NONNEGATIVE matrix factorization finds its application

Size: px
Start display at page:

Download "NONNEGATIVE matrix factorization finds its application"

Transcription

1 Multilicative Udates for Convolutional NMF Under -Divergence Pedro J. Villasana T., Stanislaw Gorlow, Meber, IEEE and Arvind T. Hariraan arxiv: v2 [cs.lg 5 May 208 Abstract In this letter, we generalize the convolutional NMF by taking the -divergence as the contrast function and resent the correct ultilicative udates for its factors in closed for. The new udates unify the -NMF and the convolutional NMF. We state why alost all of the existing udates are inexact and aroxiative w.r.t. the convolutional data odel. We show that our udates are stable and that their convergence erforance is consistent across the ost coon values of. Index Ters Convolution, nonnegative atrix factorization, ultilicative udate rules, -divergence. I. INTRODUCTION NONNEGATIVE atrix factorization finds its alication in the fields of achine learning and in connection with inverse robles, ostly. It becae iensely oular after Lee and Seung derived ultilicative udate rules that ade the u until then additive stes in the direction of the negative gradient obsolete [. In [2, they gave eirical evidence of their convergence to a stationary oint, using a the squared Euclidean distance, and b the generalized Kullback Leibler divergence as the contrast function. The factorization s origins can be traced back to [3, [4. A convolutional variant of the factorization is introduced in [5 based on the Kullback Leibler divergence. The ain idea is to exloit teoral deendencies in the neighborhood of a oint in the tie-frequency lane. In their original for, the udates result in a biased factorization. To rovide a reedy, ultile coefficient atrices are udated in [6, one for each translation, and the final udate is by taking the average over all coefficient atrices. A nonnegative atrix factor deconvolution in 2D based on the Kullback Leibler divergence is found in [7. Not only do the authors give a derivation of the udate rules, they show a sile way of aking the udate rules ultilicative. It ay be ointed out that the udate rule for the coefficient atrix is different fro the one in [6. The sae guiding rinciles are alied to derive the convolutional factorization based on the squared Euclidean distance in [8. But in the attet to give a foral roof for their udate rules, the authors largely reforulate a biased factorization coarable to [5. In [9, nonnegative atrix factorization is generalized to a faily of α-divergences under the constraints of sarsity and soothness, while the unconstrained -divergence is brought into focus in [0. For both cases ultilicative udate rules were given. The roerties of the -divergence are discussed in detail in [0, [. The cobined α--divergence together with the corresonding ultilicative udates can be found in [2. In this letter, we rovide ultilicative udate rules for the factorial deconvolution under the -divergence. Furtherore, we argue that the udates in [5, [6 and in [7 are eirical and/or inexact w.r.t. the convolutional data odel. According to our siulation, the udates in [8 do not signify any extra iroveent over [6 desite the additional load. Finally, we show that the exact udates are stable and that their behavior is consistent for {0,, 2}. II. NONNEGATIVE MATRIX FACTORIZATION The nonnegative atrix factorization NMF is an ubrella ter for a low-rank atrix aroxiation of the for V W H with V R K N, W R K I, and H RI N, where I is the redeterined rank of the factorization. The letters above hel distinguish between visible v and hidden variables h that are ut in relation through weights w. The factorization is usually forulated as a convex iniization roble with a dedicated cost function C according to with iniize W, H CW, H subject to w ki, h in 0 2 CW, H = LV, W H, 3 where L is a loss function that assesses the error between V and the factorization W H. A. -Divergence The loss in 3 can be exressed by eans of a contrast or distance function between the eleents of V and W H. Due to its robustness with resect to outliers for certain values of the inut araeter R, we resort to the -divergence [3 as a subclass of the Bregan divergence [, [4, which for the oints and q is given by [ + q q, 0,, d, q = log + q, =, 4 q q log, = 0. q Accordingly, the -divergence for atrices V and W H can be defined entrywise, as D V W H = def K N k= n= d v, i w ki h in, 5 which can further be viewed as the -divergence between two unnoralized arginal robability ass functions with k as the arginal variable. Note that the -divergence has a single global iniu for n v = i,n w ki h in, k, although it is strictly convex only for [, 2 [0, [.

2 2 B. Discrete Convolution As can be seen exlicitly in 5, the weight w ki for the ith hidden variable h in at oint k, n is alied using the scalar roduct. Given that h i evolves with n, we can assue that h i is correlated with its ast and future states. We can take this into account by extending the dot roduct to a convolution in our odel. Postulating causality and letting the weights have finite suort of cardinality M, the convolution writes M un = w ki h i n = def w ki h i,n. 6 =0 The oeration can be converted to a atrix ultilication by lining u the states h in in a truncated Toelitz atrix: h i, h i,2 h i,3 h i,n h i,n 0 h i, h i,2 h i,n 2 h i,n h i,n M h i,n M+ In accordance with, the convolutional NMF CNMF can be forulated as follows to accoodate the structure given in 7, see also [5, [6: V U = M =0 W H, 8 where is a colunwise right-shift oeration siilar to a logical shift in rograing languages that shifts all the coluns of H by ositions to the right, and fills the vacant ositions with zeros. The oeration is size-reserving. It can be seen that the CNMF has M ties as any weights as the regular NMF, whereas the nuber of hidden states is equal. III. -CNMF Ensuing fro the reliinary considerations in Section II, we adot the forulation of the CNMF fro 8 and derive ultilicative udate rules for gradient descent, while taking the entrywise -divergence fro 5 as the loss function. The result is referred to as the -CNMF. A. Proble Stateent Under the reise that V is factorizable into {W } and H, = 0,,..., M, and given the cost function C{W }, H = D V U, 9 we seek to find the ultilicative equivalents of the iterative udate rules for gradient descent: W t+ = W t κ H t+ = H t µ W t C { W t }, H t, 0a H t C{ W t }, H t, 0b where 0a and 0b alternate at each iteration t 0. The ste sizes κ and µ are allowed to change at every iteration. B. Multilicative Udates Rules Couting the artial derivatives of C w.r.t. W and H, and by choosing aroriate values for κ and µ, the iterative udate rules fro 0 becoe ultilicative in the for W t+ = W t [U t H tt [ V U t 2 H tt, H t+ = H t [ WtT Ũt [ WtT V for = 0,,..., M, with Ũt 2 a, b Ũ t = Wt+ H t, 2 where denotes the Hadaard, i.e. entrywise, roduct, is equivalent to entrywise exonentiation and, resectively, stands for the entrywise inverse. The oerator is the leftshift counterart of the right-shift oerator. The details of the derivation of can be found in the Aendix. C. Relation to Previous Works Several ultilicative udates for the CNMF can be found in the existing literature using different loss functions. In [5, the loss function is stated as LV, U = K N v log v 2 v + u u 3 k= n= and the corresonding udate rules for W and H are [ W t+ = W t [ H tt V U t H tt, 4a [ [ H t+ = H t W tt W t T V Ũt, 4b [ where the -atrix is of size K-by-N with =. At t, for every, H is udated first using W t and subsequently W t+ is couted fro the udated H, or vice versa. In [5 it is entioned and ore exlicitly stated in [6 that to avoid bias in H towards W M it is best to first udate all {W } using H t and to udate H according to H t+ = M M =0 W t+t H t [W t+t [ V Ũt. 5 Coaring 4 and 5 with, it can be noted that both udate rules have the sae factors for =. The resective loss function in this case is the generalized Kullback Leibler divergence D V U and not 3. Moreover, the -atrix in 4b is not aligned with the U-atrix. Given that =, 4 and are identical for M =, i.e. when the NMF is nonconvolutional. For M >, 4 is equal to the udates in [2 for D V W H for different W -atrices where the H-atrix is tie-aligned via. Eqs. 5 and b are also different for M > because unlike b 4b is the udate derived fro a nonconvolutional odel. H in 5 brings out

