Multi-Scale Analysis of Long Range Dependent Traffic for Anomaly Detection in Wireless Sensor Networks
|
|
- Gregory Sims
- 5 years ago
- Views:
Transcription
1 20 50th IEEE Conference on Decision and Contro and European Contro Conference (CDC-ECC) Orando FL USA December Muti-Scae Anaysis of Long Range Dependent Traffic for Anomay Detection in Wireess Sensor Networs Shanshan Zheng and John S. Baras Abstract Anomay detection is important for the correct functioning of wireess sensor networs. Recent studies have shown that node mobiity aong with spatia correation of the monitored phenomenon in sensor networs can ead to observation data that have ong range dependency which coud significanty increase the difficuty of anomay detection. In this paper we deveop an anomay detection scheme based on mutiscae anaysis of the ong range dependent traffic to address this chaenge. In this proposed detection scheme discrete waveet transform is used to approximatey de-correate the traffic data and capture data characteristics in different time scaes. The remaining dependencies are then captured by a muti-eve hidden Marov mode in the waveet domain. To estimate the mode parameters we propose an onine discounting Expectation Maximization (EM) agorithm which aso tracs variations of the estimated modes over time. Networ anomaies are detected as abrupt changes in the traced mode variation scores. We evauate our detection scheme numericay using typica ong range dependent time series. I. INTRODUCTION A wireess sensor networ consists of a set of spatiay scattered sensors that can be used to monitor and protect critica infrastructure assets such as power grids automated rairoad contro water and gas distribution etc. However due to the unattended operating environment of sensor networs it coud be easy for attacers to compromise sensors and conduct maicious behaviors. Anomay detection is thus critica to ensure the effective functioning of sensor networs. An anomay detection procedure usuay consists of two steps: first coect networ measurements and mode the norma traffic as a reference; second appy a decision rue to detect whether current networ traffic deviates from the reference.traditiona anomay detection methods usuay assume that the networ measurements are either independent or short range dependent. However recent studies have shown that node mobiity aong with spatia correation of the monitored phenomenon in sensor networs can ead to Long Range Dependent (LRD) traffic [] which coud ead to high fase aarms using traditiona methods. In this paper we deveop an anomay detection scheme based on muti-scae anaysis of the ong range dependent traffic. Discrete Waveet Transform (DWT) is used to approximatey de-correate autocorreated stochastic processes. Most of the iterature wor on using DWT for anomay detection use the first order or second order statistics (mean or variance) of the waveet coefficients for anomay detection e.g. detect changes in the mean or variance of the waveet coefficients over a moving window [2] [3]. In contrast we buid a probabiistic mode for the waveet coefficients through a muti-eve hidden Marov mode (HMM) in the expect to capture the remaining dependency This wor was supported by the Defense Advanced Research Projects Agency (DARPA) under award number for the Muti-Scae Systems Center (MuSyC) through the FRCP of SRC and DARPA US Air Force Office of Scientific Research MURI award FA and Army Research Office MURI award W9-NF Shanshan Zheng and John S. Baras are with the Institute for Systems Research and the Department of Eectrica and Computer Engineering University of Maryand Coege Par {sszheng baras}@umd.edu among the transformed data thus better mode the networ traffic and improve detection accuracy. We design a forwardbacward decomposition scheme and an onine discounting Expectation Maximization (EM) agorithm to estimate mode parameters. The onine EM agorithm can aso trac the structure changes of the estimated HMMs by evauating a mode variation score using the symmetric reative entropy between the current estimated mode and a previous estimated mode. Networ anomaies are then detected as abrupt changes in the traced mode variation scores. II. RELATED WORK Networ anomay detection is an important probem and has received numerous research efforts. It invoves modeing of the norma traffic as a reference and computing the statistica distance between the anayzed traffic and the reference. The modeing of the norma traffic can be based on various statistica characteristics of the data. For exampe Lahina et a. [4] proposed to use Principa Component Anaysis to identify an orthogona basis aong which the networ measurements exhibit the highest variance. The principa components with high variance mode the norma behavior of a networ whereas the remaining components of sma variance are used to identify and cassify anomaies. Spectra density [5] and covariance [6] have aso been used for modeing norma networ traffic. Besides these methods waveet transform is another popuar technique used for modeing networ traffic especiay for the LRD traffic. Abry et a. [7] proposed a waveetbased too for anayzing LRD time series and a reated semi-parametric estimator for estimating LRD parameters. Barford et a. [2] assume that the ow frequency band signa of a waveet transform represents the norma traffic pattern. They then normaize both medium and high frequency band signas to compute a weighted sum. An aarm is raised if the variance of the combined signa exceeds a pre-seected threshod. The ey feature of these waveet-based methods ies in the fact that waveet transform can turn the ong range dependency among the data sampes into a short memory structure among the waveet coefficients [7]. In our wor we buid a waveet-domain muti-eve hidden Marov mode for the LRD networ traffic. The merit of our method is the mode s mathematica tractabiity and its capabiity of capturing data dependency. To measure the deviation of the anayzed traffic from the reference mode various statistica distances can be used incuding simpe threshod mean quadratic distances [8] and entropy[9]. Entropy is a measure of the uncertainty of a probabiity distribution. It can be used to compare certain quaitative differences of probabiity distributions. In our detection scheme we appy the symmetric reative entropy as a distance measure. The onine EM agorithm can efficienty compute the symmetric reative entropy between the current estimated HMM mode and a previous estimated one. An anomay is then detected as abrupt changes in the symmetric reative entropy measurements //$ IEEE 4060
