Aalborg Universitet. Published in: IEEE International Conference on Acoustics, Speech, and Signal Processing. Creative Commons License Unspecified

Size: px
Start display at page:

Download "Aalborg Universitet. Published in: IEEE International Conference on Acoustics, Speech, and Signal Processing. Creative Commons License Unspecified"

Transcription

1 Aalborg Universitet Model-based Noise PSD Estimation from Speech in Non-stationary Noise Nielsen, Jesper jær; avalekalam, Mathew Shaji; Christensen, Mads Græsbøll; Boldt, Jesper Bünsow Published in: IEEE International Conference on Acoustics, Speech, and Signal Processing Creative Commons License Unspecified Publication date: 2018 Document Version Other version Link to publication from Aalborg University Citation for published version (APA): Nielsen, J.., avalekalam, M. S., Christensen, M. G., & Boldt, J. B. (2018). Model-based Noise PSD Estimation from Speech in Non-stationary Noise. In IEEE International Conference on Acoustics, Speech, and Signal Processing Calgary, Canada: IEEE. I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.? Users may download and print one copy of any publication from the public portal for the purpose of private study or research.? You may not further distribute the material or use it for any profit-making activity or commercial gain? You may freely distribute the URL identifying the publication in the public portal? Take down policy If you believe that this document breaches copyright please contact us at vbn@aub.aau.dk providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from vbn.aau.dk on: december 07, 2018

2 Variational Bayesian Inference for Model-based Noise PSD Estimation Jesper jær Nielsen 1, Mathew Shaji avalekam 1, Mads Græsbøll Christensen 1, and Jesper Boldt 2 1 Audio Analysis Lab, CREATE, Aalborg University 2 GN ReSound Last compiled: April 17, Introduction This document gives the details of the variational Bayesian (VB) noise power spectral density (PSD) estimator proposed in our ICASSP-paper 2. In the first section of the document, a general description of the VB approach is given to joint parameter estimation and model comparison. If the reader is familiar with the VB approach, this section can be skipped. In the second section, the VB algorithm is applied to the model described in 2. We strongly encourage the reader to read the ICASSP paper before continuing reading this document. 2 A VB tutorial Suppose we wish to compute the posterior distribution p(θ x) = p(θ x, M k )p(m k x), (1) for the model parameters θ, but cannot do this analytically. Sometimes, we can simplify the problem considerably if we introduce latent variables s. An example of this is for Gaussian mixture models where the introduction of latent variables allows us to use the EM-algorithm to produce a solution. If we let z = s T and the model parameters is p(z x) = θ T T, the joint posterior over the latent parameters p(z x, M k )p(m k x). (2) 1

3 Often, this posterior cannot be computed analytically, so we assume that it factorises for each model M k as p(z x, M k ) q k (z) = M i=1 q ik (z i ) (3) where we have divided z into M non-overlapping subgroups. In the sequel, we will use the more compact notation q ik = q ik (z i ). Also note that even though we do not assume any factorisation for p(m k x), we will only obtain an approximation q(m k ) of it since we use q k (z) instead of p(z x, M k ). Thus, we have that p(z, M k x) q(z, M k ) = q(m k )q k = q(m k ) M i=1 q ik. (4) Thus, we have now reformulated the problem from finding p(z, M k x) to that of finding the different factors q ik and q(m k ). If we select the factorisation in a clever way, it is much easier to find these factors (using VB) than it is to find the exact posterior p(z, M k x). The art of applying the VB framework is, therefore, to elicit a signal model (i.e., an observation model p(x z, M k ) and prior distributions p(z M k ) and p(m k )) and a factorisation so that the various factors can easily be computed while still being a good approximation to the exact posterior. In the sequel, we assume that we have elicited these and focus on describing how the various factors of the elicited factorisation are computed from the signal model. For any pdf q, we have that the log of the model evidence can be written as 1, p. 473 ln p(x) = L(q) q(m k ) q k ln p(z, M k x) dz (5) q k q(m k ) where L(q) = q(m k ) q k ln p(z, x, M k) dz. (6) q k q(m k ) It turns out that maximising the lower bound L(q) w.r.t. q produces the best approximation to the posterior p(z, M k x). To perform this maximisation, the lower bound is first rewritten as L(q) = q(m k )L k (q k ) + q(m k ) ln p(m k) q(m k ) (7) 2

4 where L k (q k ) = = = = = q k ln p(z, x M k) dz q k (8) q k ln p(z, x M k )dz q k ln q k dz (9) ( ) M ( ) q ik dz i dz j ln q ik q ik dz i i=1 i=1 q jk ln p(z, x M k ) (10) ( ) M q jk ln p(z, x M k ) q ik dz i dz j q ik ln q ik dz i (11) i=1 ( ) q jk ln p(z, x M k ) q ik dz i ln q jk dz j q ik ln q ik dz i. (12) Given the data x and the pdfs q ik for i = j, the integral in the bracket is a function of z j. However, this function does not necessarily integrate to one, but such a function can easily be defined as ln p(z j x, M k ) = E ln p(z, x M k ) ln Z jk (13) where E ln p(z, x M k ) and Z jk are the expectation operator w.r.t. to q ik and the normalisation constant so that p(z j x, M k ) integrates to one, respectively. These are given by ( ) E ln p(z, x M k ) = ln p(z, x M k ) q ik dz i (14) Z jk = exp { E ln p(z, x M k ) } dz j. (15) Consequently, we can now write L k (q k ) as L k (q k ) = q jk ln p(zj x, M k ) + ln Z jk ln q jk dzj q ik ln q ik dz i (16) = L(q jk p(z j x, M k )) + ln Z jk + Hq ik (17) where L(q jk p(z j x, M k )) and Hq ik are the ullback-leibler (L) divergence and the entropy, respectively. We have now written the lower bound L(q) in a form which allows us to maximise it w.r.t. to q jk. Since the L divergence is the only term which depends on the functional q jk, we maximise the lower bound by minimising the L divergence. Fortunately, this is easy since the minimum of the L divergence is zero, and this minimum value is obtained if and only if q jk = p(z j x, M k ) (18) 3

5 and the solution to the optimisation problem, therefore, is that ln q jk = E ln p(z, x M k ) + const. (19) The above solution is not a closed-form solution since the expectation operator depends on {q ik } which are also unknown. However, if we optimise for the individual q jk s iteratively, we are guaranteed to converge to a solution 1, p The VB Lower Bound For every model order, we can keep an eye on the convergence by monitoring the lower bound L k (q k ). For the optimal form of q jk, the L divergence vanishes, and the lower bound is given by L k (q k ) = ln Z jk + Hq ik. (20) Since this should hold for all j s, we can stop the algorithm when L k (q k ) is nearly the same for all j s. An alternative formulation for the lower bound can also be constructed directly from the definition of L k (q k ) as L k (q k ) = q k ln p(z, x M k )dz q k ln q k dz (21) = 2.2 The VB Model Comparison q k ln p(x z, M k )dz + q k ln p(z M k )dz + = E qk ln p(x z, M k ) + E qk ln p(z M k ) + M i=1 M i=1 Hq ik (22) Hq ik. (23) We can also do model comparison in the VB framework. To do this, we rewrite L(q) as L(q) = = q(m k )L k (q k ) + q(m k ) ln p(m k) q(m k ) (24) q(m k ) ln (p(m k ) exp(l k (q k ))) ln q(m k ). (25) Given the data x and the joint pdf q k, the first term in the bracket is a function of the model M k. This function, however, does not necessarily integrate to one, but such a function can easily be defined as ln p(m k x) = L k (q k ) + ln p(m k ) ln Z k (26) where Z k is a normalisation constant given by Z k = p(m k ) exp L k (q k ). (27) 4