3 VILLASANA et al.: -CNMF 3 the central tendency of the eleents in H but does not ake the factorization in the original loss convolutional. For all the reasons given above, the udates in 4b and 5 are at best an aroxiation of the udate in b for =. In [7, ultilicative udates are given for a CNMF in 2D tie and frequency with the generalized Kullback Leibler divergence and the squared Euclidean distance as the loss or cost function. In the diension of tie, the udates are very uch the sae as the udates in for = 2. For =, there is the inor difference that the -atrix is not aligned with the U-atrix, just like in in 4b. Other ultilicative udates for a CNMF with the squared Euclidean distance can be found in [8. In essence, they are derived in the exact sae anner as the udates in 4, but for a different loss function, which is D 2 V U. Thus, the udates are equal to the ones in [2 with W = W, U as in 8 and H tie-aligned. For the sae reasons that the udates in 4b and 5 are aroxiative of the udate in b for =, the udates in [8 are aroxiative of the udate in b for = 2. Beyond, 2 is udated ore efficiently as Ũ t = Ũt + W t+ W t H t 6 in between the udates of W and W +. D. Interretation The two udate rules given in are of significant value because, aart fro being exact: They are ultilicative, and thus, they converge fast and are easy to ileent. Eq. a extends the udate rule of the -NMF for W to a set of M weight atrices that are linked through a convolution oeration. It also extends the corresonding udate rule of the existing convolutional NMFs with the squared Euclidean distance or the generalized Kullback Leibler divergence to the faily of -divergences. Eq. b is even ore iortant, as it yields a colete udate of the hidden states at every iteration taking all M weight atrices into account at once. The udate rule in b can be viewed as an equivalent to a descent in the direction of the Reynolds gradient, which is to take the average over the artial derivatives under the grou of tie translations in. The average oerator reduces the gradient sread as a function of W at each iteration t such that the loss function converges to a single oint that is ost likely. The -NMF being referred to here is the heuristic -NMF derived in [9. It was shown in [0 to converge faster than a coutationally equivalent axiization-iniization MM algorith for [, 2 and equally fast in the oosite case. The heuristic udates were roven to converge for [0, 2, which is the interval of ractical value. IV. SIMULATION In this section, we coare our roosed udates with the existing ones in ters of convergence behavior and run tie TABLE I AVERAGE RUN TIME OF THE EXISTING CONVOLUTIONAL UPDATES RELATIVE TO THE PROPOSED UPDATES FOR DIFFERENT BETAS Biased Saragdis Schidt et al. Wang et al. Average = = = for 000 iterations. To that end, we generate 00 distinct V- atrices fro M = 6 χ 2 -distributed W -atrices, 2 w ki = wki 2 χ 2 2 w ki N 0,, 7 = and a uniforly distributed H-atrix, h in U0,. 8 The factorizations are reeated with 0 rando initializations of {W} t0 and H t0 with non-zero entries. The results shown in Fig. thus are couted over ensebles of 000 losses at each iteration. The nuber of visible and hidden variables is K = 000 and I = 0, and the nuber of realizations tie sales is N = 00. The run tie was easured on an Intel Xeon E v3 CPU at 3.5 GHz with 6 GB of RAM. In Table I, the figures reresent the averages over 000 runs ut into relation to the average run tie of the roosed udates for 000 iterations. In Fig. it can be seen that the biased udates are clearly least stable under the divergence for which they were eant in the first lace =. Already in [7, convergence issues with these udates were reorted. For {0, 2} at less than 00 iterations they can converge faster because H and U are udated M ties er iteration, which exlains the significant increase in run tie. Between 00 and 000 iterations, other udates show better erforance. Wang s udates are siilar in erforance to Saragdis average udates in site of the additional interediate udates of U for =, and slightly worse otherwise. As stated above, Schidt s udates are the sae as ours for, and so is their behavior. For =, our udates show the sallest variance overall and yield the the lowest cost below 00 iterations, which is a tyical uer bound in ractice. The longer run tie is due to the shifting of the -atrix. For {0, 2}, the loss distributions of our, Wang s, and Saragdis average udates look like they have the sae ean but a slightly different standard deviation. To test the hyothesis that the costs are statistically equivalent in resect of the ean, we eloy Welch s t-test. The -values suggest that the null hyothesis can be rejected alost surely for = 0, whereas for = 2 in general it cannot. V. CONCLUSION To the best of our owledge, our letter is the only one to rovide a colete and exact derivation of the ultilicative udates for the convolutional NMF. Above, the cost function is generalized to the faily of -divergences. It is shown by

4 4 Cost Cost Welch s t-test = 0 = Villasana et al. Saragdis biased Saragdis average Schidt et al. Wang et al. = Fig.. Siulation results including the ean and the standard deviation of the loss distribution and the -value fro Welch s t-test for the hyothesis that the roosed and the existing udate rules on average have the sae cost and eventually converge to the sae iniu as a function of iteration. siulation that the udates are stable and that their behavior is consistent for {0,, 2}. APPENDIX Let u = i, w ki h i,n and U = [u R K N. Then, for any {, 2,..., K}, q {, 2,..., I}, and r {0,,..., M }: w q r C{W }, H = w q r = u n n Choosing κ fro 0a as leads to the first udate rule w t+ ki = wki t n u n v n u 2 n v n u n hq,n r. 9 w q r κ =, 20 n u n h q,n r n v u t 2 n ut ht i,n ht i,n. 2 Further, for any {, 2,..., I} and q {, 2,..., N}: h q C{W }, H = h q k,n It is straightforward to show that u v u. 22 h q u = w k n q 23 by setting n = q = n q. As a result, lugging in q + for n in 22 and using 23, we finally obtain h q C{W }, H = k, w k Choosing µ fro 0b as µ = u k,q+ v k,q+ u 2 k,q+. 24 h q k, w k u, 25 k,q+ leads to the second udate rule k, wt ki v k,n+ u t 2 h t+ in = ht in k, wt ki ut k,n+ k,n+. 26