2 III. LONG RANGE DEPENDENT TRAFFIC IN WIRELESS SENSOR NETWORKS A time series {x(t)} N t is considered to be ong range dependent if its autocorreation function ρ() decays at a rate sower than an exponentia decay. Typicay ρ() asymptoticay behaves as c 2H 2 for 0.5 < H < where c > 0 is a constant and H is the Hurst parameter. The intensity of LRD is expressed as the speed of the decay for the autocorreation function and is measured by the Hurst parameter i.e. as H the dependence among data becomes stronger. It can be shown that ρ(). Intuitivey LRD impies that the process has infinite memory i.e. individuay sma highag correations have an important cumuative effect. This is contrast to the conventiona Short Range Dependent (SRD) process which are characterized by an exponentia decay of the autocorreations resuting in a summabe autocorreation function. LRD is an important property for traffic modeing as it is iey to be responsibe for the decrease in both the networ performance and the quaity of service [8]. A wireess sensor networ operates on the IEEE standard. It has been shown recenty [] that the traffic generated from a singe mobie node in the wireess sensor networ can be represented by an ON/OFF process X(t) where the probabiity density function of the ON period τ a can be approximated by a truncated Pareto distribution [] f τa (x) γ ax (γa+) t γa min t γa max with t min (t max ) denoting the minimum (maximum) ON time and γ a denoting the tai index. The vaue of γ a depends on the variabiity of node mobiity pattern and the spatia correation of the monitored phenomena by the networ. Using this traffic mode for wireess sensor networs we anayze the dynamic behaviors of networ traffic measurements such as the pacet round trip time the number of received pacets per second etc. by conducting experiments using Networ simuator 2 (NS-2). It is found that the LRD property do exit in the simuated data traces with typica Hurst parameters between 0.8 and 0.9. Incorporating LRD precisey in the design of anomay detection schemes for wireess sensor networs is critica. IV. WAVELET DOMAIN HIDDEN MARKOV MODEL FOR LONG RANGE DEPENDENT TRAFFIC Waveet transforms have been popuar for anayzing autocorreated measurements due to their capabiity to compress muti-scae features and approximatey de-correate the autocorreated stochastic processes. We buid a Hidden Marov Mode (HMM) for the waveet transform coefficients of the networ traffic. The basic idea for transform domain mode is that a inear invertibe transform can often restructure an signa generating transform coefficients whose structure is simper to mode. A. Waveet domain hidden Marov mode In waveet transform the measurements x(t) t... N are decomposed into mutipe scaes by a weighted sum of a certain orthonorma basis functions N L x(t) a L φ L (t) + d m ψ m (t) m where φ j ψ j are the orthonorma basis a L d m are the approximation and detai coefficients. The approximation coefficients a L provide the genera shape of the signa whie the detai coefficients d m from different scaes provide different eves of detais for the signa content with d providing the finest detais and d L providing the coarsest detais. In our wor we appy the Discrete Waveet Transform (DWT) to the networ traffic. A discrete waveet transform is a waveet transform for which the basis functions are discretey samped. DWT can be expained using a pair of quadrature mirror fiters which incudes a high pass fiter h[n] and a ow pass fiter g[n]. Efficient methods have been deveoped for decomposing a signa using a famiy of waveet basis functions based on convoution with the corresponding quadrature mirror fiters. A 2-eve discrete waveet transform using the corresponding quadrature mirror fiters is iustrated in Fig.. Fig. : 2-eve discrete waveet transform However the waveet transform cannot competey decorreate rea-word signas i.e. a residua dependency structure aways remains among the waveet coefficients. We use a hidden Marov mode to capture the remaining dependency. It is based on two properties of the waveet transform observed in iterature [0] []: first is the Custering property which means that if a particuar waveet coefficient is arge/sma the adjacent coefficients are very iey to aso be arge/sma; second is the Persistence property which means that arge/sma vaues of waveet coefficients tend to propagate across scaes. For a signa x(t) that is decomposed into L scaes with waveet coefficients d j... n j and j... L we assume that each waveet coefficient is associated with a hidden state s j. We then use a hidden Marov mode to characterize the waveet coefficients through the factorization P ({d i s i } n i... {d Li s Li } n L i ) p(s L ) L n i i j2 n L j2 p(s Lj s Lj ) L i p(s ij s ij s i+ j/2 ) p(s i s i+ ) L n i p(d ij s ij ). () i j This factorization invoves three main conditiona independence assumptions: first conditioned on the states at the previous coarser scae i + the states at the scae i form a first order Marov chain; second conditioned on the corresponding state at the previous coarser scae i + i.e. s i+ j/2 and the previous state at the same scae i.e. s ij the state s ij is independent of a states in coarser scaes; third the waveet coefficients are independent of everything ese given their hidden states. The three independence assumptions are critica for deriving the inference agorithms for this waveet domain HMM. Fig. 2 iustrate a hidden Marov mode for a 3-eve waveet decomposition. B. Estimating mode parameters using an Expectation- Maximization (EM) agorithm Denote the set of the waveet coefficients and their hidden states by D {{d Li } n L i... {d i} n i } and S {{s Li } n L i... {s i} n i } respectivey where n i is the number of waveet coefficients in the i th scae. The parameters of the HMM incude the foowing three probabiities: first is the initia probabiity for the state s L i.e. π 406
3 O(N K L+ ) where L is the waveet decomposition eve and much smaer than N. Fig. 2: HMM for 3-eve waveet decomposition P (s L ); second is the two types of state transition probabiities i.e. π i 2 P (s i s i+ 2 ) for i < L π i 2 3 P (s ij s ij 2 s i+ j/2 3 ); and third is the conditiona probabiity of the waveet coefficients given their hidden states at the i th scae i.e. P (d ij s ij ) which can be modeed by a mixture Gaussian distribution. For simpicity and presentation cearity we use a singe Gaussian distribution to catpure P (d ij s ij ) i.e P (d ij s ij ) N (µ i σi ) where µi and σi are the mean and the standard deviation for the state in the i th scae. The extension to mixture Gaussian distribution is straightforward. These mode parameters denoted by θ {π π i 2 π i 2 3 µ i σi } can be estimated from the rea data using the maximum ieihood criterion. Due to the intractabiity of direct maximization of the ieihood function we appy an Expectation Maximization (EM) agorithm to estimate the parameters. The EM agorithm provides an maximum ieihood estimation of mode parameters by iterativey appying an E-step and a M-step. In the E-step the expected vaue of the og ieihood function Q(θ θ (t) ) E S Dθ (t)[og P θ (S D)] is computed. Then in the M-step the parameter that maximizes Q(θ θ (t) ) is computed i.e. θ (t+) arg max θ Q(θ θ (t) ). To impement the two steps we define the foowing posterior probabiities γ ij P (s ij D) γ i 2 P (s i s i+ 2 D) for i < L γ ij 2 3 P (s ij s ij 2 s i j/2 3 D). According to equation () maximizing Q(θ θ (t) ) using Lagrange method eads to the foowing estimation of θ π γ L π i 2 γi 2 K γi ni µ i j γij d ij ni j γij ni π i 2 3 ni 2 K (σ i ) 2 ni j γij j2 γij 2 3 j2 γij 2 3 (d ij µ i )2 ni j γij where K represents the domain of the hidden states. The computation of the posterior probabiities is a itte more invoved. Using a brute force computation by direct marginaization wi tae O(N K N ) operations where N is the ength of the input signa. However by expoiting the sparse factorization in equation () and manipuating the distributive property of + and we are abe to design an forward-bacward decomposition agorithm to compute these posterior probabiities with computationa compexity C. Forward-bacward decomposition Our agorithm extends the cassica forward-bacward decomposition agorithm for a one-eve hidden Marov mode to our muti-eve case. The ey point is to ony maintain L appropriate hidden states in both the forward and bacward variabes for computationa efficiency. ) Forward decomposition: Let S ij {s L 2 i L j... s i+ 2 j s ij s i 2(j )... s 2 i (j )} and D