6 Using these definitions, we can write L(q) as L(q) = L(q(M k ) p(m k x)) + ln Z k (28) which is clearly maximised for or, equivalently, q(m k ) = p(m k x) (29) ln q(m k ) = L k (q k ) + ln p(m k ) + const. (30) = ln Z jk + Hq ik + ln p(m k ) + const. (31) for any j. 3 Noise PSD Estimation Using VB We will now apply the VB framework to solve the problem of computing the posterior distribution on the excitaiton noise variances from a mixture of two periodic AR processes. That is, we observe y = s + e (32) where p(e σ 2 e, M k ) = N (0, σ 2 e R e (b k )) (33) p(s σ 2 s, M k ) = N (0, σ 2 s R s (a k )) (34) R s (a k ) = N 1 FD s (a k )F H (35) R e (b k ) = N 1 FD e (b k )F H (36) F nl = exp(j2π(n 1)(l 1)/N), n, l = 1,..., N (37) Λ s (a k ) = diag(f H a T k 0 T ) (38) T) Λ e (b k ) = diag(f H bk 0 T (39) ( 1 D s (a k ) = Λs H (a k )Λ s (a k )) (40) ( 1 D e (b k ) = Λe H (b k )Λ e (b k )). (41) Moreover, we assume that the prior for the excitation noise variances are given by p(σ 2 e M k ) = Inv-G(α e,k, β e,k ) (42) p(σ 2 s M k ) = Inv-G(α s,k, β s,k ) (43) 5

7 whose hyperparameters we initially assume known. The latens variable of our model is s whereas the model parameters are σ 2 s and σ 2 e. Note that the AR parameters are given by the model. From the above model, it also follows that p(y s, σ 2 e, M k ) = N (s, σ 2 e R e (b k )) (44) so that joint distribution over the observations, the latent variables, and the model parameters factorise as p(y, s, σ 2 s, σ 2 e M k ) = p(y s, σ 2 e, M k )p(s σ 2 s, M k )p(σ 2 s M k )p(σ 2 e M k ). (45) In our VB algorithm, we seek to compute and approximation of the joint posterior p(s, σ 2 s, σ 2 e M k ) using the factorisation p(s, σ 2 s, σ 2 e M k ) q k (s, σ 2 s, σ 2 e ) = q 1k (s)q 2k (σ 2 s )q 3k (σ 2 e ) (46) where the factors can be viewed as approximations to the marginal posteriors with q 1k (s) p(s y, M k ) (47) q 2k (σ 2 s ) p(σ 2 s y, M k ) (48) q 3k (σ 2 e ) p(σ 2 e y, M k ). (49) We also wish to compute an approximation q(m k ) of the posterior pmf for the models. Before we can do that, however, we first have to find the functional expression of the factors q ik, their entropies Hq ik, and the normalisation constants Z ik. 3.1 Functional expressions for the factor q 1k (s) According to the above tutorial, we have that ln q 1k (s) = E i =1 ln p(z, y M k ) + const. (50) Inserting the expression for p(z, y M k ) in the expectation operator, we obtain E i =1 ln p(z, y M k ) = E i =1 ln p(y s, σ 2 e, M k ) + E i =1 ln p(s σ 2 s, M k ) + E i =1 ln p(σ 2 s M k ) + E i =1 ln p(σ 2 e M k ). (51) Of the four terms, the first two terms depend on s whereas the last two do not. However, we still have to compute the last two terms in order to compute the normalisation constant Z 1k. 6

8 Since ln p(y s, σ 2 e, M k ) = (N/2) ln(2πσ 2 e ) (1/2) ln det(r e (b k )) 1 2σe 2 (y s) T Re 1 (b k )(y s) (52) ln p(s σs 2, M k ) = (N/2) ln(2πσs 2 ) (1/2) ln det(r s (a k )) 1 2σs 2 s T Rs 1 (a k )s (53) ln p(σ 2 s M k ) = α s,k ln β s,k ln Γ(α s,k ) (α s,k + 1) ln σ 2 s β s,k σ 2 s ln p(σ 2 e M k ) = α e,k ln β e,k ln Γ(α e,k ) (α e,k + 1) ln σ 2 e β e,k σ 2 e (54), (55) we have for the four terms that E i =1 ln p(y s, σe 2, M k ) = q 3k (σe 2 ) ln p(y s, σe 2, M k )dσe 2 (56) = (N/2) ln(2π) (N/2) E q3k ln σ 2 e (1/2) ln det(r e (b k )) (1/2)(y s) T Re 1 (b k )(y s) E q3k σe 2 (57) = ln N (s, E 1 q 3k σe 2 R e (b k )) (N/2) { ln E q3k σe 2 + E q3k ln σe 2 } (58) E i =1 ln p(s σs 2, M k ) = q 2k (σs 2 ) ln p(s σs 2, M k )dσs 2 (59) = (N/2) ln(2π) (N/2) E q2k ln σ 2 s (1/2) ln det(r s (a k )) (1/2)s T Rs 1 (a k )s E q2k σs 2 (60) = ln N (0, E 1 q 2k σs 2 R s (a k )) (N/2) { ln E q2k σ 2 s + E q2k ln σ 2 s } (61) E i =1 ln p(σ 2 s M k ) = α s,k ln β s,k ln Γ(α s,k ) (α s,k + 1) E q2k ln σ 2 s β s,k E q2k σ 2 s (62) E i =1 ln p(σ 2 e M k ) = α e,k ln β e,k ln Γ(α e,k ) (α e,k + 1) E q3k ln σ 2 e β e,k E q3k σ 2 e (63) By only retaining the terms from these four expressions which depend on s, we obtain that q 1k (s) N (s, E 1 from which we can derive that q 3k σe 2 R e (b k ))N (0, E 1 q 2k σ 2 s R s (a k )) (64) q 1k (s) = N (ŝ, Σŝ) (65) 1 Σŝ = E q2k σs 2 Rs 1 (a k ) + E q3k σe 2 Re 1 (b k ) (66) ŝ = Σŝ E q3k σe 2 Re 1 (b k )y. (67) 7