5 VILLASANA et al.: -CNMF 5 REFERENCES [ D. D. Lee and H. S. Seung, Learning the arts of objects by nonnegative atrix factorization, Nature, vol. 40, , 999. [2, Algoriths for non-negative atrix factorization, in Advances in Neural Inforation Processing Systes, 200, [3 P. Paatero and U. Taer, Positive atrix factorization: A non-negative factor odel with otial utilization of error estiates of data values, Environetrics, vol. 5, no. 2,. 26, 994. [4 P. Paatero, Least squares forulation of robust non-negative factor analysis, Cheoetrics and Intelligent Laboratory Systes, vol. 37, no., , 997. [5 P. Saragdis, Non-negative atrix factor deconvolution; extraction of ultile sound sources fro onohonic inuts, in Indeendent Coonent Analysis and Blind Signal Searation, 2004, [6, Convolutive seech bases and their alication to suervised seech searation, IEEE Transactions on Audio, Seech, and Language Processing, vol. 5, no.,. 2, [7 M. N. Schidt and M. Møru, Nonnegative atrix factor 2-D deconvolution for blind single channel source searation, in Indeendent Coonent Analysis and Blind Signal Searation, 2006, [8 W. Wang, A. Cichocki, and J. A. Chabers, A ultilicative algorith for convolutive non-negative atrix factorization based on squared Euclidean distance, IEEE Transactions on Signal Processing, vol. 57, no. 7, , [9 A. Cichocki, R. Zdunek, and S. Aari, Csiszár s divergences for nonnegative atrix factorization: Faily of new algoriths, in Indeendent Coonent Analysis and Blind Signal Searation, 2006, [0 C. Févotte and J. Idier, Algoriths for nonnegative atrix factorization with the -divergence, Neural Coutation, vol. 23, no. 9, , 20. [ A. Cichocki and S.-i. Aari, Failies of alha- beta- and gaadivergences: Flexible and robust easures of siilarities, Entroy, vol. 2, no. 6, , 200. [2 A. Cichocki, S. Cruces, and S.-i. Aari, Generalized alha-beta divergences and their alication to robust nonnegative atrix factorization, Entroy, vol. 3, no., , 20. [3 A. Basu, I. R. Harris, N. L. Hjort, and M. C. Jones, Robust and efficient estiation by iniising a density ower divergence, Bioetrika, vol. 85, no. 3, , 998. [4 L. M. Bregan, The relaxation ethod of finding the coon oint of convex sets and its alication to the solution of robles in convex rograing, USSR Coutational Matheatics and Matheatical Physics, vol. 7, no. 3, , 967.

[95/95] APPROACH FOR DESIGN LIMITS ANALYSIS IN VVER. Shishkov L., Tsyganov S. Russian Research Centre Kurchatov Institute Russian Federation, Moscow

[95/95] APPROACH FOR DESIGN LIMITS ANALYSIS IN VVER. Shishkov L., Tsyganov S. Russian Research Centre Kurchatov Institute Russian Federation, Moscow [95/95] APPROACH FOR DESIGN LIMITS ANALYSIS IN VVER Shishkov L., Tsyganov S. Russian Research Centre Kurchatov Institute Russian Federation, Moscow ABSTRACT The aer discusses a well-known condition [95%/95%],

More information

5. Dimensional Analysis. 5.1 Dimensions and units

5. Dimensional Analysis. 5.1 Dimensions and units 5. Diensional Analysis In engineering the alication of fluid echanics in designs ake uch of the use of eirical results fro a lot of exerients. This data is often difficult to resent in a readable for.

More information

arxiv: v4 [math.st] 9 Aug 2017

arxiv: v4 [math.st] 9 Aug 2017 PARALLELIZING SPECTRAL ALGORITHMS FOR KERNEL LEARNING GILLES BLANCHARD AND NICOLE MÜCKE arxiv:161007487v4 [athst] 9 Aug 2017 Abstract We consider a distributed learning aroach in suervised learning for

More information

Parallelizing Spectrally Regularized Kernel Algorithms

Parallelizing Spectrally Regularized Kernel Algorithms Journal of Machine Learning Research 19 (2018) 1-29 Subitted 11/16; Revised 8/18; Published 8/18 Parallelizing Sectrally Regularized Kernel Algoriths Nicole Mücke nicole.uecke@atheatik.uni-stuttgart.de

More information

Minimizing Machinery Vibration Transmission in a Lightweight Building using Topology Optimization

Minimizing Machinery Vibration Transmission in a Lightweight Building using Topology Optimization 1 th World Congress on Structural and Multidiscilinary Otiization May 19-4, 13, Orlando, Florida, USA Miniizing Machinery Vibration ransission in a Lightweight Building using oology Otiization Niels Olhoff,

More information

Handout 6 Solutions to Problems from Homework 2

Handout 6 Solutions to Problems from Homework 2 CS 85/185 Fall 2003 Lower Bounds Handout 6 Solutions to Probles fro Hoewor 2 Ait Charabarti Couter Science Dartouth College Solution to Proble 1 1.2: Let f n stand for A 111 n. To decide the roerty f 3

More information

Lecture 3: October 2, 2017

Lecture 3: October 2, 2017 Inforation and Coding Theory Autun 2017 Lecturer: Madhur Tulsiani Lecture 3: October 2, 2017 1 Shearer s lea and alications In the revious lecture, we saw the following stateent of Shearer s lea. Lea 1.1

More information

Control and Stability of the Time-delay Linear Systems

Control and Stability of the Time-delay Linear Systems ISSN 746-7659, England, UK Journal of Inforation and Couting Science Vol., No. 4, 206,.29-297 Control and Stability of the Tie-delay Linear Systes Negras Tahasbi *, Hojjat Ahsani Tehrani Deartent of Matheatics,

More information

Numerical Method for Obtaining a Predictive Estimator for the Geometric Distribution

Numerical Method for Obtaining a Predictive Estimator for the Geometric Distribution British Journal of Matheatics & Couter Science 19(5): 1-13, 2016; Article no.bjmcs.29941 ISSN: 2231-0851 SCIENCEDOMAIN international www.sciencedoain.org Nuerical Method for Obtaining a Predictive Estiator

More information

Design of Linear-Phase Two-Channel FIR Filter Banks with Rational Sampling Factors

Design of Linear-Phase Two-Channel FIR Filter Banks with Rational Sampling Factors R. Bregović and. Saraäi, Design of linear hase two-channel FIR filter bans with rational saling factors, Proc. 3 rd Int. Sy. on Iage and Signal Processing and Analysis, Roe, Italy, Set. 3,. 749 754. Design

More information

Frequency Domain Analysis of Rattle in Gear Pairs and Clutches. Abstract. 1. Introduction