ij {d L 2 i L j... d i+ 2 j d i j d i 2(j )... d 2 i (j )} we define the forward variabe to be α ij P (S ij D ij ). Denote [α 2 h j] f(h α hj ) for h j Z + to be a dynamic programming agorithm with input parameters (h α hj ) and output parameter α 2 h j. The pseudo-code for computing the forward variabes using dynamic programming is shown in Tabe I. Its correctness can be proved using the three conditiona independence assumptions in our HMM. For simpicity and presentation carity in Tabe I we assume that the input data ength N is an order of 2 and denote the conditiona probabiity P (d ij s ij ) by g (d ij ) and P (s ij s ij s i+ j/2 ) by g 2 (s ij ). TABLE I: Computing forward variabes Initiaization: α L P (s L d L ) For L to 2 L N α 2 L L f(l α LL ) α LL + g (d LL +) s LL [g 2 (s LL +) α 2 L L ] end function [α 2 j] f(h α h hj ) If h 2 α 2j g (d 2j ) s 2j 2 [g 2 (s 2j ) α 2j ] α 2j g (d 2j s 2j ) s 2j [g 2 (s 2j ) α 2j ] ese α h 2j g (d h 2j ) [g 2 (s h 2j ) α hj ] s h 2j 2 α 2 h 2 (2j ) f(h α h 2j ) α h 2j g (d h 2j ) [g 2 (s h 2j ) α 2 h 2 (2j )] s h 2j α 2 h 2 (2j) f(h α h 2j ) End There is one impementation issue for the agorithm in Tabe I namey the numerica under- or over-fow of α ij as P (S ij D ij ) becomes smaer and smaer with the increasing number of observations. Therefore it is necessary to scae the forward variabes by positive rea numbers to eep the numeric vaues within reasonabe bounds. One soution is to use a scaed version ᾱ ij αij c ij where c ij S ij α ij. In this way ᾱ ij represents the probabiity P (S ij D ij ) and c ij represents the probabiity P (d ij D ij \d ij ). It is straightforward to prove that both c ij and ᾱ ij do not depend on the number of observations. The agorithm for computing (ᾱ ij c ij ) can be obtained by adding a normaization step after each update of α ij for the agorithm in Tabe I. 4062
4 2) Bacward decomposition: Let D c ij D D ij we define the bacward variabe to be β ij P (D c ij S ij) which can be computed using a simiar dynamic programming agorithm as the one in Tabe I. To avoid the numerica underor over-fow probem instead of computing β ij we compute a scaed version β ij as is shown in Tabe II. The scaed bacward variabe represents the probabiity P (Dc ij Sij) P (D c ij Dij). The correctness of the agorithm in Tabe II can be verified using the three conditiona independence assumptions in our HMM. Due to space imit we omit the proof here. TABLE II: Computing scaed bacward variabes Initiaization: β N/2 For L 2 L N to β LL f(l β 2 L L ) end β 2 L ( L ) s LL g (d LL )g 2(s LL ) β LL c LL function [ β hj ] f(h β 2 h j) If h 2 β 2j s 2j g (d 2j)g 2(s 2j) β 2j c 2j g (d 2j )g 2(s 2j ) β 2j c 2j β 2j s 2j ese β h 2j f(h β 2 h 2 2j) β 2 h 2 (2j ) s h 2j β h 2j f(h β 2 h 2 (2j )) End β hj s h 2j g (d h 2j )g 2(s h 2j ) β h 2j c h 2j g (d h 2j )g 2(s h 2j ) β h 2j c h 2j 3) Computing posterior probabiities: Since ᾱ ij P (S ij D ij ) and β ij P (Dc ij Sij) P (D c ij Dij) we have ᾱ ij β ij P (S ij D) according to the Marovian property of our HMM. Then the posterior probabiity γ( ) can be computed as γ ij ᾱ ij β ij γ i 2 ᾱ i β i γ Lj 2 ᾱ 2 L (j ) β Lj g(d Lj )g 2 (s Lj ) c Lj { β γ ij ᾱ2 i (j ) ij g(dij) g2(sij) c 2 3 ij if j is even ᾱi+ j/2 βij g(dij) g2(sij) c ij if j is odd. Without confusion we omit the variabes under for the above equations. The correctness of these equations can be derived according to the three conditiona independence assumptions in our HMM. V. ANOMALY DETECTION BY TRACKING HMM MODEL VARIATIONS A first thought on the anomay detection probem is to treat the anomaies as abrupt changes in the HMM modeed data and then appy change-point detection methods to detect these abrupt changes. However it is found that directy appying change-point detection methods to the HMM modeed data is computationay expensive. We designed here a ightweight anomay detection scheme based on detecting the structure changes of the estimated HMM. An important requirement for anomay detection is to mae the decision maing process onine. Therefore we first deveop an onine EM agorithm for HMM mode estimation. A. An onine discounting EM agorithm Before we present the onine EM agorithm we first introduce the so caed imiting EM agorithm [2]. Let x denote the hidden states and y denote the observations. If the joint probabiity distribution p θ (x y) beongs to an exponentia famiy such that p θ (x y) h(x y) exp( φ(θ) ss(x y) A(θ)) where denotes the scaar product ss(x y) is the sufficient statistics and A(θ) is some og-partition function. If the equation θ φ(θ) ss θ A(θ) 0 has a unique soution denoted by θ θ(ss) then the imiting EM agorithm obeys the simpe recursion ss + E θ [E θ(ss )[ss(x y) y]] where θ represents the true mode parameter. Since E θ [E θ [ss(x y) y]] may be estimated consistenty from the observations by n n t E θ[ss(x t y t ) y t ] an onine EM agorithm can be obtained by using the conventiona stochastic approximation procedure ŝs + γ + E θ(ŝs )[ss(x + y + ) y + ]+( γ + )ŝs where γ is a time discounting factor. The estimation of mode parameters can then be derived from the sufficient statistics ŝs. It is proved [2] that under suitabe assumptions this onine EM agorithm is an asymptoticay efficient estimator of the mode parameter θ. It is not difficut to see that the joint probabiity distribution P (S D) for our HMM mode satisfies the above mentioned conditions for appying the imiting EM agorithm. For each waveet coefficient d ij we have the foowing sufficient statistics for computing the HMM mode parameters n ij m P (s m S ij D ij ) ˆ n ij m P (s m S ij D ij ) d m n ij m P (s m S ij D ij ) d 2 m 2 3 n ij m2 P (s m s m 2 s + m/2 3 S ij D ij ) where {... L} is the scae index K is the hidden state index and n ij represents the number of observed waveet coefficients in scae after d ij arrives i.e. { n ij 2 i j if i 2 i j if < i. It is straightforward to prove that the HMM mode parameters {π 2 3 µ σ } can be updated using { ij ij ˆτ τ 2 3 } as foows π 2 3 µ (σ ) 2 S ij 2 3 S ij S ij ˆ S ij S ij S ij 2 3 (2) (3) ( S ij ˆ S ij ) 2. (4) Note that the HMM parameter π and π i 2 can be updated using the sufficient statistics L P (s L S ij D ij ) and P (s s + 2 S ij D ij ). But we omit 4063
5 the reated discussions here as the computations are simiar. The next step is to design recursive (onine) updates of the sufficient statistics ij ij ij ˆτ τ and τ 2 3. According to the Marovian property of our HMM the onine updates of the sufficient statistics can be achieved by foowing a simiar dynamic programming procedure as the one for computing the scaed forward variabes ᾱ ij. Reca that ᾱ ij is computed by adding a normaization step after α ij is computed in the agorithm in Tabe I. The sufficient statistics are updated once ᾱ ij is updated. For iustration purpose we ony show here how to update the sufficient statistics when α h2j in Tabe I is computed. Updates for the other cases are simiar. For {... L} et γ h 2j be a time discounting factor and δh be the Dirac Deta function such that { δh if h 0 if h. Define rh 2j δ h γ h 2j ᾱ h 2j t h 2j ( δ h γ h 2j )g (d h 2j )/c hj q h 2j δ h γ h 2j g(d h 2j )g 2(s h 2j )ᾱ hj c h 2j. We then have the foowing equations for updating the sufficient statistics τ h 2j ˆτ h 2j τ h 2j τ h 2j 2 3 r h 2j + t h 2j s h 2j 2 g 2 (s h 2j )τ hj (5) rh 2j d h 2j + t h 2j g 2 (s h 2j )ˆτ hj (6) s h 2j 2 rh 2j d 2 h 2j + t h 2j g 2 (s h 2j ) τ hj (7) s h 2j 2 qh 2j + t h 2j s h 2j 2 g 2 (s h 2j ) τ hj 2 3.