9 and that N (s, E 1 q 3k σe 2 R e (b k ))N (0, E 1 q 2k σ 2 s R s (a k ))ds = N (0, E 1 q 2k σs 2 R s (a k ) + E 1 q 3k σ 2 e R e (b k )) (68) by using standard Bayesian inference. Thus, the log-normalisation factor ln Z 1k is given by ln Z 1 = ln exp { E i =1ln p(z, y M k ) } ds (69) = ln N (0, E 1 q 2k σs 2 R s (a k ) + E 1 q 3k σ 2 e R e (b k )) (N/2) { ln E q2k σs 2 + E q2k ln σs 2 + ln E q3k σe 2 + E q3k ln σe 2 } (70) E i =1 ln p(σs 2 M k ) + E i =1 ln p(σe 2 M k ). (71) Finally, since q 1k (s) is a multivariate Gaussian distribution, its entropy is given by Hq 1k = (N/2) ln(2π exp(1)) + (1/2) ln det(σŝ). (72) 3.2 Functional expressions for the factor q 2k (σ 2 s ) According to the above tutorial, we have that ln q 2k (σ 2 s ) = E i =2 ln p(z, y M k ) + const. (73) Inserting the expression for p(z, y M k ) in the expectation operator, we obtain E i =2 ln p(z, y M k ) = E i =2 ln p(y s, σ 2 e, M k ) + E i =2 ln p(s σ 2 s, M k ) + E i =2 ln p(σ 2 s M k ) + E i =2 ln p(σ 2 e M k ). (74) Of the four terms, the second and third terms depend on σs 2 whereas the first and fourth do not. However, we still have to compute all the terms to obtain the normalisation constant Z 2k. For the four terms, we have that E i =2 ln p(y s, σe 2, M k ) = q 1k (s)q 3k (σe 2 ) ln p(y s, σe 2, M k )dsdσe 2 (75) = (N/2) ln(2π) (N/2) E q3k ln σe 2 (1/2) ln det(r e (b k )) (1/2) E q1k (y s) T Re 1 (b k )(y s) E q3k σe 2 (76) E i =2 ln p(s σs 2, M k ) = q 1k (s) ln p(s σs 2, M k )ds (77) = (N/2) ln(2π) (N/2) ln σs 2 (1/2) ln det(r s (a k )) 1 2σs 2 E q1k s T Rs 1 (a k )s E i =2 ln p(σs 2 M k ) = α s,k ln β s,k ln Γ(α s,k ) (α s,k + 1) ln σs 2 β s,k σs 2 (79) E i =2 ln p(σe 2 M k ) = E i =1 ln p(σe 2 M k ) (80) (78) 8

10 From this, we have that ln q 2k (σs 2 ) = (N/2 + α s,k + 1) ln σs 2 (1/2) E q 1k s T Rs 1 (a k )s + β s,k σs 2 + const. (81) where all the terms independent of σ 2 s are absorbed in the constant. Thus, where q 2k (σ 2 s ) = Inv-G(a s,k, b s,k ) (82) a s,k = N/2 + α s,k (83) b s,k = (1/2) E q1k s T Rs 1 (a k )s + β s,k. (84) The log-normalisation factor ln Z 2 can now also be found to ln Z 2 = ln = ln exp { E i =2ln p(z, y M k ) } dσs 2 ( (σs 2 ) (as,k+1) exp ) dσ 2 s + α s,k ln β s,k ln Γ(α s,k ) b s,k σs 2 (N/2) ln(2π) (N/2) ln σs 2 (1/2) ln det(r s (a k )) (85) + E i =2 ln p(y s, σ 2 e, M k ) + E i =2 ln p(σ 2 e M k ) (86) = a s,k ln b s,k + ln Γ(a s,k ) + α s,k ln β s,k ln Γ(α s,k ) (N/2) ln(2π) (N/2) ln σ 2 s (1/2) ln det(r s (a k )) + E i =2 ln p(y s, σ 2 e, M k ) + E i =2 ln p(σ 2 e M k ) (87) Finally, since q 2k (σ 2 s ) is an inverse Gamma distribution, its entropy is given by where Ψ( ) is the digamma function. Hq 2k = a s,k + ln(b s,k Γ(a s,k )) (1 + a s,k )Ψ(a s,k ) (88) 3.3 Functional expressions for the factor q 3k (σ 2 e ) According to the above tutorial, we have that ln q 3k (σ 2 e ) = E i =3 ln p(z, y M k ) + const. (89) Inserting the expression for p(z, y M k ) in the expectation operator, we obtain E i =3 ln p(z, y M k ) = E i =3 ln p(y s, σ 2 e, M k ) + E i =3 ln p(s σ 2 s, M k ) + E i =3 ln p(σ 2 s M k ) + E i =3 ln p(σ 2 e M k ). (90) 9

11 Of the four terms, the first and the fourth terms depend on σe 2 whereas the second and third do not. However, we still have to compute all the terms to obtain the normalisation constant Z 3k. For the four terms, we have that E i =3 ln p(y s, σe 2, M k ) = q 1k (s) ln p(y s, σe 2, M k )ds (91) = (N/2) ln(2π) (N/2) ln σe 2 (1/2) ln det(r e (b k )) 1 2σe 2 E q1k (y s) T Re 1 (b k )(y s) (92) E i =3 ln p(s σs 2, M k ) = q 1k (s)q 2k (σs 2 ) ln p(s σs 2, M k )dsdσs 2 (93) = (N/2) ln(2π) (N/2) E q2k ln σs 2 (1/2) ln det(r s (a k )) (1/2) E q1k s T Rs 1 (a k )s E q2k σs 2 (94) E i =3 ln p(σ 2 s M k ) = E i =1 ln p(σ 2 s M k ) (95) E i =3 ln p(σ 2 e M k ) = α e,k ln β e,k ln Γ(α e,k ) (α e,k + 1) ln σ 2 e β e,k σ 2 e (96) (97) From this, we have that ln q 3k (σ 2 e ) = (N/2 + α e,k + 1) ln σ 2 e (1/2) E q1k (y s) T Re 1 (b k )(y s) + β e,k σe 2 + const. (98) where all the terms independent of σ 2 e are absorbed in the constant. Thus, where q 3k (σ 2 s ) = Inv-G(a e,k, b e,k ) (99) a e,k = N/2 + α e,k (100) b e,k = (1/2) E q1k (y s) T Re 1 (b k )(y s) + β e,k. (101) 10