Frequency Domain Analysis of Rattle in Gear Pairs and Clutches. Abstract. 1. Introduction The 00 International Congress and Exosition on Noise Control Engineering Dearborn, MI, USA. August 9-, 00 Frequency Doain Analysis of Rattle in Gear Pairs and Clutches T. C. Ki and R. Singh Acoustics and

More information

ACCURACY OF THE DISCRETE FOURIER TRANSFORM AND THE FAST FOURIER TRANSFORM

ACCURACY OF THE DISCRETE FOURIER TRANSFORM AND THE FAST FOURIER TRANSFORM SIAM J. SCI. COMPUT. c 1996 Society for Industrial and Alied Matheatics Vol. 17, o. 5,. 1150 1166, Seteber 1996 008 ACCURACY OF THE DISCRETE FOURIER TRASFORM AD THE FAST FOURIER TRASFORM JAMES C. SCHATZMA

More information

Exploiting Matrix Symmetries and Physical Symmetries in Matrix Product States and Tensor Trains

Exploiting Matrix Symmetries and Physical Symmetries in Matrix Product States and Tensor Trains Exloiting Matrix Syetries and Physical Syetries in Matrix Product States and Tensor Trains Thoas K Huckle a and Konrad Waldherr a and Thoas Schulte-Herbrüggen b a Technische Universität München, Boltzannstr

More information

1. (2.5.1) So, the number of moles, n, contained in a sample of any substance is equal N n, (2.5.2)

1. (2.5.1) So, the number of moles, n, contained in a sample of any substance is equal N n, (2.5.2) Lecture.5. Ideal gas law We have already discussed general rinciles of classical therodynaics. Classical therodynaics is a acroscoic science which describes hysical systes by eans of acroscoic variables,

More information

Mistiming Performance Analysis of the Energy Detection Based ToA Estimator for MB-OFDM

Mistiming Performance Analysis of the Energy Detection Based ToA Estimator for MB-OFDM IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS Mistiing Perforance Analysis of the Energy Detection Based ToA Estiator for MB-OFDM Huilin Xu, Liuqing Yang contact author, Y T Jade Morton and Mikel M Miller

More information

Phase field modelling of microstructural evolution using the Cahn-Hilliard equation: A report to accompany CH-muSE

Phase field modelling of microstructural evolution using the Cahn-Hilliard equation: A report to accompany CH-muSE Phase field odelling of icrostructural evolution using the Cahn-Hilliard equation: A reort to accoany CH-uSE 1 The Cahn-Hilliard equation Let us consider a binary alloy of average coosition c 0 occuying

More information

INTERIOR BALLISTIC PRINCIPLE OF HIGH/LOW PRESSURE CHAMBERS IN AUTOMATIC GRENADE LAUNCHERS

INTERIOR BALLISTIC PRINCIPLE OF HIGH/LOW PRESSURE CHAMBERS IN AUTOMATIC GRENADE LAUNCHERS XXXX IB08 19th International Syosiu of Ballistics, 7 11 May 001, Interlaken, Switzerland INTERIOR BALLISTIC PRINCIPLE OF HIGH/LOW PRESSURE CHAMBERS IN AUTOMATIC GRENADE LAUNCHERS S. Jaraaz1, D. Micković1,

More information

Boosting with log-loss

Boosting with log-loss Boosting with log-loss Marco Cusuano-Towner Septeber 2, 202 The proble Suppose we have data exaples {x i, y i ) i =... } for a two-class proble with y i {, }. Let F x) be the predictor function with the

More information

The CIA (consistency in aggregation) approach A new economic approach to elementary indices

The CIA (consistency in aggregation) approach A new economic approach to elementary indices The CIA (consistency in aggregation) aroach A new econoic aroach to eleentary indices Dr Jens ehrhoff*, Head of Section Business Cycle and Structural Econoic Statistics * Jens This ehrhoff, resentation

More information

AN EXPLICIT METHOD FOR NUMERICAL SIMULATION OF WAVE EQUATIONS

AN EXPLICIT METHOD FOR NUMERICAL SIMULATION OF WAVE EQUATIONS The 4 th World Conference on Earthquake Engineering October -7, 8, Beiing, China AN EXPLICIT ETHOD FOR NUERICAL SIULATION OF WAVE EQUATIONS Liu Heng and Liao Zheneng Doctoral Candidate, Det. of Structural

More information

Proceedings of Meetings on Acoustics

Proceedings of Meetings on Acoustics Proceedings of Meetings on Acoustics Volue 19, 2013 htt://acousticalsociety.org/ ICA 2013 Montreal Montreal, Canada 2-7 June 2013 Physical Acoustics Session 1PAb: Acoustics in Microfluidics and for Particle

More information

A STUDY OF UNSUPERVISED CHANGE DETECTION BASED ON TEST STATISTIC AND GAUSSIAN MIXTURE MODEL USING POLSAR SAR DATA

A STUDY OF UNSUPERVISED CHANGE DETECTION BASED ON TEST STATISTIC AND GAUSSIAN MIXTURE MODEL USING POLSAR SAR DATA A STUDY OF UNSUPERVISED CHANGE DETECTION BASED ON TEST STATISTIC AND GAUSSIAN MIXTURE MODEL USING POLSAR SAR DATA Yang Yuxin a, Liu Wensong b* a Middle School Affiliated to Central China Noral University,

More information

Suppress Parameter Cross-talk for Elastic Full-waveform Inversion: Parameterization and Acquisition Geometry

Suppress Parameter Cross-talk for Elastic Full-waveform Inversion: Parameterization and Acquisition Geometry Suress Paraeter Cross-talk for Elastic Full-wavefor Inversion: Paraeterization and Acquisition Geoetry Wenyong Pan and Kris Innanen CREWES Project, Deartent of Geoscience, University of Calgary Suary Full-wavefor

More information

SUPPORTING INFORMATION FOR. Mass Spectrometrically-Detected Statistical Aspects of Ligand Populations in Mixed Monolayer Au 25 L 18 Nanoparticles

SUPPORTING INFORMATION FOR. Mass Spectrometrically-Detected Statistical Aspects of Ligand Populations in Mixed Monolayer Au 25 L 18 Nanoparticles SUPPORTIG IFORMATIO FOR Mass Sectroetrically-Detected Statistical Asects of Lig Poulations in Mixed Monolayer Au 25 L 8 anoarticles Aala Dass,,a Kennedy Holt, Joseh F. Parer, Stehen W. Feldberg, Royce

More information

Computationally Efficient Control System Based on Digital Dynamic Pulse Frequency Modulation for Microprocessor Implementation

Computationally Efficient Control System Based on Digital Dynamic Pulse Frequency Modulation for Microprocessor Implementation IJCSI International Journal of Couter Science Issues, Vol. 0, Issue 3, No 2, May 203 ISSN (Print): 694-084 ISSN (Online): 694-0784 www.ijcsi.org 20 Coutationally Efficient Control Syste Based on Digital