(8) The correctness of these updates can be proved using the three conditiona independence assumptions in our HMM. Due to space imit we omit the proof here. B. Change-point detection on mode variations To measure the structure changes of the estimated HMM modes over time we use the concept of the symmetric reative entropy to define a mode variation score. Denote the mode at time t and t by P t and P t respectivey then the mode variation score is defined to be v t im n n D(P t P t ) + im n n D(P t P t ) where D(p q) represents the reative entropy of distribution p to q and n represents the ength of the input data. It is natura to et n as we can then compare the two modes under the stationary states in the imit of n. It can be proved that im n n D(P t P t ) + i L i 2 i D((πi 2 3 ) t (π i 2 3 ) t ) L 2 i (π) i t D(N ((µ i ) t (σ) i t ) N ((µ i ) t (σ) i t )) where π i P (s ij ). Therefore besides the probabiity distributions π i 2 3 and N (µ i σi ) provided by the onine EM agorithm the computation of im n n D(P t P t ) aso invoves the computation of the probabiity distributions π i 2 3 P (s ij s ij 2 s i+ j/2 3 ) and π i. The estimation of πi and πi 2 3 can be obtained from the sufficient statistics τ m m i and τi 2 3 as foows π i τ m i πi 2 3 τ m i 2 3. S m S m The other reative entropy im n n D(P t P t ) can be computed simiary. We can see that the symmetric mode variation score actuay captures two types of changes. The first is the changes in the transition probabiities of the hidden states whie the second is the changes in the generation pattern of the observed data from a fixed state. By using the symmetric reative entropy as a distance measure between two HMM modes it is expected that not ony the changes of the data generation pattern wi be detected but aso the changes in the hidden states can aso be detected. VI. EVALUATIONS We evauate the performance of our anomay detection agorithm numericay. The agorithm performances are evauated using the detection atency and the Receiver Operating Characteristic (ROC) curve which is a pot of the detection rate versus the fase aarm rate at different threshod vaues. The seection of the waveet basis used in our anomay detection scheme is based on a baance between its time ocaization and frequency ocaization characteristics [2]. In our experiments we found that for the Daubechies famiy waveets the D2 (Haar waveets) and D4 waveets can give us reasonaby good performance. Hence we use the Haar waveets for a the experiments in this paper. In the numerica experiments two we nown LRD time series incuding the Fractiona Gaussian Noise (FGN) and the Autoregressive Fractionay Integrated Moving Average (ARFIMA) mode are used for generating data. We then inject two types of mode variations into the time series as anomaies. First is the mean eve shift i.e. a step function with a constant ampitude wi be imposed on the origina signa. Second is to vary mode parameters for the data generation process incuding the standard deviation and the Hurst parameter for FGN and ARFIMA. The duration for the norma state and the anomay state is generated from exponentia distribution with different mean vaues. We first discuss the detection performance on mean eve shifts in the synthetic LRD time series. Fig. 3 shows one representative exampe. The top figure iustrates the time series which is generated from an ARFIMA mode with Hurst parameter 0.9 and ength 2 5. The standard deviation for the generated data sequence is set to be. The mean eve shift occurs at the first quarter of the time series and ends at the midde with intensity 0.75 which is ess than the standard deviation. The bottom two figures show the corresponding mode variation scores computed by our onine EM agorithm with 5-eve and 4-eve waveet decomposition respectivey. The x axis represents the scaed time due to waveet transform and the y axis represents the mode variation score. From Fig. 3 we can see that the visua inspection of the injected mean eve shift from the time series directy can be difficut. However in the traced mode variation scores there are abrupt changes at the two time ocations where the mean eve shift starts and ends. These two abrupt changes suggest where the injection starts and ends. Especiay when 4064
6 data Time series data with mean eve shift origina time x 0 4 Decomposition eve 5 detection. It is observed that our method can aways beat this baseine method. For exampe Fig. 5 shows the ROC curves and detection atency for our method and the baseine methods on the detection of Hurst parameter changing from 0.9 to 0.7. score score scaed time Decomposition eve scaed time Fig. 3: Effects of different decomposition eves the waveet decomposition eve is 5 these are the ony 2 abrupt changes that exist. When the waveet decomposition eve is 4 there are some fase aarms therefore the decomposition eve of the waveet transform is an important parameter for the anomay detection scheme. A higher eve decomposition can give higher detection rate and smaer fase aarm rate. However since an L-eve decomposition has a 2 L time aggregation scae i.e. it transforms the data sampes within a 2 L time window to the waveet domain so the waveet coefficients within this window is timeindistinguishabe a higher eve decomposition woud often give onger detection atency than that of a ower eve decomposition. In our experiments we found that a 5-eve waveet decomposition can give a reasonabe good baance between detection accuracy and atency. The intensity of the injected mean eve shift aso affects the detection performance. Fig. 4 shows the ROC curve and detection atency for the injected mean shift with different intensities. Each curve is obtained over 000 simuation traces with the 5-eve waveet decomposition. As is expected for higher injected mean eve shifts the detection becomes much easier in terms of ower fase aarms higher detection rates and smaer detection atency. Detection rate Detection atency ROC curve mean shift0.75 mean shift mean shift Fase aarm rate FAR vs. Latency mean shift0.75 mean shift mean shift Fase aarm rate Fig. 4: Effects of different injected intensities For the second type of anomaies i.e. the one that varies mode parameters for the data generation process simiar good performances are observed for our detection method. When the variation becomes arger the detection becomes easier. Due to space imit we omit the resuts here. For performance comparison we impement a baseine method adapted from [2] in which ony the mean and variance of the waveet coefficients is used for anomay Detection rate Detection atency ROC our method baseine method Fase aarm rate FAR vs. Detection atency our method baseine method Fase aarm rate Fig. 5: Comparison of our method and the baseine method VII. CONCLUSIONS In this paper we studied the anomay detection probem for LRD traffic in wireess sensor networs. We proposed a waveet-domain hidden Marov mode for capturing the properties of networ traffic. The waveet transform is abe to turn the ong range dependency that exists among the sampe data into a short memory structure among its waveet coefficients. The HMM in the waveet-domain is used to further capture the remaining dependency among the waveet coefficients thus mode the traffic variabiity more accuratey. Networ anomaies are then detected as abrupt changes in the traced HMM mode structures. We evauate the performance of our agorithm numericay using typica LRD time series. In the future wor we pan to study the optimization of mode parameters for the waveet domain HMM mode in order to achieve better performance. REFERENCES [] P. Wang and I. F. Ayidiz Spatia correation and