12 The log-normalisation factor ln Z 3 can now also be found to ln Z 3 = ln exp { E i =3ln p(z, y M k ) } dσ3 2 = ln ( (σe 2 ) (ae,k+1) exp ) dσ 2 e + α e,k ln β e,k ln Γ(α e,k ) b e,k σe 2 (N/2) ln(2π) (N/2) ln σe 2 (1/2) ln det(r e (b k )) (102) + E i =3 ln p(s σ 2 e, M k ) + E i =3 ln p(σ 2 s M k ) (103) = a e,k ln b e,k + ln Γ(a e,k ) + α e,k ln β e,k ln Γ(α e,k ) (N/2) ln(2π) (N/2) ln σ 2 e (1/2) ln det(r e (b k )) + E i =3 ln p(s σ 2 e, M k ) + E i =3 ln p(σ 2 s M k ) (104) Finally, since q 3k (σ 2 e ) is an inverse Gamma distribution, its entropy is given by 3.4 Evaluating the expectations Hq 3k = a e,k + ln(b e,k Γ(a e,k )) (1 + a e,k )Ψ(a e,k ). (105) To summarize the main result above, we have that where Σŝ = E q2k σ 2 s q 1k (s) = N (ŝ, Σŝ) (106) q 2k (σ 2 s ) = Inv-G(a s,k, b s,k ) (107) q 3k (σ 2 e ) = Inv-G(a e,k, b e,k ) (108) R 1 s 1 (a k ) + E q3k σe 2 Re 1 (b k ) (109) ŝ = Σŝ E q3k σe 2 Re 1 (b k )y (110) a s,k = N/2 + α s,k (111) b s,k = (1/2) E q1k s T Rs 1 (a k )s + β s,k (112) a e,k = N/2 + α e,k (113) b e,k = (1/2) E q1k (y s) T Re 1 (b k )(y s) + β e,k. (114) Since we now have derived the form of the distributions for the factors, we can evaluate the expectations. These are E q2k σs 2 = a s,k (115) b s,k E q3k σe 2 = a e,k (116) b e,k ( ) E q1k s T Rs 1 (a k )s = ŝ T Rs 1 (a k )ŝ + tr Rs 1 (a k )Σŝ (117) ( ) E q1k (y s) T Rs 1 (a k )(y s) = ê T Re 1 (b k )ê + tr Re 1 (b k )Σŝ (118) 11

13 where ê = y ŝ. For the evaluation of the normalisation constants, we also have that 3.5 Computing the model factor q(m k ) E q2k ln σ 2 s = ln b s,k Ψ(a s,k ) (119) E q3k ln σ 2 e = ln b e,k Ψ(a e,k ). (120) To evaluate the model factor q(m k ), we have to compute the lower bound L k (q k ) for all models. This lower bound can be computed from the normalisation factors {Z ik } 3 i=1 and the entropies for q 1k (s), q 2k (σs 2 ), and q 3k (σe 2 ). Alternatively, we also have from (23) that L k (q k ) can be written as L k (q k ) = E qk ln p(y s, σ 2 e, M k ) + E qk ln p(s σ 2 s, M k )+ E qk ln p(σ 2 s M k ) + E qk ln p(σ 2 e M k ) + Since none of the expectations depends on all three sets of unknowns, we have that M i=1 Hq ik. (121) E qk ln p(y s, σ 2 e, M k ) = E i =2 ln p(y s, σ 2 e, M k ) (122) = (N/2) ln(2π) (N/2) E q3k ln σe 2 (1/2) ln det(r e (b k )) (1/2) E q1k (y s) T Re 1 (b k )(y s) E q3k σe 2 (123) E qk ln p(s σ 2 s, M k ) = E i =3 ln p(s σ 2 s, M k ) (124) References = (N/2) ln(2π) (N/2) E q2k ln σs 2 (1/2) ln det(r s (a k )) (1/2) E q1k s T Rs 1 (a k )s E q2k σs 2 (125) E qk ln p(σ 2 s M k ) = E i =1 ln p(σ 2 s M k ) = E i =3 ln p(σ 2 s M k ) (126) = α s,k ln β s,k ln Γ(α s,k ) (α s,k + 1) E q2k ln σ 2 s β s,k E q2k σs 2 (127) E qk ln p(σe 2 M k ) = E i =1 ln p(σe 2 M k ) = E i =2 ln p(σe 2 M k ) (128) = α e,k ln β e,k ln Γ(α e,k ) (α e,k + 1) E q3k ln σ 2 e β e,k E q3k σ 2 e. (129) 1 C. M. Bishop. Pattern Recognition and Machine Learning. New York, NY, USA: Springer, isbn: J.. Nielsen et al. Model-based Noise PSD Estimation from Speech in Non-stationary Noise. In: Proc. IEEE Int. Conf. Acoust., Speech, Signal Process

Sound absorption properties of a perforated plate and membrane ceiling element Nilsson, Anders C.; Rasmussen, Birgit

Sound absorption properties of a perforated plate and membrane ceiling element Nilsson, Anders C.; Rasmussen, Birgit Aalborg Universitet Sound absorption properties of a perforated plate and membrane ceiling element Nilsson, Anders C.; Rasmussen, Birgit Published in: Proceedings of Inter-Noise 1983 Publication date:

More information

Aalborg Universitet. Lecture 14 - Introduction to experimental work Kramer, Morten Mejlhede. Publication date: 2015

Aalborg Universitet. Lecture 14 - Introduction to experimental work Kramer, Morten Mejlhede. Publication date: 2015 Aalborg Universitet Lecture 14 - Introduction to experimental work Kramer, Morten Mejlhede Publication date: 2015 Document Version Publisher's PDF, also known as Version of record Link to publication from

More information

Published in: Proceedings of the 21st Symposium on Mathematical Theory of Networks and Systems

Published in: Proceedings of the 21st Symposium on Mathematical Theory of Networks and Systems Aalborg Universitet Affine variety codes are better than their reputation Geil, Hans Olav; Martin, Stefano Published in: Proceedings of the 21st Symposium on Mathematical Theory of Networks and Systems

More information

Cheng Soon Ong & Christian Walder. Canberra February June 2017

Cheng Soon Ong & Christian Walder. Canberra February June 2017 Cheng Soon Ong & Christian Walder Research Group and College of Engineering and Computer Science Canberra February June 2017 (Many figures from C. M. Bishop, "Pattern Recognition and ") 1of 679 Part XIX

More information

Aalborg Universitet. Bounds on information combining for parity-check equations Land, Ingmar Rüdiger; Hoeher, A.; Huber, Johannes

Aalborg Universitet. Bounds on information combining for parity-check equations Land, Ingmar Rüdiger; Hoeher, A.; Huber, Johannes Aalborg Universitet Bounds on information combining for parity-check equations Land, Ingmar Rüdiger; Hoeher, A.; Huber, Johannes Published in: 2004 International Seminar on Communications DOI link to publication

More information

Re-estimation of Linear Predictive Parameters in Sparse Linear Prediction

Re-estimation of Linear Predictive Parameters in Sparse Linear Prediction Downloaded from vbnaaudk on: januar 12, 2019 Aalborg Universitet Re-estimation of Linear Predictive Parameters in Sparse Linear Prediction Giacobello, Daniele; Murthi, Manohar N; Christensen, Mads Græsbøll;

More information

Tectonic Thinking Beim, Anne; Bech-Danielsen, Claus; Bundgaard, Charlotte; Madsen, Ulrik Stylsvig