More information

#A62 INTEGERS 16 (2016) REPRESENTATION OF INTEGERS BY TERNARY QUADRATIC FORMS: A GEOMETRIC APPROACH

#A62 INTEGERS 16 (2016) REPRESENTATION OF INTEGERS BY TERNARY QUADRATIC FORMS: A GEOMETRIC APPROACH #A6 INTEGERS 16 (016) REPRESENTATION OF INTEGERS BY TERNARY QUADRATIC FORMS: A GEOMETRIC APPROACH Gabriel Durha Deartent of Matheatics, University of Georgia, Athens, Georgia gjdurha@ugaedu Received: 9/11/15,

More information

Edinburgh Research Explorer

Edinburgh Research Explorer Edinburgh Research Exlorer ALMOST-ORTHOGONALITY IN THE SCHATTEN-VON NEUMANN CLASSES Citation for ublished version: Carbery, A 2009, 'ALMOST-ORTHOGONALITY IN THE SCHATTEN-VON NEUMANN CLASSES' Journal of

More information

COS 424: Interacting with Data. Written Exercises

COS 424: Interacting with Data. Written Exercises COS 424: Interacting with Data Hoework #4 Spring 2007 Regression Due: Wednesday, April 18 Written Exercises See the course website for iportant inforation about collaboration and late policies, as well

More information

Approximation by Piecewise Constants on Convex Partitions

Approximation by Piecewise Constants on Convex Partitions Aroxiation by Piecewise Constants on Convex Partitions Oleg Davydov Noveber 4, 2011 Abstract We show that the saturation order of iecewise constant aroxiation in L nor on convex artitions with N cells

More information

Optimal Adaptive Computations in the Jaffard Algebra and Localized Frames

Optimal Adaptive Computations in the Jaffard Algebra and Localized Frames www.oeaw.ac.at Otial Adative Coutations in the Jaffard Algebra and Localized Fraes M. Fornasier, K. Gröchenig RICAM-Reort 2006-28 www.rica.oeaw.ac.at Otial Adative Coutations in the Jaffard Algebra and

More information

Fundamentals of Astrodynamics and Applications 3 rd Ed

Fundamentals of Astrodynamics and Applications 3 rd Ed Fundaentals of Astrodynaics and Alications 3 rd Ed Errata June 0, 0 This listing is an on-going docuent of corrections and clarifications encountered in the book. I areciate any coents and questions you

More information

The Semantics of Data Flow Diagrams. P.D. Bruza. Th.P. van der Weide. Dept. of Information Systems, University of Nijmegen

The Semantics of Data Flow Diagrams. P.D. Bruza. Th.P. van der Weide. Dept. of Information Systems, University of Nijmegen The Seantics of Data Flow Diagras P.D. Bruza Th.P. van der Weide Det. of Inforation Systes, University of Nijegen Toernooiveld, NL-6525 ED Nijegen, The Netherlands July 26, 1993 Abstract In this article

More information

CALCULATION of CORONA INCEPTION VOLTAGES in N 2 +SF 6 MIXTURES via GENETIC ALGORITHM

CALCULATION of CORONA INCEPTION VOLTAGES in N 2 +SF 6 MIXTURES via GENETIC ALGORITHM CALCULATION of COONA INCPTION VOLTAGS in N +SF 6 MIXTUS via GNTIC ALGOITHM. Onal G. Kourgoz e-ail: onal@elk.itu.edu.tr e-ail: guven@itu.edu..edu.tr Istanbul Technical University, Faculty of lectric and

More information

arxiv: v2 [math.st] 13 Feb 2018

arxiv: v2 [math.st] 13 Feb 2018 A data-deendent weighted LASSO under Poisson noise arxiv:1509.08892v2 [ath.st] 13 Feb 2018 Xin J. Hunt, SAS Institute Inc., Cary, NC USA Patricia Reynaud-Bouret University of Côte d Azur, CNRS, LJAD, Nice,

More information

An Investigation into the Effects of Roll Gyradius on Experimental Testing and Numerical Simulation: Troubleshooting Emergent Issues

An Investigation into the Effects of Roll Gyradius on Experimental Testing and Numerical Simulation: Troubleshooting Emergent Issues An Investigation into the Effects of Roll Gyradius on Exeriental esting and Nuerical Siulation: roubleshooting Eergent Issues Edward Dawson Maritie Division Defence Science and echnology Organisation DSO-N-140

More information

EXACT BOUNDS FOR JUDICIOUS PARTITIONS OF GRAPHS

EXACT BOUNDS FOR JUDICIOUS PARTITIONS OF GRAPHS EXACT BOUNDS FOR JUDICIOUS PARTITIONS OF GRAPHS B. BOLLOBÁS1,3 AND A.D. SCOTT,3 Abstract. Edwards showed that every grah of size 1 has a biartite subgrah of size at least / + /8 + 1/64 1/8. We show that

More information

Intelligent Systems: Reasoning and Recognition. Artificial Neural Networks

Intelligent Systems: Reasoning and Recognition. Artificial Neural Networks Intelligent Systes: Reasoning and Recognition Jaes L. Crowley MOSIG M1 Winter Seester 2018 Lesson 7 1 March 2018 Outline Artificial Neural Networks Notation...2 Introduction...3 Key Equations... 3 Artificial

More information

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization Recent Researches in Coputer Science Support Vector Machine Classification of Uncertain and Ibalanced data using Robust Optiization RAGHAV PAT, THEODORE B. TRAFALIS, KASH BARKER School of Industrial Engineering

More information

A Constraint View of IBD Graphs

A Constraint View of IBD Graphs A Constraint View of IBD Grahs Rina Dechter, Dan Geiger and Elizabeth Thoson Donald Bren School of Inforation and Couter Science University of California, Irvine, CA 92697 1 Introduction The reort rovides

More information

CHAPTER 2 THERMODYNAMICS

CHAPTER 2 THERMODYNAMICS CHAPER 2 HERMODYNAMICS 2.1 INRODUCION herodynaics is the study of the behavior of systes of atter under the action of external fields such as teerature and ressure. It is used in articular to describe

More information

Polynomials and Cryptography. Michele Elia Politecnico di Torino

Polynomials and Cryptography. Michele Elia Politecnico di Torino Polynoials and Crytograhy Michele Elia Politecnico di Torino Trento 0 arzo 0 Preable Polynoials have always occuied a roinent osition in atheatics. In recent tie their use has becoe unavoidable in crytograhy.