mobiity aware traffic modeing for wireess sensor networs in Proceedings of Gobecom [2] P. Barford J. Kine D. Pona and A. Ron A signa anaysis of networ traffic anomaies in ACM SIGCOMM Internet Measurement Worshop [3] P. Zuraniewsi and D. Rincon Waveet transforms and chengepoint detection agorithms for tracing networ traffic fractaity in Proceedings of 2nd Conference on Next Generation Internet Design and Engineering [4] A. Lahina M. Crovea and C. Diot Mining anomaies using trafc feature distributions in Proceedings of ACM SIGCOMM [5] A. Hussain03 J. Heidemann and C. Papadopouos A framewor for cassifying denia of service attacs in Proceedings of ACM SIGCOMM [6] S. Jin and D. Yeung A covariance anaysis mode for DDoS attac detection in Proceedings of IEEE ICC [7] P. Abry and D. Veitch Waveet anaysis of ong range dependent traffic IEEE Transactions of Information Theory 998. [8] L. Rabiner Non-Gaussian and ong memory statistica characterisations for Internet traffic with anomaies IEEE Transactions on Dependabe and Secure Computing [9] Y. Gu A. McCaum and D. Towsey Detecting anomaies in networ traffic using maximum entropy estimation in Proceedings of the 5th ACM SIGCOMM conference on Internet Measurement [0] M. Crouse R. Nowa and R. Baraniu Waveet-based statistica signa processing using hidden Marov modes IEEE Transactions on signa processing 998. [] S. Maat and S. Zhong Characterization of signas from mutiscae edges IEEE Transactions on Pattern Anaysis and Machine Inteigence 992. [2] O. Cappe Onine EM agorithm for hidden Marov modes
A Brief Introduction to Markov Chains and Hidden Markov Models
A Brief Introduction to Markov Chains and Hidden Markov Modes Aen B MacKenzie Notes for December 1, 3, &8, 2015 Discrete-Time Markov Chains You may reca that when we first introduced random processes,
More informationExpectation-Maximization for Estimating Parameters for a Mixture of Poissons
Expectation-Maximization for Estimating Parameters for a Mixture of Poissons Brandon Maone Department of Computer Science University of Hesini February 18, 2014 Abstract This document derives, in excrutiating
More informationSTA 216 Project: Spline Approach to Discrete Survival Analysis
: Spine Approach to Discrete Surviva Anaysis November 4, 005 1 Introduction Athough continuous surviva anaysis differs much from the discrete surviva anaysis, there is certain ink between the two modeing
More informationAsynchronous Control for Coupled Markov Decision Systems
INFORMATION THEORY WORKSHOP (ITW) 22 Asynchronous Contro for Couped Marov Decision Systems Michae J. Neey University of Southern Caifornia Abstract This paper considers optima contro for a coection of
More informationOptimality of Inference in Hierarchical Coding for Distributed Object-Based Representations
Optimaity of Inference in Hierarchica Coding for Distributed Object-Based Representations Simon Brodeur, Jean Rouat NECOTIS, Département génie éectrique et génie informatique, Université de Sherbrooke,
More informationApproach to Identifying Raindrop Vibration Signal Detected by Optical Fiber
Sensors & Transducers, o. 6, Issue, December 3, pp. 85-9 Sensors & Transducers 3 by IFSA http://www.sensorsporta.com Approach to Identifying Raindrop ibration Signa Detected by Optica Fiber ongquan QU,
More informationA. Distribution of the test statistic
A. Distribution of the test statistic In the sequentia test, we first compute the test statistic from a mini-batch of size m. If a decision cannot be made with this statistic, we keep increasing the mini-batch
More informationSequential Anomaly Detection in Wireless Sensor Networks and Effects of Long-Range Dependent Data
This article was downloaded by: [Nationwide Childrens Hospital], [John Baras] On: 16 October 2012, At: 19:28 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954
More informationASummaryofGaussianProcesses Coryn A.L. Bailer-Jones
ASummaryofGaussianProcesses Coryn A.L. Baier-Jones Cavendish Laboratory University of Cambridge caj@mrao.cam.ac.uk Introduction A genera prediction probem can be posed as foows. We consider that the variabe
More informationAST 418/518 Instrumentation and Statistics
AST 418/518 Instrumentation and Statistics Cass Website: http://ircamera.as.arizona.edu/astr_518 Cass Texts: Practica Statistics for Astronomers, J.V. Wa, and C.R. Jenkins, Second Edition. Measuring the
More informationarxiv: v1 [cs.lg] 31 Oct 2017
ACCELERATED SPARSE SUBSPACE CLUSTERING Abofaz Hashemi and Haris Vikao Department of Eectrica and Computer Engineering, University of Texas at Austin, Austin, TX, USA arxiv:7.26v [cs.lg] 3 Oct 27 ABSTRACT
More informationAn Algorithm for Pruning Redundant Modules in Min-Max Modular Network
An Agorithm for Pruning Redundant Modues in Min-Max Moduar Network Hui-Cheng Lian and Bao-Liang Lu Department of Computer Science and Engineering, Shanghai Jiao Tong University 1954 Hua Shan Rd., Shanghai
More informationFORECASTING TELECOMMUNICATIONS DATA WITH AUTOREGRESSIVE INTEGRATED MOVING AVERAGE MODELS
FORECASTING TEECOMMUNICATIONS DATA WITH AUTOREGRESSIVE INTEGRATED MOVING AVERAGE MODES Niesh Subhash naawade a, Mrs. Meenakshi Pawar b a SVERI's Coege of Engineering, Pandharpur. nieshsubhash15@gmai.com
More informationResearch of Data Fusion Method of Multi-Sensor Based on Correlation Coefficient of Confidence Distance
Send Orders for Reprints to reprints@benthamscience.ae 340 The Open Cybernetics & Systemics Journa, 015, 9, 340-344 Open Access Research of Data Fusion Method of Muti-Sensor Based on Correation Coefficient
More informationLecture Note 3: Stationary Iterative Methods
MATH 5330: Computationa Methods of Linear Agebra Lecture Note 3: Stationary Iterative Methods Xianyi Zeng Department of Mathematica Sciences, UTEP Stationary Iterative Methods The Gaussian eimination (or
More informationBayesian Learning. You hear a which which could equally be Thanks or Tanks, which would you go with?
Bayesian Learning A powerfu and growing approach in machine earning We use it in our own decision making a the time You hear a which which coud equay be Thanks or Tanks, which woud you go with? Combine
More information6.434J/16.391J Statistics for Engineers and Scientists May 4 MIT, Spring 2006 Handout #17. Solution 7
6.434J/16.391J Statistics for Engineers and Scientists May 4 MIT, Spring 2006 Handout #17 Soution 7 Probem 1: Generating Random Variabes Each part of this probem requires impementation in MATLAB. For the
More informationBayesian Unscented Kalman Filter for State Estimation of Nonlinear and Non-Gaussian Systems
Bayesian Unscented Kaman Fiter for State Estimation of Noninear and Non-aussian Systems Zhong Liu, Shing-Chow Chan, Ho-Chun Wu and iafei Wu Department of Eectrica and Eectronic Engineering, he University
More informationFRST Multivariate Statistics. Multivariate Discriminant Analysis (MDA)
1 FRST 531 -- Mutivariate Statistics Mutivariate Discriminant Anaysis (MDA) Purpose: 1. To predict which group (Y) an observation beongs to based on the characteristics of p predictor (X) variabes, using
More informationSeasonal Time Series Data Forecasting by Using Neural Networks Multiscale Autoregressive Model
American ourna of Appied Sciences 7 (): 372-378, 2 ISSN 546-9239 2 Science Pubications Seasona Time Series Data Forecasting by Using Neura Networks Mutiscae Autoregressive Mode Suhartono, B.S.S. Uama and
More informationA proposed nonparametric mixture density estimation using B-spline functions
A proposed nonparametric mixture density estimation using B-spine functions Atizez Hadrich a,b, Mourad Zribi a, Afif Masmoudi b a Laboratoire d Informatique Signa et Image de a Côte d Opae (LISIC-EA 4491),
