 1/2)

Finding Dense Subgraphs in G(n, 1/2) Fndng Dense Subgraphs n Gn, 1/ Atsh Das Sarma 1, Amt Deshpande, and Rav Kannan 1 Georga Insttute of Technology,atsh@cc.gatech.edu Mcrosoft Research-Bangalore,amtdesh,annan@mcrosoft.com Abstract. Fndng

More information

A new construction of 3-separable matrices via an improved decoding of Macula s construction

A new construction of 3-separable matrices via an improved decoding of Macula s construction Dscrete Optmzaton 5 008 700 704 Contents lsts avalable at ScenceDrect Dscrete Optmzaton journal homepage: wwwelsevercom/locate/dsopt A new constructon of 3-separable matrces va an mproved decodng of Macula

More information

Supplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso

Supplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso Supplement: Proofs and Techncal Detals for The Soluton Path of the Generalzed Lasso Ryan J. Tbshran Jonathan Taylor In ths document we gve supplementary detals to the paper The Soluton Path of the Generalzed

More information

Singular Value Decomposition: Theory and Applications

Singular Value Decomposition: Theory and Applications Sngular Value Decomposton: Theory and Applcatons Danel Khashab Sprng 2015 Last Update: March 2, 2015 1 Introducton A = UDV where columns of U and V are orthonormal and matrx D s dagonal wth postve real

More information

Global Sensitivity. Tuesday 20 th February, 2018

Global Sensitivity. Tuesday 20 th February, 2018 Global Senstvty Tuesday 2 th February, 28 ) Local Senstvty Most senstvty analyses [] are based on local estmates of senstvty, typcally by expandng the response n a Taylor seres about some specfc values

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there

More information

The lower and upper bounds on Perron root of nonnegative irreducible matrices

The lower and upper bounds on Perron root of nonnegative irreducible matrices Journal of Computatonal Appled Mathematcs 217 (2008) 259 267 wwwelsevercom/locate/cam The lower upper bounds on Perron root of nonnegatve rreducble matrces Guang-Xn Huang a,, Feng Yn b,keguo a a College

More information

Deriving the X-Z Identity from Auxiliary Space Method

Deriving the X-Z Identity from Auxiliary Space Method Dervng the X-Z Identty from Auxlary Space Method Long Chen Department of Mathematcs, Unversty of Calforna at Irvne, Irvne, CA 92697 chenlong@math.uc.edu 1 Iteratve Methods In ths paper we dscuss teratve

More information

P R. Lecture 4. Theory and Applications of Pattern Recognition. Dept. of Electrical and Computer Engineering /

P R. Lecture 4. Theory and Applications of Pattern Recognition. Dept. of Electrical and Computer Engineering / Theory and Applcatons of Pattern Recognton 003, Rob Polkar, Rowan Unversty, Glassboro, NJ Lecture 4 Bayes Classfcaton Rule Dept. of Electrcal and Computer Engneerng 0909.40.0 / 0909.504.04 Theory & Applcatons

More information

Multigradient for Neural Networks for Equalizers 1

Multigradient for Neural Networks for Equalizers 1 Multgradent for Neural Netorks for Equalzers 1 Chulhee ee, Jnook Go and Heeyoung Km Department of Electrcal and Electronc Engneerng Yonse Unversty 134 Shnchon-Dong, Seodaemun-Ku, Seoul 1-749, Korea ABSTRACT

More information

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud Resource Allocaton wth a Budget Constrant for Computng Independent Tasks n the Cloud Wemng Sh and Bo Hong School of Electrcal and Computer Engneerng Georga Insttute of Technology, USA 2nd IEEE Internatonal

More information

Classification as a Regression Problem

Classification as a Regression Problem Target varable y C C, C,, ; Classfcaton as a Regresson Problem { }, 3 L C K To treat classfcaton as a regresson problem we should transform the target y nto numercal values; The choce of numercal class

More information

Estimating the Fundamental Matrix by Transforming Image Points in Projective Space 1

Estimating the Fundamental Matrix by Transforming Image Points in Projective Space 1 Estmatng the Fundamental Matrx by Transformng Image Ponts n Projectve Space 1 Zhengyou Zhang and Charles Loop Mcrosoft Research, One Mcrosoft Way, Redmond, WA 98052, USA E-mal: fzhang,cloopg@mcrosoft.com