Tectonic Thinking Beim, Anne; Bech-Danielsen, Claus; Bundgaard, Charlotte; Madsen, Ulrik Stylsvig Aalborg Universitet Tectonic Thinking Beim, Anne; Bech-Danielsen, Claus; Bundgaard, Charlotte; Madsen, Ulrik Stylsvig Publication date: 2011 Document Version Early version, also known as pre-print Link

More information

Aalborg Universitet. Land-use modelling by cellular automata Hansen, Henning Sten. Published in: NORDGI : Nordic Geographic Information

Aalborg Universitet. Land-use modelling by cellular automata Hansen, Henning Sten. Published in: NORDGI : Nordic Geographic Information Aalborg Universitet Land-use modelling by cellular automata Hansen, Henning Sten Published in: NORDGI : Nordic Geographic Information Publication date: 2006 Document Version Publisher's PDF, also known

More information

The Expectation Maximization or EM algorithm

The Expectation Maximization or EM algorithm The Expectation Maximization or EM algorithm Carl Edward Rasmussen November 15th, 2017 Carl Edward Rasmussen The EM algorithm November 15th, 2017 1 / 11 Contents notation, objective the lower bound functional,

More information

Matrix representation of a Neural Network

Matrix representation of a Neural Network Downloaded from orbit.dtu.dk on: May 01, 2018 Matrix representation of a Neural Network Christensen, Bjørn Klint Publication date: 2003 Document Version Publisher's PDF, also known as Version of record

More information

GIS Readiness Survey 2014 Schrøder, Anne Lise; Hvingel, Line Træholt; Hansen, Henning Sten; Skovdal, Jesper Høi

GIS Readiness Survey 2014 Schrøder, Anne Lise; Hvingel, Line Træholt; Hansen, Henning Sten; Skovdal, Jesper Høi Aalborg Universitet GIS Readiness Survey 2014 Schrøder, Anne Lise; Hvingel, Line Træholt; Hansen, Henning Sten; Skovdal, Jesper Høi Published in: Geoforum Perspektiv DOI (link to publication from Publisher):

More information

The IEA Annex 20 Two-Dimensional Benchmark Test for CFD Predictions

The IEA Annex 20 Two-Dimensional Benchmark Test for CFD Predictions Downloaded from vbn.aau.dk on: april 05, 2019 Aalborg Universitet The IEA Annex 20 Two-Dimensional Benchmark Test for CFD Predictions Nielsen, Peter V.; Rong, Li; Cortes, Ines Olmedo Published in: Clima

More information

Robust Implementation of the MUSIC algorithm Zhang, Johan Xi; Christensen, Mads Græsbøll; Dahl, Joachim; Jensen, Søren Holdt; Moonen, Marc

Robust Implementation of the MUSIC algorithm Zhang, Johan Xi; Christensen, Mads Græsbøll; Dahl, Joachim; Jensen, Søren Holdt; Moonen, Marc Aalborg Universitet Robust Implementation of the MUSIC algorithm Zhang, Johan Xi; Christensen, Mads Græsbøll; Dahl, Joachim; Jensen, Søren Holdt; Moonen, Marc Published in: I E E E International Conference

More information

Estimation of Peak Wave Stresses in Slender Complex Concrete Armor Units

Estimation of Peak Wave Stresses in Slender Complex Concrete Armor Units Downloaded from vbn.aau.dk on: januar 19, 2019 Aalborg Universitet Estimation of Peak Wave Stresses in Slender Complex Concrete Armor Units Howell, G.L.; Burcharth, Hans Falk; Rhee, Joon R Published in:

More information

The Variational Gaussian Approximation Revisited

The Variational Gaussian Approximation Revisited The Variational Gaussian Approximation Revisited Manfred Opper Cédric Archambeau March 16, 2009 Abstract The variational approximation of posterior distributions by multivariate Gaussians has been much

More information

Variational inference

Variational inference Simon Leglaive Télécom ParisTech, CNRS LTCI, Université Paris Saclay November 18, 2016, Télécom ParisTech, Paris, France. Outline Introduction Probabilistic model Problem Log-likelihood decomposition EM

More information

Unknown input observer based scheme for detecting faults in a wind turbine converter Odgaard, Peter Fogh; Stoustrup, Jakob

Unknown input observer based scheme for detecting faults in a wind turbine converter Odgaard, Peter Fogh; Stoustrup, Jakob Aalborg Universitet Unknown input observer based scheme for detecting faults in a wind turbine converter Odgaard, Peter Fogh; Stoustrup, Jakob Published in: Elsevier IFAC Publications / IFAC Proceedings

More information

Published in: Proceedings of IECON 2016: 42nd Annual Conference of the IEEE Industrial Electronics Society

Published in: Proceedings of IECON 2016: 42nd Annual Conference of the IEEE Industrial Electronics Society Aalborg Universitet Design of Energy Storage Control Strategy to Improve the PV System Power Quality Lei, Mingyu; Yang, Zilong; Wang, Yibo; Xu, Honghua; Meng, Lexuan; Quintero, Juan Carlos Vasquez; Guerrero,

More information

Variational Principal Components

Variational Principal Components Variational Principal Components Christopher M. Bishop Microsoft Research 7 J. J. Thomson Avenue, Cambridge, CB3 0FB, U.K. cmbishop@microsoft.com http://research.microsoft.com/ cmbishop In Proceedings

More information

Room Allocation Optimisation at the Technical University of Denmark

Room Allocation Optimisation at the Technical University of Denmark Downloaded from orbit.dtu.dk on: Mar 13, 2018 Room Allocation Optimisation at the Technical University of Denmark Bagger, Niels-Christian Fink; Larsen, Jesper; Stidsen, Thomas Jacob Riis Published in:

More information

Control Configuration Selection for Multivariable Descriptor Systems Shaker, Hamid Reza; Stoustrup, Jakob

Control Configuration Selection for Multivariable Descriptor Systems Shaker, Hamid Reza; Stoustrup, Jakob Aalborg Universitet Control Configuration Selection for Multivariable Descriptor Systems Shaker, Hamid Reza; Stoustrup, Jakob Published in: 2012 American Control Conference (ACC) Publication date: 2012

More information

Plane Wave Medical Ultrasound Imaging Using Adaptive Beamforming

Plane Wave Medical Ultrasound Imaging Using Adaptive Beamforming Downloaded from orbit.dtu.dk on: Jul 14, 2018 Plane Wave Medical Ultrasound Imaging Using Adaptive Beamforming Voxen, Iben Holfort; Gran, Fredrik; Jensen, Jørgen Arendt Published in: Proceedings of 5th

More information

Development of a Dainish infrastructure for spatial information (DAISI) Brande-Lavridsen, Hanne; Jensen, Bent Hulegaard

Development of a Dainish infrastructure for spatial information (DAISI) Brande-Lavridsen, Hanne; Jensen, Bent Hulegaard Aalborg Universitet Development of a Dainish infrastructure for spatial information (DAISI) Brande-Lavridsen, Hanne; Jensen, Bent Hulegaard Published in: NORDGI : Nordic Geographic Information Publication