More informationImproving the Accuracy of Boolean Tomography by Exploiting Path Congestion Degrees
Improving the Accuracy of Booean Tomography by Expoiting Path Congestion Degrees Zhiyong Zhang, Gaoei Fei, Fucai Yu, Guangmin Hu Schoo of Communication and Information Engineering, University of Eectronic
More informationScalable Spectrum Allocation for Large Networks Based on Sparse Optimization
Scaabe Spectrum ocation for Large Networks ased on Sparse Optimization innan Zhuang Modem R&D Lab Samsung Semiconductor, Inc. San Diego, C Dongning Guo, Ermin Wei, and Michae L. Honig Department of Eectrica
More informationStatistical Learning Theory: A Primer
Internationa Journa of Computer Vision 38(), 9 3, 2000 c 2000 uwer Academic Pubishers. Manufactured in The Netherands. Statistica Learning Theory: A Primer THEODOROS EVGENIOU, MASSIMILIANO PONTIL AND TOMASO
More informationA Simple and Efficient Algorithm of 3-D Single-Source Localization with Uniform Cross Array Bing Xue 1 2 a) * Guangyou Fang 1 2 b and Yicai Ji 1 2 c)
A Simpe Efficient Agorithm of 3-D Singe-Source Locaization with Uniform Cross Array Bing Xue a * Guangyou Fang b Yicai Ji c Key Laboratory of Eectromagnetic Radiation Sensing Technoogy, Institute of Eectronics,
More informationHYDROGEN ATOM SELECTION RULES TRANSITION RATES
DOING PHYSICS WITH MATLAB QUANTUM PHYSICS Ian Cooper Schoo of Physics, University of Sydney ian.cooper@sydney.edu.au HYDROGEN ATOM SELECTION RULES TRANSITION RATES DOWNLOAD DIRECTORY FOR MATLAB SCRIPTS
More informationCS229 Lecture notes. Andrew Ng
CS229 Lecture notes Andrew Ng Part IX The EM agorithm In the previous set of notes, we taked about the EM agorithm as appied to fitting a mixture of Gaussians. In this set of notes, we give a broader view
More informationStochastic Complement Analysis of Multi-Server Threshold Queues. with Hysteresis. Abstract
Stochastic Compement Anaysis of Muti-Server Threshod Queues with Hysteresis John C.S. Lui The Dept. of Computer Science & Engineering The Chinese University of Hong Kong Leana Goubchik Dept. of Computer
More informationMONTE CARLO SIMULATIONS
MONTE CARLO SIMULATIONS Current physics research 1) Theoretica 2) Experimenta 3) Computationa Monte Caro (MC) Method (1953) used to study 1) Discrete spin systems 2) Fuids 3) Poymers, membranes, soft matter
More informationTraffic data collection
Chapter 32 Traffic data coection 32.1 Overview Unike many other discipines of the engineering, the situations that are interesting to a traffic engineer cannot be reproduced in a aboratory. Even if road
More informationFitting Algorithms for MMPP ATM Traffic Models
Fitting Agorithms for PP AT Traffic odes A. Nogueira, P. Savador, R. Vaadas University of Aveiro / Institute of Teecommunications, 38-93 Aveiro, Portuga; e-mai: (nogueira, savador, rv)@av.it.pt ABSTRACT
More informationXSAT of linear CNF formulas
XSAT of inear CN formuas Bernd R. Schuh Dr. Bernd Schuh, D-50968 Kön, Germany; bernd.schuh@netcoogne.de eywords: compexity, XSAT, exact inear formua, -reguarity, -uniformity, NPcompeteness Abstract. Open
More informationTurbo Codes. Coding and Communication Laboratory. Dept. of Electrical Engineering, National Chung Hsing University
Turbo Codes Coding and Communication Laboratory Dept. of Eectrica Engineering, Nationa Chung Hsing University Turbo codes 1 Chapter 12: Turbo Codes 1. Introduction 2. Turbo code encoder 3. Design of intereaver
More informationFFTs in Graphics and Vision. Spherical Convolution and Axial Symmetry Detection
FFTs in Graphics and Vision Spherica Convoution and Axia Symmetry Detection Outine Math Review Symmetry Genera Convoution Spherica Convoution Axia Symmetry Detection Math Review Symmetry: Given a unitary
More informationBP neural network-based sports performance prediction model applied research
Avaiabe onine www.jocpr.com Journa of Chemica and Pharmaceutica Research, 204, 6(7:93-936 Research Artice ISSN : 0975-7384 CODEN(USA : JCPRC5 BP neura networ-based sports performance prediction mode appied
More informationEfficiently Generating Random Bits from Finite State Markov Chains
1 Efficienty Generating Random Bits from Finite State Markov Chains Hongchao Zhou and Jehoshua Bruck, Feow, IEEE Abstract The probem of random number generation from an uncorreated random source (of unknown
More informationII. PROBLEM. A. Description. For the space of audio signals
CS229 - Fina Report Speech Recording based Language Recognition (Natura Language) Leopod Cambier - cambier; Matan Leibovich - matane; Cindy Orozco Bohorquez - orozcocc ABSTRACT We construct a rea time
More informationA Statistical Framework for Real-time Event Detection in Power Systems
1 A Statistica Framework for Rea-time Event Detection in Power Systems Noan Uhrich, Tim Christman, Phiip Swisher, and Xichen Jiang Abstract A quickest change detection (QCD) agorithm is appied to the probem
More informationMARKOV CHAINS AND MARKOV DECISION THEORY. Contents
MARKOV CHAINS AND MARKOV DECISION THEORY ARINDRIMA DATTA Abstract. In this paper, we begin with a forma introduction to probabiity and expain the concept of random variabes and stochastic processes. After
More informationSoft Clustering on Graphs
Soft Custering on Graphs Kai Yu 1, Shipeng Yu 2, Voker Tresp 1 1 Siemens AG, Corporate Technoogy 2 Institute for Computer Science, University of Munich kai.yu@siemens.com, voker.tresp@siemens.com spyu@dbs.informatik.uni-muenchen.de
More informationAlberto Maydeu Olivares Instituto de Empresa Marketing Dept. C/Maria de Molina Madrid Spain
CORRECTIONS TO CLASSICAL PROCEDURES FOR ESTIMATING THURSTONE S CASE V MODEL FOR RANKING DATA Aberto Maydeu Oivares Instituto de Empresa Marketing Dept. C/Maria de Moina -5 28006 Madrid Spain Aberto.Maydeu@ie.edu
More informationData Mining Technology for Failure Prognostic of Avionics
IEEE Transactions on Aerospace and Eectronic Systems. Voume 38, #, pp.388-403, 00. Data Mining Technoogy for Faiure Prognostic of Avionics V.A. Skormin, Binghamton University, Binghamton, NY, 1390, USA
More informationA Novel Learning Method for Elman Neural Network Using Local Search
Neura Information Processing Letters and Reviews Vo. 11, No. 8, August 2007 LETTER A Nove Learning Method for Eman Neura Networ Using Loca Search Facuty of Engineering, Toyama University, Gofuu 3190 Toyama
More informationPhysics 566: Quantum Optics Quantization of the Electromagnetic Field
Physics 566: Quantum Optics Quantization of the Eectromagnetic Fied Maxwe's Equations and Gauge invariance In ecture we earned how to quantize a one dimensiona scaar fied corresponding to vibrations on
More informationStochastic Variational Inference with Gradient Linearization
Stochastic Variationa Inference with Gradient Linearization Suppementa Materia Tobias Pötz * Anne S Wannenwetsch Stefan Roth Department of Computer Science, TU Darmstadt Preface In this suppementa materia,
More informationCentralized Coded Caching of Correlated Contents
Centraized Coded Caching of Correated Contents Qianqian Yang and Deniz Gündüz Information Processing and Communications Lab Department of Eectrica and Eectronic Engineering Imperia Coege London arxiv:1711.03798v1
More information<C 2 2. λ 2 l. λ 1 l 1 < C 1
Teecommunication Network Contro and Management (EE E694) Prof. A. A. Lazar Notes for the ecture of 7/Feb/95 by Huayan Wang (this document was ast LaT E X-ed on May 9,995) Queueing Primer for Muticass Optima
More informationTranslation Microscopy (TRAM) for super-resolution imaging.