More information

On spinors and their transformation

On spinors and their transformation AMERICAN JOURNAL OF SCIENTIFIC AND INDUSTRIAL RESEARCH, Science Huβ, htt:www.scihub.orgajsir ISSN: 5-69X On sinors and their transforation Anaitra Palit AuthorTeacher, P5 Motijheel Avenue, Flat C,Kolkata

More information

Uniform Deviation Bounds for k-means Clustering

Uniform Deviation Bounds for k-means Clustering Unifor Deviation Bounds for k-means Clustering Olivier Bache Mario Lucic S Haed Hassani Andreas Krause Abstract Unifor deviation bounds liit the difference between a odel s exected loss and its loss on

More information

Modi ed Local Whittle Estimator for Long Memory Processes in the Presence of Low Frequency (and Other) Contaminations

Modi ed Local Whittle Estimator for Long Memory Processes in the Presence of Low Frequency (and Other) Contaminations Modi ed Local Whittle Estiator for Long Meory Processes in the Presence of Low Frequency (and Other Containations Jie Hou y Boston University Pierre Perron z Boston University March 5, 203; Revised: January

More information

J.B. LASSERRE AND E.S. ZERON

J.B. LASSERRE AND E.S. ZERON L -NORMS, LOG-BARRIERS AND CRAMER TRANSFORM IN OPTIMIZATION J.B. LASSERRE AND E.S. ZERON Abstract. We show that the Lalace aroxiation of a sureu by L -nors has interesting consequences in otiization. For

More information

On Maximizing the Convergence Rate for Linear Systems With Input Saturation

On Maximizing the Convergence Rate for Linear Systems With Input Saturation IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 48, NO. 7, JULY 2003 1249 On Maxiizing the Convergence Rate for Linear Systes With Inut Saturation Tingshu Hu, Zongli Lin, Yacov Shaash Abstract In this note,

More information

Nonlinear Active Noise Control Using NARX Model Structure Selection

Nonlinear Active Noise Control Using NARX Model Structure Selection 2009 Aerican Control Conference Hyatt Regency Riverfront, St. Louis, MO, USA June 10-12, 2009 FrC13.6 Nonlinear Active Noise Control Using NARX Model Structure Selection R. Naoli and L. Piroddi, Meber,

More information

Anomalous heat capacity for nematic MBBA near clearing point

Anomalous heat capacity for nematic MBBA near clearing point Journal of Physics: Conference Series Anoalous heat caacity for neatic MA near clearing oint To cite this article: D A Lukashenko and M Khasanov J. Phys.: Conf. Ser. 394 View the article online for udates

More information

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis E0 370 tatistical Learning Theory Lecture 6 (Aug 30, 20) Margin Analysis Lecturer: hivani Agarwal cribe: Narasihan R Introduction In the last few lectures we have seen how to obtain high confidence bounds

More information

Pattern Recognition and Machine Learning. Artificial Neural networks

Pattern Recognition and Machine Learning. Artificial Neural networks Pattern Recognition and Machine Learning Jaes L. Crowley ENSIMAG 3 - MMIS Fall Seester 2017 Lessons 7 20 Dec 2017 Outline Artificial Neural networks Notation...2 Introduction...3 Key Equations... 3 Artificial

More information

Ch 12: Variations on Backpropagation

Ch 12: Variations on Backpropagation Ch 2: Variations on Backpropagation The basic backpropagation algorith is too slow for ost practical applications. It ay take days or weeks of coputer tie. We deonstrate why the backpropagation algorith

More information

VACUUM chambers have wide applications for a variety of

VACUUM chambers have wide applications for a variety of JOURNAL OF THERMOPHYSICS AND HEAT TRANSFER Vol. 2, No., January March 27 Free Molecular Flows Between Two Plates Equied with Pus Chunei Cai ZONA Technology, Inc., Scottsdale, Arizona 85258 Iain D. Boyd

More information

Uniform Deviation Bounds for k-means Clustering

Uniform Deviation Bounds for k-means Clustering Unifor Deviation Bounds for k-means Clustering Olivier Bache Mario Lucic S. Haed Hassani Andreas Krause Abstract Unifor deviation bounds liit the difference between a odel s exected loss and its loss on

More information

A Simple Regression Problem

A Simple Regression Problem A Siple Regression Proble R. M. Castro March 23, 2 In this brief note a siple regression proble will be introduced, illustrating clearly the bias-variance tradeoff. Let Y i f(x i ) + W i, i,..., n, where

More information

Analysis of low rank matrix recovery via Mendelson s small ball method

Analysis of low rank matrix recovery via Mendelson s small ball method Analysis of low rank atrix recovery via Mendelson s sall ball ethod Maryia Kabanava Chair for Matheatics C (Analysis) ontdriesch 0 kabanava@athc.rwth-aachen.de Holger Rauhut Chair for Matheatics C (Analysis)

More information

CALIFORNIA INSTITUTE OF TECHNOLOGY

CALIFORNIA INSTITUTE OF TECHNOLOGY CALIFORNIA INSIUE OF ECHNOLOGY Control and Dynaical Systes Course Project CDS 270 Instructor: Eugene Lavretsky, eugene.lavretsky@boeing.co Sring 2007 Project Outline: his roject consists of two flight

More information

Chordal Sparsity, Decomposing SDPs and the Lyapunov Equation

Chordal Sparsity, Decomposing SDPs and the Lyapunov Equation Chordal Sarsity, Decoosing SDPs and the Lyaunov Equation Richard P. Mason and Antonis Paachristodoulou Abstract Analysis questions in control theory are often forulated as Linear Matrix Inequalities and

More information

DISCRETE DUALITY FINITE VOLUME SCHEMES FOR LERAY-LIONS TYPE ELLIPTIC PROBLEMS ON GENERAL 2D MESHES

DISCRETE DUALITY FINITE VOLUME SCHEMES FOR LERAY-LIONS TYPE ELLIPTIC PROBLEMS ON GENERAL 2D MESHES ISCRETE UALITY FINITE VOLUME SCHEMES FOR LERAY-LIONS TYPE ELLIPTIC PROBLEMS ON GENERAL 2 MESHES BORIS ANREIANOV, FRANCK BOYER AN FLORENCE HUBERT Abstract. iscrete duality finite volue schees on general

More information

Input-Output (I/O) Stability. -Stability of a System

Input-Output (I/O) Stability. -Stability of a System Inut-Outut (I/O) Stability -Stability of a Syste Outline: Introduction White Boxes and Black Boxes Inut-Outut Descrition Foralization of the Inut-Outut View Signals and Signal Saces he Notions of Gain

More information

Bootstrapping Dependent Data

Bootstrapping Dependent Data Bootstrapping Dependent Data One of the key issues confronting bootstrap resapling approxiations is how to deal with dependent data. Consider a sequence fx t g n t= of dependent rando variables. Clearly

More information

A Sequential Tracking Filter without Requirement of Measurement Decorrelation

A Sequential Tracking Filter without Requirement of Measurement Decorrelation A Sequential racing Filter without Requient of Measuent Decorlation Gongjian Zhou, Changjun Yu, aifan Quan Deartent of Electronic Engineering Harbin Institute of echnology Harbin, China zhougj@hit.edu.cn