More information

Reducing Uncertainty of Near-shore wind resource Estimates (RUNE) using wind lidars and mesoscale models

Reducing Uncertainty of Near-shore wind resource Estimates (RUNE) using wind lidars and mesoscale models Downloaded from orbit.dtu.dk on: Dec 16, 2018 Reducing Uncertainty of Near-shore wind resource Estimates (RUNE) using wind lidars and mesoscale models Floors, Rogier Ralph; Vasiljevic, Nikola; Lea, Guillaume;

More information

Multi (building)physics modeling

Multi (building)physics modeling Multi (building)physics modeling van Schijndel, A.W.M. Published: 01/01/2010 Document Version Publisher s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check

More information

Properties and approximations of some matrix variate probability density functions

Properties and approximations of some matrix variate probability density functions Technical report from Automatic Control at Linköpings universitet Properties and approximations of some matrix variate probability density functions Karl Granström, Umut Orguner Division of Automatic Control

More information

Aalborg Universitet. All P3-equipackable graphs Randerath, Bert; Vestergaard, Preben Dahl. Publication date: 2008

Aalborg Universitet. All P3-equipackable graphs Randerath, Bert; Vestergaard, Preben Dahl. Publication date: 2008 Aalborg Universitet All P-equipackable graphs Randerath, Bert; Vestergaard, Preben Dahl Publication date: 2008 Document Version Publisher's PD, also known as Version of record Link to publication from

More information

Foundations of Statistical Inference

Foundations of Statistical Inference Foundations of Statistical Inference Julien Berestycki Department of Statistics University of Oxford MT 2016 Julien Berestycki (University of Oxford) SB2a MT 2016 1 / 32 Lecture 14 : Variational Bayes

More information

Multiple Scenario Inversion of Reflection Seismic Prestack Data

Multiple Scenario Inversion of Reflection Seismic Prestack Data Downloaded from orbit.dtu.dk on: Nov 28, 2018 Multiple Scenario Inversion of Reflection Seismic Prestack Data Hansen, Thomas Mejer; Cordua, Knud Skou; Mosegaard, Klaus Publication date: 2013 Document Version

More information

IEOR E4570: Machine Learning for OR&FE Spring 2015 c 2015 by Martin Haugh. The EM Algorithm

IEOR E4570: Machine Learning for OR&FE Spring 2015 c 2015 by Martin Haugh. The EM Algorithm IEOR E4570: Machine Learning for OR&FE Spring 205 c 205 by Martin Haugh The EM Algorithm The EM algorithm is used for obtaining maximum likelihood estimates of parameters when some of the data is missing.

More information

Pitch Estimation and Tracking with Harmonic Emphasis On The Acoustic Spectrum

Pitch Estimation and Tracking with Harmonic Emphasis On The Acoustic Spectrum Downloaded from vbn.aau.dk on: marts 31, 2019 Aalborg Universitet Pitch Estimation and Tracking with Harmonic Emphasis On The Acoustic Spectrum Karimian-Azari, Sam; Mohammadiha, Nasser; Jensen, Jesper

More information

Citation for published version (APA): Burcharth, H. F. (1988). Stress Analysis. Aalborg: Department of Civil Engineering, Aalborg University.

Citation for published version (APA): Burcharth, H. F. (1988). Stress Analysis. Aalborg: Department of Civil Engineering, Aalborg University. Aalborg Universitet Stress Analysis Burcharth, Hans Falk Publication date: 1988 Document Version Accepted author manuscript, peer reviewed version Link to publication from Aalborg University Citation for

More information

Clustering K-means. Clustering images. Machine Learning CSE546 Carlos Guestrin University of Washington. November 4, 2014.

Clustering K-means. Clustering images. Machine Learning CSE546 Carlos Guestrin University of Washington. November 4, 2014. Clustering K-means Machine Learning CSE546 Carlos Guestrin University of Washington November 4, 2014 1 Clustering images Set of Images [Goldberger et al.] 2 1 K-means Randomly initialize k centers µ (0)

More information

Upper and Lower Bounds of Frequency Interval Gramians for a Class of Perturbed Linear Systems Shaker, Hamid Reza

Upper and Lower Bounds of Frequency Interval Gramians for a Class of Perturbed Linear Systems Shaker, Hamid Reza Aalborg Universitet Upper and Lower Bounds of Frequency Interval Gramians for a Class of Perturbed Linear Systems Shaker, Hamid Reza Published in: 7th IFAC Symposium on Robust Control Design DOI (link

More information

Bayesian Inference Course, WTCN, UCL, March 2013

Bayesian Inference Course, WTCN, UCL, March 2013 Bayesian Course, WTCN, UCL, March 2013 Shannon (1948) asked how much information is received when we observe a specific value of the variable x? If an unlikely event occurs then one would expect the information

More information

An Introduction to Expectation-Maximization

An Introduction to Expectation-Maximization An Introduction to Expectation-Maximization Dahua Lin Abstract This notes reviews the basics about the Expectation-Maximization EM) algorithm, a popular approach to perform model estimation of the generative

More information

The M/G/1 FIFO queue with several customer classes

The M/G/1 FIFO queue with several customer classes The M/G/1 FIFO queue with several customer classes Boxma, O.J.; Takine, T. Published: 01/01/2003 Document Version Publisher s PDF, also known as Version of Record (includes final page, issue and volume

More information

An Indicator for Separation of Structural and Harmonic Modes in Output-Only Modal Testing Brincker, Rune; Andersen, P.; Møller, N.

An Indicator for Separation of Structural and Harmonic Modes in Output-Only Modal Testing Brincker, Rune; Andersen, P.; Møller, N. Aalborg Universitet An Indicator for Separation of Structural and Harmonic Modes in Output-Only Modal Testing Brincker, Rune; Andersen, P.; Møller, N. Published in: Proceedings of the European COST F3

More information

Design possibilities for impact noise insulation in lightweight floors A parameter study

Design possibilities for impact noise insulation in lightweight floors A parameter study Downloaded from orbit.dtu.dk on: Dec 23, 218 Design possibilities for impact noise insulation in lightweight floors A parameter study Brunskog, Jonas; Hammer, Per Published in: Euronoise Publication date:

More information

Characterization of microporous silica-based membranes by calorimetric analysis Boffa, Vittorio; Yue, Yuanzheng

Characterization of microporous silica-based membranes by calorimetric analysis Boffa, Vittorio; Yue, Yuanzheng Aalborg Universitet Characterization of microporous silica-based membranes by calorimetric analysis Boffa, Vittorio; Yue, Yuanzheng Published in: 6th Black Sea Conference on Analytical Chemistry - Book

More information

Spatial decay rate of speech in open plan offices: the use of D2,S and Lp,A,S,4m as building requirements Wenmaekers, R.H.C.; Hak, C.C.J.M.