ransation Microscopy (RAM) for super-resoution imaging. Zhen Qiu* 1,,3, Rhodri S Wison* 1,3, Yuewei Liu 1,3, Aison Dun 1,3, Rebecca S Saeeb 1,3, Dongsheng Liu 4, Coin Ricman 1,3, Rory R Duncan 1,3,5, Weiping
More informationNEW DEVELOPMENT OF OPTIMAL COMPUTING BUDGET ALLOCATION FOR DISCRETE EVENT SIMULATION
NEW DEVELOPMENT OF OPTIMAL COMPUTING BUDGET ALLOCATION FOR DISCRETE EVENT SIMULATION Hsiao-Chang Chen Dept. of Systems Engineering University of Pennsyvania Phiadephia, PA 904-635, U.S.A. Chun-Hung Chen
More informationSome Measures for Asymmetry of Distributions
Some Measures for Asymmetry of Distributions Georgi N. Boshnakov First version: 31 January 2006 Research Report No. 5, 2006, Probabiity and Statistics Group Schoo of Mathematics, The University of Manchester
More informationTiming Attacks on Cognitive Authentication Schemes
Timing Attacks on Cognitive Authentication Schemes Mario Čagaj, Toni Perković, Member, IEEE, Marin Bugarić Department of Eectrica Engineering, FESB, University of Spit, Croatia {mcagaj, toperkov, mbugaric}@fesb.hr
More informationA Comparison Study of the Test for Right Censored and Grouped Data
Communications for Statistica Appications and Methods 2015, Vo. 22, No. 4, 313 320 DOI: http://dx.doi.org/10.5351/csam.2015.22.4.313 Print ISSN 2287-7843 / Onine ISSN 2383-4757 A Comparison Study of the
More informationA Robust Voice Activity Detection based on Noise Eigenspace Projection
A Robust Voice Activity Detection based on Noise Eigenspace Projection Dongwen Ying 1, Yu Shi 2, Frank Soong 2, Jianwu Dang 1, and Xugang Lu 1 1 Japan Advanced Institute of Science and Technoogy, Nomi
More informationHaar Decomposition and Reconstruction Algorithms
Jim Lambers MAT 773 Fa Semester 018-19 Lecture 15 and 16 Notes These notes correspond to Sections 4.3 and 4.4 in the text. Haar Decomposition and Reconstruction Agorithms Decomposition Suppose we approximate
More informationOptimal Control of Assembly Systems with Multiple Stages and Multiple Demand Classes 1
Optima Contro of Assemby Systems with Mutipe Stages and Mutipe Demand Casses Saif Benjaafar Mohsen EHafsi 2 Chung-Yee Lee 3 Weihua Zhou 3 Industria & Systems Engineering, Department of Mechanica Engineering,
More informationNumerical solution of one dimensional contaminant transport equation with variable coefficient (temporal) by using Haar wavelet
Goba Journa of Pure and Appied Mathematics. ISSN 973-1768 Voume 1, Number (16), pp. 183-19 Research India Pubications http://www.ripubication.com Numerica soution of one dimensiona contaminant transport
More informationEfficient Generation of Random Bits from Finite State Markov Chains
Efficient Generation of Random Bits from Finite State Markov Chains Hongchao Zhou and Jehoshua Bruck, Feow, IEEE Abstract The probem of random number generation from an uncorreated random source (of unknown
More informationWidth of Percolation Transition in Complex Networks
APS/23-QED Width of Percoation Transition in Compex Networs Tomer Kaisy, and Reuven Cohen 2 Minerva Center and Department of Physics, Bar-Ian University, 52900 Ramat-Gan, Israe 2 Department of Computer
More informationSource and Relay Matrices Optimization for Multiuser Multi-Hop MIMO Relay Systems
Source and Reay Matrices Optimization for Mutiuser Muti-Hop MIMO Reay Systems Yue Rong Department of Eectrica and Computer Engineering, Curtin University, Bentey, WA 6102, Austraia Abstract In this paper,
More informationIE 361 Exam 1. b) Give *&% confidence limits for the bias of this viscometer. (No need to simplify.)
October 9, 00 IE 6 Exam Prof. Vardeman. The viscosity of paint is measured with a "viscometer" in units of "Krebs." First, a standard iquid of "known" viscosity *# Krebs is tested with a company viscometer
More informationInductive Bias: How to generalize on novel data. CS Inductive Bias 1
Inductive Bias: How to generaize on nove data CS 478 - Inductive Bias 1 Overfitting Noise vs. Exceptions CS 478 - Inductive Bias 2 Non-Linear Tasks Linear Regression wi not generaize we to the task beow
More informationTarget Location Estimation in Wireless Sensor Networks Using Binary Data
Target Location stimation in Wireess Sensor Networks Using Binary Data Ruixin Niu and Pramod K. Varshney Department of ectrica ngineering and Computer Science Link Ha Syracuse University Syracuse, NY 344
More information(This is a sample cover image for this issue. The actual cover is not yet available at this time.)
(This is a sampe cover image for this issue The actua cover is not yet avaiabe at this time) This artice appeared in a journa pubished by Esevier The attached copy is furnished to the author for interna
More informationCryptanalysis of PKP: A New Approach
Cryptanaysis of PKP: A New Approach Éiane Jaumes and Antoine Joux DCSSI 18, rue du Dr. Zamenhoff F-92131 Issy-es-Mx Cedex France eiane.jaumes@wanadoo.fr Antoine.Joux@ens.fr Abstract. Quite recenty, in
More informationOptimum Design Method of Viscous Dampers in Building Frames Using Calibration Model
The 4 th Word Conference on Earthquake Engineering October -7, 8, Beijing, China Optimum Design Method of iscous Dampers in Buiding Frames sing Caibration Mode M. Yamakawa, Y. Nagano, Y. ee 3, K. etani
More informationMode in Output Participation Factors for Linear Systems
2010 American ontro onference Marriott Waterfront, Batimore, MD, USA June 30-Juy 02, 2010 WeB05.5 Mode in Output Participation Factors for Linear Systems Li Sheng, yad H. Abed, Munther A. Hassouneh, Huizhong
More informationWAVELET FREQUENCY DOMAIN APPROACH FOR TIME-SERIES MODELING
1. Introduction WAVELET FREQUENCY DOMAIN APPROACH FOR TIME-SERIES MODELING Ranit Kumar Pau Indian Agricutura Statistics Research Institute Library Avenue, New Dehi 11001 ranitstat@iasri.res.in Autoregressive
More informationSteepest Descent Adaptation of Min-Max Fuzzy If-Then Rules 1
Steepest Descent Adaptation of Min-Max Fuzzy If-Then Rues 1 R.J. Marks II, S. Oh, P. Arabshahi Λ, T.P. Caude, J.J. Choi, B.G. Song Λ Λ Dept. of Eectrica Engineering Boeing Computer Services University
More informationc 2016 Georgios Rovatsos
c 2016 Georgios Rovatsos QUICKEST CHANGE DETECTION WITH APPLICATIONS TO LINE OUTAGE DETECTION BY GEORGIOS ROVATSOS THESIS Submitted in partia fufiment of the requirements for the degree of Master of Science
More informationOn the Performance of Mismatched Data Detection in Large MIMO Systems
On the Performance of Mismatched Data Detection in Large MIMO Systems Chares Jeon, Arian Maei, and Christoph Studer arxiv:1605.034v [cs.it] Jun 016 Abstract We investigate the performance of mismatched
More informationEffective Appearance Model and Similarity Measure for Particle Filtering and Visual Tracking
Effective Appearance Mode and Simiarity Measure for Partice Fitering and Visua Tracking Hanzi Wang David Suter and Konrad Schinder Institute for Vision Systems Engineering Department of Eectrica and Computer
More information17 Lecture 17: Recombination and Dark Matter Production
PYS 652: Astrophysics 88 17 Lecture 17: Recombination and Dark Matter Production New ideas pass through three periods: It can t be done. It probaby can be done, but it s not worth doing. I knew it was
More informationThe Binary Space Partitioning-Tree Process Supplementary Material
The inary Space Partitioning-Tree Process Suppementary Materia Xuhui Fan in Li Scott. Sisson Schoo of omputer Science Fudan University ibin@fudan.edu.cn Schoo of Mathematics and Statistics University of
More informationCombining reaction kinetics to the multi-phase Gibbs energy calculation
7 th European Symposium on Computer Aided Process Engineering ESCAPE7 V. Pesu and P.S. Agachi (Editors) 2007 Esevier B.V. A rights reserved. Combining reaction inetics to the muti-phase Gibbs energy cacuation
More informationWavelet Methods for Time Series Analysis. Wavelet-Based Signal Estimation: I. Part VIII: Wavelet-Based Signal Extraction and Denoising
Waveet Methods for Time Series Anaysis Part VIII: Waveet-Based Signa Extraction and Denoising overview of key ideas behind waveet-based approach description of four basic modes for signa estimation discussion
More informationLimits on Support Recovery with Probabilistic Models: An Information-Theoretic Framework
Limits on Support Recovery with Probabiistic Modes: An Information-Theoretic Framewor Jonathan Scarett and Voan Cevher arxiv:5.744v3 cs.it 3 Aug 6 Abstract The support recovery probem consists of determining
More informationOnline Appendices for The Economics of Nationalism (Xiaohuan Lan and Ben Li)
Onine Appendices for The Economics of Nationaism Xiaohuan Lan and Ben Li) A. Derivation of inequaities 9) and 10) Consider Home without oss of generaity. Denote gobaized and ungobaized by g and ng, respectivey.