More information

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices CS71 Randoness & Coputation Spring 018 Instructor: Alistair Sinclair Lecture 13: February 7 Disclaier: These notes have not been subjected to the usual scrutiny accorded to foral publications. They ay

More information

Chapter 8 Markov Chains and Some Applications ( 馬哥夫鏈 )

Chapter 8 Markov Chains and Some Applications ( 馬哥夫鏈 ) Chater 8 arkov Chains and oe Alications ( 馬哥夫鏈 Consider a sequence of rando variables,,, and suose that the set of ossible values of these rando variables is {,,,, }, which is called the state sace. It

More information

Pattern Recognition and Machine Learning. Artificial Neural networks

Pattern Recognition and Machine Learning. Artificial Neural networks Pattern Recognition and Machine Learning Jaes L. Crowley ENSIMAG 3 - MMIS Fall Seester 2016/2017 Lessons 9 11 Jan 2017 Outline Artificial Neural networks Notation...2 Convolutional Neural Networks...3

More information

The Generalized Integer Gamma DistributionA Basis for Distributions in Multivariate Statistics

The Generalized Integer Gamma DistributionA Basis for Distributions in Multivariate Statistics Journal of Multivariate Analysis 64, 8610 (1998) Article No. MV971710 The Generalized Inteer Gaa DistributionA Basis for Distributions in Multivariate Statistics Carlos A. Coelho Universidade Te cnica

More information

Feature Extraction Techniques

Feature Extraction Techniques Feature Extraction Techniques Unsupervised Learning II Feature Extraction Unsupervised ethods can also be used to find features which can be useful for categorization. There are unsupervised ethods that

More information

FRESNEL FORMULAE FOR SCATTERING OPERATORS

FRESNEL FORMULAE FOR SCATTERING OPERATORS elecounications and Radio Engineering, 70(9):749-758 (011) MAHEMAICAL MEHODS IN ELECROMAGNEIC HEORY FRESNEL FORMULAE FOR SCAERING OPERAORS I.V. Petrusenko & Yu.K. Sirenko A. Usikov Institute of Radio Physics

More information

arxiv: v1 [physics.soc-ph] 5 Jun 2011

arxiv: v1 [physics.soc-ph] 5 Jun 2011 el draft Phase transition in the Sznajd odel with indeendence. K. Sznajd-Weron, M. Tabiszewski, A. M. Tianaro 2 arxiv:6.934v [hysics.soc-h] 5 Jun 2 Institute of Theoretical Physics, University of Wroc

More information

Stochastic Subgradient Methods

Stochastic Subgradient Methods Stochastic Subgradient Methods Lingjie Weng Yutian Chen Bren School of Inforation and Coputer Science University of California, Irvine {wengl, yutianc}@ics.uci.edu Abstract Stochastic subgradient ethods

More information

Binomial and Poisson Probability Distributions

Binomial and Poisson Probability Distributions Binoial and Poisson Probability Distributions There are a few discrete robability distributions that cro u any ties in hysics alications, e.g. QM, SM. Here we consider TWO iortant and related cases, the

More information

Non-Parametric Non-Line-of-Sight Identification 1

Non-Parametric Non-Line-of-Sight Identification 1 Non-Paraetric Non-Line-of-Sight Identification Sinan Gezici, Hisashi Kobayashi and H. Vincent Poor Departent of Electrical Engineering School of Engineering and Applied Science Princeton University, Princeton,

More information

A Subspace Iteration for Calculating a Cluster of Exterior Eigenvalues

A Subspace Iteration for Calculating a Cluster of Exterior Eigenvalues Advances in Linear Algebra & Matrix heory 05 5 76-89 Published Online Seteber 05 in SciRes htt://wwwscirorg/ournal/alat htt://dxdoiorg/0436/alat0553008 A Subsace Iteration for Calculating a Cluster of

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Notes for EE227C (Spring 2018): Convex Optiization and Approxiation Instructor: Moritz Hardt Eail: hardt+ee227c@berkeley.edu Graduate Instructor: Max Sichowitz Eail: sichow+ee227c@berkeley.edu October

More information

Some simple continued fraction expansions for an in nite product Part 1. Peter Bala, January ax 4n+3 1 ax 4n+1. (a; x) =

Some simple continued fraction expansions for an in nite product Part 1. Peter Bala, January ax 4n+3 1 ax 4n+1. (a; x) = Soe sile continued fraction exansions for an in nite roduct Part. Introduction The in nite roduct Peter Bala, January 3 (a; x) = Y ax 4n+3 ax 4n+ converges for arbitrary colex a rovided jxj

More information

An Iterative Substructuring Approach to the Calculation of. Eigensolution and Eigensensitivity

An Iterative Substructuring Approach to the Calculation of. Eigensolution and Eigensensitivity his is the Pre-Published Version. An Iterative Substructuring Aroach to the Calculation of Eigensolution and Eigensensitivity Shun Weng PhD student, Deartent of Civil and Structural Engineering, he Hong

More information

A GENERAL THEORY OF PARTICLE FILTERS IN HIDDEN MARKOV MODELS AND SOME APPLICATIONS. By Hock Peng Chan National University of Singapore and

A GENERAL THEORY OF PARTICLE FILTERS IN HIDDEN MARKOV MODELS AND SOME APPLICATIONS. By Hock Peng Chan National University of Singapore and Subitted to the Annals of Statistics A GENERAL THEORY OF PARTICLE FILTERS IN HIDDEN MARKOV MODELS AND SOME APPLICATIONS By Hock Peng Chan National University of Singaore and By Tze Leung Lai Stanford University

More information

Optimization of Dynamic Reactive Power Sources Using Mesh Adaptive Direct Search

Optimization of Dynamic Reactive Power Sources Using Mesh Adaptive Direct Search Acceted by IET Generation, Transission & Distribution on 6/2/207 Otiization of Dynaic Reactive Power Sources Using Mesh Adative Direct Search Weihong Huang, Kai Sun,*, Junjian Qi 2, Jiaxin Ning 3 Electrical

More information

Pattern Recognition and Machine Learning. Artificial Neural networks

Pattern Recognition and Machine Learning. Artificial Neural networks Pattern Recognition and Machine Learning Jaes L. Crowley ENSIMAG 3 - MMIS Fall Seester 2016 Lessons 7 14 Dec 2016 Outline Artificial Neural networks Notation...2 1. Introduction...3... 3 The Artificial

More information

Feedback-error control

Feedback-error control Chater 4 Feedback-error control 4.1 Introduction This chater exlains the feedback-error (FBE) control scheme originally described by Kawato [, 87, 8]. FBE is a widely used neural network based controller