More information

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,

More information

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity Week3, Chapter 4 Moton n Two Dmensons Lecture Quz A partcle confned to moton along the x axs moves wth constant acceleraton from x =.0 m to x = 8.0 m durng a 1-s tme nterval. The velocty of the partcle

More information

1 Derivation of Point-to-Plane Minimization

1 Derivation of Point-to-Plane Minimization 1 Dervaton of Pont-to-Plane Mnmzaton Consder the Chen-Medon (pont-to-plane) framework for ICP. Assume we have a collecton of ponts (p, q ) wth normals n. We want to determne the optmal rotaton and translaton

More information

Tracking with Kalman Filter

Tracking with Kalman Filter Trackng wth Kalman Flter Scott T. Acton Vrgna Image and Vdeo Analyss (VIVA), Charles L. Brown Department of Electrcal and Computer Engneerng Department of Bomedcal Engneerng Unversty of Vrgna, Charlottesvlle,

More information

Convexity preserving interpolation by splines of arbitrary degree

Convexity preserving interpolation by splines of arbitrary degree Computer Scence Journal of Moldova, vol.18, no.1(52), 2010 Convexty preservng nterpolaton by splnes of arbtrary degree Igor Verlan Abstract In the present paper an algorthm of C 2 nterpolaton of dscrete

More information

4DVAR, according to the name, is a four-dimensional variational method.

4DVAR, according to the name, is a four-dimensional variational method. 4D-Varatonal Data Assmlaton (4D-Var) 4DVAR, accordng to the name, s a four-dmensonal varatonal method. 4D-Var s actually a drect generalzaton of 3D-Var to handle observatons that are dstrbuted n tme. The

More information

Uncertainty in measurements of power and energy on power networks

Uncertainty in measurements of power and energy on power networks Uncertanty n measurements of power and energy on power networks E. Manov, N. Kolev Department of Measurement and Instrumentaton, Techncal Unversty Sofa, bul. Klment Ohrdsk No8, bl., 000 Sofa, Bulgara Tel./fax:

More information

Lecture 4: Constant Time SVD Approximation

Lecture 4: Constant Time SVD Approximation Spectral Algorthms and Representatons eb. 17, Mar. 3 and 8, 005 Lecture 4: Constant Tme SVD Approxmaton Lecturer: Santosh Vempala Scrbe: Jangzhuo Chen Ths topc conssts of three lectures 0/17, 03/03, 03/08),

More information

More metrics on cartesian products

More metrics on cartesian products More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of

More information

DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM

DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM Ganj, Z. Z., et al.: Determnaton of Temperature Dstrbuton for S111 DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM by Davood Domr GANJI

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem

More information

Lecture 4: November 17, Part 1 Single Buffer Management

Lecture 4: November 17, Part 1 Single Buffer Management Lecturer: Ad Rosén Algorthms for the anagement of Networs Fall 2003-2004 Lecture 4: November 7, 2003 Scrbe: Guy Grebla Part Sngle Buffer anagement In the prevous lecture we taled about the Combned Input

More information

VQ widely used in coding speech, image, and video

VQ widely used in coding speech, image, and video at Scalar quantzers are specal cases of vector quantzers (VQ): they are constraned to look at one sample at a tme (memoryless) VQ does not have such constrant better RD perfomance expected Source codng

More information

Solving Nonlinear Differential Equations by a Neural Network Method

Solving Nonlinear Differential Equations by a Neural Network Method Solvng Nonlnear Dfferental Equatons by a Neural Network Method Luce P. Aarts and Peter Van der Veer Delft Unversty of Technology, Faculty of Cvlengneerng and Geoscences, Secton of Cvlengneerng Informatcs,

More information

The Expectation-Maximization Algorithm

The Expectation-Maximization Algorithm The Expectaton-Maxmaton Algorthm Charles Elan elan@cs.ucsd.edu November 16, 2007 Ths chapter explans the EM algorthm at multple levels of generalty. Secton 1 gves the standard hgh-level verson of the algorthm.

More information

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming EEL 6266 Power System Operaton and Control Chapter 3 Economc Dspatch Usng Dynamc Programmng Pecewse Lnear Cost Functons Common practce many utltes prefer to represent ther generator cost functons as sngle-

More information

Neural Networks & Learning

Neural Networks & Learning Neural Netorks & Learnng. Introducton The basc prelmnares nvolved n the Artfcal Neural Netorks (ANN) are descrbed n secton. An Artfcal Neural Netorks (ANN) s an nformaton-processng paradgm that nspred

More information