More informationStudy on Fusion Algorithm of Multi-source Image Based on Sensor and Computer Image Processing Technology
Sensors & Transducers 3 by IFSA http://www.sensorsporta.com Study on Fusion Agorithm of Muti-source Image Based on Sensor and Computer Image Processing Technoogy Yao NAN, Wang KAISHENG, 3 Yu JIN The Information
More informationDecentralized Event-Driven Algorithms for Multi-Agent Persistent Monitoring Tasks
2017 IEEE 56th Annua Conference on Decision and Contro (CDC) December 12-15, 2017, Mebourne, Austraia Decentraized Event-Driven Agorithms for Muti-Agent Persistent Monitoring Tasks Nan Zhou 1, Christos
More informationFast methods for spatially correlated multilevel functional data
Biostatistics Advance Access pubished January 19, 2010 Biostatistics (2010), 0, 0, pp. 1 18 doi:10.1093/biostatistics/xp058 Fast methods for spatiay correated mutieve functiona data ANA-MARIA STAICU Department
More informationInternational Journal of Mass Spectrometry
Internationa Journa of Mass Spectrometry 280 (2009) 179 183 Contents ists avaiabe at ScienceDirect Internationa Journa of Mass Spectrometry journa homepage: www.esevier.com/ocate/ijms Stark mixing by ion-rydberg
More informationarxiv: v1 [math.ca] 6 Mar 2017
Indefinite Integras of Spherica Besse Functions MIT-CTP/487 arxiv:703.0648v [math.ca] 6 Mar 07 Joyon K. Boomfied,, Stephen H. P. Face,, and Zander Moss, Center for Theoretica Physics, Laboratory for Nucear
More informationMore Scattering: the Partial Wave Expansion
More Scattering: the Partia Wave Expansion Michae Fower /7/8 Pane Waves and Partia Waves We are considering the soution to Schrödinger s equation for scattering of an incoming pane wave in the z-direction
More informationRadar/ESM Tracking of Constant Velocity Target : Comparison of Batch (MLE) and EKF Performance
adar/ racing of Constant Veocity arget : Comparison of Batch (LE) and EKF Performance I. Leibowicz homson-csf Deteis/IISA La cef de Saint-Pierre 1 Bd Jean ouin 7885 Eancourt Cede France Isabee.Leibowicz
More informationFast Blind Recognition of Channel Codes
Fast Bind Recognition of Channe Codes Reza Moosavi and Erik G. Larsson Linköping University Post Print N.B.: When citing this work, cite the origina artice. 213 IEEE. Persona use of this materia is permitted.
More informationReliability: Theory & Applications No.3, September 2006
REDUNDANCY AND RENEWAL OF SERVERS IN OPENED QUEUING NETWORKS G. Sh. Tsitsiashvii M.A. Osipova Vadivosto, Russia 1 An opened queuing networ with a redundancy and a renewa of servers is considered. To cacuate
More informationIntroduction to Simulation - Lecture 14. Multistep Methods II. Jacob White. Thanks to Deepak Ramaswamy, Michal Rewienski, and Karen Veroy
Introduction to Simuation - Lecture 14 Mutistep Methods II Jacob White Thans to Deepa Ramaswamy, Micha Rewiensi, and Karen Veroy Outine Sma Timestep issues for Mutistep Methods Reminder about LTE minimization
More informationBICM Performance Improvement via Online LLR Optimization
BICM Performance Improvement via Onine LLR Optimization Jinhong Wu, Mostafa E-Khamy, Jungwon Lee and Inyup Kang Samsung Mobie Soutions Lab San Diego, USA 92121 Emai: {Jinhong.W, Mostafa.E, Jungwon2.Lee,
More informationConverting Z-number to Fuzzy Number using. Fuzzy Expected Value
ISSN 1746-7659, Engand, UK Journa of Information and Computing Science Vo. 1, No. 4, 017, pp.91-303 Converting Z-number to Fuzzy Number using Fuzzy Expected Vaue Mahdieh Akhbari * Department of Industria
More informationsensors Beamforming Based Full-Duplex for Millimeter-Wave Communication Article
sensors Artice Beamforming Based Fu-Dupex for Miimeter-Wave Communication Xiao Liu 1,2,3, Zhenyu Xiao 1,2,3, *, Lin Bai 1,2,3, Jinho Choi 4, Pengfei Xia 5 and Xiang-Gen Xia 6 1 Schoo of Eectronic and Information
More informationhttps://doi.org/ /epjconf/
HOW TO APPLY THE OPTIMAL ESTIMATION METHOD TO YOUR LIDAR MEASUREMENTS FOR IMPROVED RETRIEVALS OF TEMPERATURE AND COMPOSITION R. J. Sica 1,2,*, A. Haefee 2,1, A. Jaai 1, S. Gamage 1 and G. Farhani 1 1 Department
More informationCategories and Subject Descriptors B.7.2 [Integrated Circuits]: Design Aids Verification. General Terms Algorithms
5. oward Eicient arge-scae Perormance odeing o Integrated Circuits via uti-ode/uti-corner Sparse Regression Wangyang Zhang entor Graphics Corporation Ridder Park Drive San Jose, CA 953 wangyan@ece.cmu.edu
More informationBALANCING REGULAR MATRIX PENCILS
BALANCING REGULAR MATRIX PENCILS DAMIEN LEMONNIER AND PAUL VAN DOOREN Abstract. In this paper we present a new diagona baancing technique for reguar matrix pencis λb A, which aims at reducing the sensitivity
More informationAlgorithms to solve massively under-defined systems of multivariate quadratic equations
Agorithms to sove massivey under-defined systems of mutivariate quadratic equations Yasufumi Hashimoto Abstract It is we known that the probem to sove a set of randomy chosen mutivariate quadratic equations
More informationA GENERAL METHOD FOR EVALUATING OUTAGE PROBABILITIES USING PADÉ APPROXIMATIONS
A GENERAL METHOD FOR EVALUATING OUTAGE PROBABILITIES USING PADÉ APPROXIMATIONS Jack W. Stokes, Microsoft Corporation One Microsoft Way, Redmond, WA 9852, jstokes@microsoft.com James A. Ritcey, University
More informationTRAVEL TIME ESTIMATION FOR URBAN ROAD NETWORKS USING LOW FREQUENCY PROBE VEHICLE DATA
TRAVEL TIME ESTIMATIO FOR URBA ROAD ETWORKS USIG LOW FREQUECY PROBE VEHICLE DATA Erik Jeneius Corresponding author KTH Roya Institute of Technoogy Department of Transport Science Emai: erik.jeneius@abe.kth.se
More informationTELECOMMUNICATION DATA FORECASTING BASED ON ARIMA MODEL
TEECOMMUNICATION DATA FORECASTING BASED ON ARIMA MODE Anjuman Akbar Muani 1, Prof. Sachin Muraraka 2, Prof. K. Sujatha 3 1Student ME E&TC,Shree Ramchandra Coege of Engineering, onikand,pune,maharashtra
More informationMelodic contour estimation with B-spline models using a MDL criterion
Meodic contour estimation with B-spine modes using a MDL criterion Damien Loive, Ney Barbot, Oivier Boeffard IRISA / University of Rennes 1 - ENSSAT 6 rue de Kerampont, B.P. 80518, F-305 Lannion Cedex
More informationA Solution to the 4-bit Parity Problem with a Single Quaternary Neuron
Neura Information Processing - Letters and Reviews Vo. 5, No. 2, November 2004 LETTER A Soution to the 4-bit Parity Probem with a Singe Quaternary Neuron Tohru Nitta Nationa Institute of Advanced Industria
More informationSupporting Information for Suppressing Klein tunneling in graphene using a one-dimensional array of localized scatterers
Supporting Information for Suppressing Kein tunneing in graphene using a one-dimensiona array of ocaized scatterers Jamie D Was, and Danie Hadad Department of Chemistry, University of Miami, Cora Gabes,
More information