Spatial decay rate of speech in open plan offices: the use of D2,S and Lp,A,S,4m as building requirements Wenmaekers, R.H.C.; Hak, C.C.J.M. Spatial decay rate of speech in open plan offices: the use of D2,S and Lp,A,S,4m as building requirements Wenmaekers, R.H.C.; Hak, C.C.J.M. Published in: Euronoise 2015, the 10th European Congress and

More information

Heriot-Watt University

Heriot-Watt University Heriot-Watt University Heriot-Watt University Research Gateway Prediction of settlement delay in critical illness insurance claims by using the generalized beta of the second kind distribution Dodd, Erengul;

More information

The dipole moment of a wall-charged void in a bulk dielectric

The dipole moment of a wall-charged void in a bulk dielectric Downloaded from orbit.dtu.dk on: Dec 17, 2017 The dipole moment of a wall-charged void in a bulk dielectric McAllister, Iain Wilson Published in: Proceedings of the Conference on Electrical Insulation

More information

A Bayesian Treatment of Linear Gaussian Regression

A Bayesian Treatment of Linear Gaussian Regression A Bayesian Treatment of Linear Gaussian Regression Frank Wood December 3, 2009 Bayesian Approach to Classical Linear Regression In classical linear regression we have the following model y β, σ 2, X N(Xβ,

More information

Introduction to Probabilistic Graphical Models: Exercises

Introduction to Probabilistic Graphical Models: Exercises Introduction to Probabilistic Graphical Models: Exercises Cédric Archambeau Xerox Research Centre Europe cedric.archambeau@xrce.xerox.com Pascal Bootcamp Marseille, France, July 2010 Exercise 1: basics

More information

Joint Direction-of-Arrival and Order Estimation in Compressed Sensing using Angles between Subspaces

Joint Direction-of-Arrival and Order Estimation in Compressed Sensing using Angles between Subspaces Aalborg Universitet Joint Direction-of-Arrival and Order Estimation in Compressed Sensing using Angles between Subspaces Christensen, Mads Græsbøll; Nielsen, Jesper Kjær Published in: I E E E / S P Workshop

More information

Modelling Salmonella transfer during grinding of meat

Modelling Salmonella transfer during grinding of meat Downloaded from orbit.dtu.dk on: Nov 24, 2018 Modelling Salmonella transfer during grinding of meat Hansen, Tina Beck Publication date: 2017 Document Version Publisher's PDF, also known as Version of record

More information

Radioactivity in the Risø District January-June 2014

Radioactivity in the Risø District January-June 2014 Downloaded from orbit.dtu.dk on: Dec 20, 2017 Radioactivity in the Risø District January-June 2014 Nielsen, Sven Poul; Andersson, Kasper Grann; Miller, Arne Publication date: 2014 Document Version Publisher's

More information

Reflection of sound from finite-size plane and curved surfaces

Reflection of sound from finite-size plane and curved surfaces Downloaded from orbit.dtu.dk on: Sep, 08 Reflection of sound from finite-size plane and curved surfaces Rindel, Jens Holger Published in: Journal of the Acoustical Society of America Publication date:

More information

A beam theory fracture mechanics approach for strength analysis of beams with a hole

A beam theory fracture mechanics approach for strength analysis of beams with a hole A beam theory fracture mechanics approach for strength analysis of beams with a hole Danielsson, Henrik; Gustafsson, Per-Johan Published in: International Network on Timber Engineering Research - Proceedings

More information

G8325: Variational Bayes

G8325: Variational Bayes G8325: Variational Bayes Vincent Dorie Columbia University Wednesday, November 2nd, 2011 bridge Variational University Bayes Press 2003. On-screen viewing permitted. Printing not permitted. http://www.c

More information

Series 7, May 22, 2018 (EM Convergence)

Series 7, May 22, 2018 (EM Convergence) Exercises Introduction to Machine Learning SS 2018 Series 7, May 22, 2018 (EM Convergence) Institute for Machine Learning Dept. of Computer Science, ETH Zürich Prof. Dr. Andreas Krause Web: https://las.inf.ethz.ch/teaching/introml-s18

More information

Aalborg Universitet. Encounter Probability of Significant Wave Height Liu, Z.; Burcharth, Hans Falk. Publication date: 1998

Aalborg Universitet. Encounter Probability of Significant Wave Height Liu, Z.; Burcharth, Hans Falk. Publication date: 1998 Aalborg Universitet Encounter Probability of Significant Wave Height Liu, Z.; Burcharth, Hans Falk Publication date: 1998 Document Version Early version, also known as pre-print Link to publication from

More information

Optimizing Reliability using BECAS - an Open-Source Cross Section Analysis Tool

Optimizing Reliability using BECAS - an Open-Source Cross Section Analysis Tool Downloaded from orbit.dtu.dk on: Dec 20, 2017 Optimizing Reliability using BECAS - an Open-Source Cross Section Analysis Tool Bitsche, Robert; Blasques, José Pedro Albergaria Amaral Publication date: 2012

More information

An Equivalent Source Method for Modelling the Lithospheric Magnetic Field Using Satellite and Airborne Magnetic Data

An Equivalent Source Method for Modelling the Lithospheric Magnetic Field Using Satellite and Airborne Magnetic Data Downloaded from orbit.dtu.dk on: Oct 03, 2018 An Equivalent Source Method for Modelling the Lithospheric Magnetic Field Using Satellite and Airborne Magnetic Data Kother, Livia Kathleen; D. Hammer, Magnus;

More information

The Wien Bridge Oscillator Family

The Wien Bridge Oscillator Family Downloaded from orbit.dtu.dk on: Dec 29, 207 The Wien Bridge Oscillator Family Lindberg, Erik Published in: Proceedings of the ICSES-06 Publication date: 2006 Link back to DTU Orbit Citation APA): Lindberg,

More information

Computer Vision Group Prof. Daniel Cremers. 6. Mixture Models and Expectation-Maximization

Computer Vision Group Prof. Daniel Cremers. 6. Mixture Models and Expectation-Maximization Prof. Daniel Cremers 6. Mixture Models and Expectation-Maximization Motivation Often the introduction of latent (unobserved) random variables into a model can help to express complex (marginal) distributions

More information

Aalborg Universitet. Reparametrizations with given stop data Raussen, Martin Hubert. Publication date: 2008

Aalborg Universitet. Reparametrizations with given stop data Raussen, Martin Hubert. Publication date: 2008 Aalborg Universitet Reparametrizations with given stop data Raussen, Martin Hubert Publication date: 2008 Document Version Publisher's PDF, also known as Version of record Link to publication from Aalborg

More information

Variational Inference. Sargur Srihari

Variational Inference. Sargur Srihari Variational Inference Sargur srihari@cedar.buffalo.edu 1 Plan of Discussion Functionals Calculus of Variations Maximizing a Functional Finding Approximation to a Posterior Minimizing K-L divergence Factorized

More information

Aalborg Universitet. Comparison between Different Air Distribution Systems Nielsen, Peter Vilhelm. Publication date: 2006

Aalborg Universitet. Comparison between Different Air Distribution Systems Nielsen, Peter Vilhelm. Publication date: 2006 Aalborg Universitet Comparison between Different Air Distribution Systems Nielsen, Peter Vilhelm Publication date: 2006 Document Version Publisher's PDF, also known as Version of record Link to publication

More information

Bayesian Linear Models

Bayesian Linear Models Bayesian Linear Models Sudipto Banerjee 1 and Andrew O. Finley 2 1 Department of Forestry & Department of Geography, Michigan State University, Lansing Michigan, U.S.A. 2 Biostatistics, School of Public

More information

Lidar calibration What s the problem?