More information

Distributed Subgradient Methods for Multi-agent Optimization

Distributed Subgradient Methods for Multi-agent Optimization 1 Distributed Subgradient Methods for Multi-agent Optiization Angelia Nedić and Asuan Ozdaglar October 29, 2007 Abstract We study a distributed coputation odel for optiizing a su of convex objective functions

More information

4 A Survey of Congruent Results 12

4 A Survey of Congruent Results 12 4 A urvey of Congruent Results 1 ECTION 4.5 Perfect Nubers and the iga Function By the end of this section you will be able to test whether a given Mersenne nuber is rie understand what is eant be a erfect

More information

The Number of Information Bits Related to the Minimum Quantum and Gravitational Masses in a Vacuum Dominated Universe

The Number of Information Bits Related to the Minimum Quantum and Gravitational Masses in a Vacuum Dominated Universe Wilfrid Laurier University Scholars Coons @ Laurier Physics and Couter Science Faculty Publications Physics and Couter Science 01 The uber of Inforation Bits Related to the Miniu Quantu and Gravitational

More information

3 Thermodynamics and Statistical mechanics

3 Thermodynamics and Statistical mechanics Therodynaics and Statistical echanics. Syste and environent The syste is soe ortion of atter that we searate using real walls or only in our ine, fro the other art of the universe. Everything outside the

More information

The Schrödinger Equation and the Scale Principle

The Schrödinger Equation and the Scale Principle Te Scrödinger Equation and te Scale Princile RODOLFO A. FRINO Jul 014 Electronics Engineer Degree fro te National Universit of Mar del Plata - Argentina rodolfo_frino@aoo.co.ar Earlier tis ear (Ma) I wrote

More information

Intelligent Systems: Reasoning and Recognition. Perceptrons and Support Vector Machines

Intelligent Systems: Reasoning and Recognition. Perceptrons and Support Vector Machines Intelligent Systes: Reasoning and Recognition Jaes L. Crowley osig 1 Winter Seester 2018 Lesson 6 27 February 2018 Outline Perceptrons and Support Vector achines Notation...2 Linear odels...3 Lines, Planes

More information

Security Transaction Differential Equation

Security Transaction Differential Equation Security Transaction Differential Equation A Transaction Volue/Price Probability Wave Model Shi, Leilei This draft: June 1, 4 Abstract Financial arket is a tyical colex syste because it is an oen trading

More information

PROJECTIONS IN VECTOR SPACES OVER FINITE FIELDS

PROJECTIONS IN VECTOR SPACES OVER FINITE FIELDS Annales Acadeiæ Scientiaru Fennicæ Matheatica Voluen 43, 2018, 171 185 PROJECTIONS IN VECTOR SPACES OVER FINITE FIELDS Changhao Chen The University of New South Wales, School of Matheatics and Statistics

More information

Fixed-to-Variable Length Distribution Matching

Fixed-to-Variable Length Distribution Matching Fixed-to-Variable Length Distribution Matching Rana Ali Ajad and Georg Böcherer Institute for Counications Engineering Technische Universität München, Gerany Eail: raa2463@gail.co,georg.boecherer@tu.de

More information

Block designs and statistics

Block designs and statistics Bloc designs and statistics Notes for Math 447 May 3, 2011 The ain paraeters of a bloc design are nuber of varieties v, bloc size, nuber of blocs b. A design is built on a set of v eleents. Each eleent

More information

A method to determine relative stroke detection efficiencies from multiplicity distributions

A method to determine relative stroke detection efficiencies from multiplicity distributions A ethod to deterine relative stroke detection eiciencies ro ultiplicity distributions Schulz W. and Cuins K. 2. Austrian Lightning Detection and Inoration Syste (ALDIS), Kahlenberger Str.2A, 90 Vienna,

More information

The Construction of Orthonormal Wavelets Using Symbolic Methods and a Matrix Analytical Approach for Wavelets on the Interval

The Construction of Orthonormal Wavelets Using Symbolic Methods and a Matrix Analytical Approach for Wavelets on the Interval The Construction of Orthonoral Wavelets Using Sybolic Methods and a Matrix Analytical Aroach for Wavelets on the Interval Frédéric Chyzak, Peter Paule, Otar Scherzer, Arin Schoisswohl, and Burkhard Zierann

More information

Analytical Analysis and Feedback Linearization Tracking Control of the General Takagi-Sugeno Fuzzy Dynamic Systems

Analytical Analysis and Feedback Linearization Tracking Control of the General Takagi-Sugeno Fuzzy Dynamic Systems 290 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS PART C: APPLICATIONS AND REVIEWS, VOL. 29, NO., FEBRUARY 999 In order to verify the erforance of the roosed ethod, exerients with the NIST nueral

More information

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion Suppleentary Material for Fast and Provable Algoriths for Spectrally Sparse Signal Reconstruction via Low-Ran Hanel Matrix Copletion Jian-Feng Cai Tianing Wang Ke Wei March 1, 017 Abstract We establish

More information

New Set of Rotationally Legendre Moment Invariants

New Set of Rotationally Legendre Moment Invariants New Set of Rotationally Legendre Moent Invariants Khalid M. Hosny Abstract Orthogonal Legendre oents are used in several attern recognition and iage rocessing alications. Translation and scale Legendre

More information

The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Parameters

The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Parameters journal of ultivariate analysis 58, 96106 (1996) article no. 0041 The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Paraeters H. S. Steyn

More information

Receiver and Performance Analysis

Receiver and Performance Analysis On the Distribution of SINR for the MMSE MIMO Receiver and Perforance Analysis Ping Li, Debashis Paul, Ravi Narasihan, Meber, IEEE, and John Cioffi, Fellow, IEEE Abstract This aer studies the statistical

More information

Multi-Dimensional Hegselmann-Krause Dynamics

Multi-Dimensional Hegselmann-Krause Dynamics Multi-Diensional Hegselann-Krause Dynaics A. Nedić Industrial and Enterprise Systes Engineering Dept. University of Illinois Urbana, IL 680 angelia@illinois.edu B. Touri Coordinated Science Laboratory

More information

Dueling Bandits with Weak Regret

Dueling Bandits with Weak Regret Dueling Bandits with Weak Regret Bangrui Chen 1 Peter I Frazier 1 Our contribution is a new algorith, Winner Stays WS), with variants designed for the weak WS-W) and strong regret WS-S) settings We rove

More information

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon Model Fitting CURM Background Material, Fall 014 Dr. Doreen De Leon 1 Introduction Given a set of data points, we often want to fit a selected odel or type to the data (e.g., we suspect an exponential

More information

Kernel Methods and Support Vector Machines

Kernel Methods and Support Vector Machines Intelligent Systes: Reasoning and Recognition Jaes L. Crowley ENSIAG 2 / osig 1 Second Seester 2012/2013 Lesson 20 2 ay 2013 Kernel ethods and Support Vector achines Contents Kernel Functions...2 Quadratic

More information