Lidar calibration What s the problem? Downloaded from orbit.dtu.dk on: Feb 05, 2018 Lidar calibration What s the problem? Courtney, Michael Publication date: 2015 Document Version Peer reviewed version Link back to DTU Orbit Citation (APA):

More information

Published in: Proceedings of the 17th International Modal Analysis Conference (IMAC), Kissimmee, Florida, USA, February 8-11, 1999

Published in: Proceedings of the 17th International Modal Analysis Conference (IMAC), Kissimmee, Florida, USA, February 8-11, 1999 Aalborg Universitet ARMA Models in Modal Space Brincker, Rune Published in: Proceedings of the 17th International Modal Analysis Conference (IMAC), Kissimmee, Florida, USA, February 8-11, 1999 Publication

More information

A Derivation of the EM Updates for Finding the Maximum Likelihood Parameter Estimates of the Student s t Distribution

A Derivation of the EM Updates for Finding the Maximum Likelihood Parameter Estimates of the Student s t Distribution A Derivation of the EM Updates for Finding the Maximum Likelihood Parameter Estimates of the Student s t Distribution Carl Scheffler First draft: September 008 Contents The Student s t Distribution The

More information

Aalborg Universitet. The number of independent sets in unicyclic graphs Pedersen, Anders Sune; Vestergaard, Preben Dahl. Publication date: 2005

Aalborg Universitet. The number of independent sets in unicyclic graphs Pedersen, Anders Sune; Vestergaard, Preben Dahl. Publication date: 2005 Aalborg Universitet The number of independent sets in unicyclic graphs Pedersen, Anders Sune; Vestergaard, Preben Dahl Publication date: 2005 Document Version Publisher's PDF, also known as Version of

More information

Pattern Recognition and Machine Learning. Bishop Chapter 9: Mixture Models and EM

Pattern Recognition and Machine Learning. Bishop Chapter 9: Mixture Models and EM Pattern Recognition and Machine Learning Chapter 9: Mixture Models and EM Thomas Mensink Jakob Verbeek October 11, 27 Le Menu 9.1 K-means clustering Getting the idea with a simple example 9.2 Mixtures

More information

Characterization of the Airflow from a Bottom Hung Window under Natural Ventilation

Characterization of the Airflow from a Bottom Hung Window under Natural Ventilation Downloaded from vbn.aau.dk on: March 2, 219 Aalborg Universitet Characterization of the Airflow from a Bottom Hung Window under Natural Ventilation Svidt, Kjeld; Heiselberg, Per; Nielsen, Peter V. Publication

More information

Phase Diagram Study on Variational Bayes Learning of Bernoulli Mixture

Phase Diagram Study on Variational Bayes Learning of Bernoulli Mixture 009 Technical Report on Information-Based Induction Sciences 009 (IBIS009) Phase Diagram Study on Variational Bayes Learning of Bernoulli Mixture Daisue Kaji Sumio Watanabe Abstract: Variational Bayes

More information

On a set of diophantine equations

On a set of diophantine equations On a set of diophantine equations Citation for published version (APA): van Lint, J. H. (1968). On a set of diophantine equations. (EUT report. WSK, Dept. of Mathematics and Computing Science; Vol. 68-WSK-03).

More information

Radioactivity in the Risø District July-December 2013

Radioactivity in the Risø District July-December 2013 Downloaded from orbit.dtu.dk on: Jan 02, 2019 Radioactivity in the Risø District July-December 2013 Nielsen, Sven Poul; Andersson, Kasper Grann; Miller, Arne Publication date: 2014 Document Version Publisher's

More information

MACHINE LEARNING AND PATTERN RECOGNITION Fall 2006, Lecture 8: Latent Variables, EM Yann LeCun

MACHINE LEARNING AND PATTERN RECOGNITION Fall 2006, Lecture 8: Latent Variables, EM Yann LeCun Y. LeCun: Machine Learning and Pattern Recognition p. 1/? MACHINE LEARNING AND PATTERN RECOGNITION Fall 2006, Lecture 8: Latent Variables, EM Yann LeCun The Courant Institute, New York University http://yann.lecun.com

More information

Expectation Maximization

Expectation Maximization Expectation Maximization Aaron C. Courville Université de Montréal Note: Material for the slides is taken directly from a presentation prepared by Christopher M. Bishop Learning in DAGs Two things could

More information

Reflection and absorption coefficients for use in room acoustic simulations

Reflection and absorption coefficients for use in room acoustic simulations Downloaded from orbit.dtu.dk on: May 1, 018 Reflection and absorption coefficients for use in room acoustic simulations Jeong, Cheol-Ho Published in: Proceedings of Spring Meeting of the Acoustical Society

More information

A Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement

A Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement A Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement Simon Leglaive 1 Laurent Girin 1,2 Radu Horaud 1 1: Inria Grenoble Rhône-Alpes 2: Univ. Grenoble Alpes, Grenoble INP,

More information

A Note on Powers in Finite Fields

A Note on Powers in Finite Fields Downloaded from orbit.dtu.dk on: Jul 11, 2018 A Note on Powers in Finite Fields Aabrandt, Andreas; Hansen, Vagn Lundsgaard Published in: International Journal of Mathematical Education in Science and Technology

More information

Bayesian Linear Models

Bayesian Linear Models Bayesian Linear Models Sudipto Banerjee 1 and Andrew O. Finley 2 1 Biostatistics, School of Public Health, University of Minnesota, Minneapolis, Minnesota, U.S.A. 2 Department of Forestry & Department

More information

On the Einstein-Stern model of rotational heat capacities

On the Einstein-Stern model of rotational heat capacities Downloaded from orbit.dtu.dk on: Jan 06, 2018 On the Einstein-Stern model of rotational heat capacities Dahl, Jens Peder Published in: Journal of Chemical Physics Link to article, DOI: 10.1063/1.77766

More information

Lecture 1a: Basic Concepts and Recaps

Lecture 1a: Basic Concepts and Recaps Lecture 1a: Basic Concepts and Recaps Cédric Archambeau Centre for Computational Statistics and Machine Learning Department of Computer Science University College London c.archambeau@cs.ucl.ac.uk Advanced

More information

Fatigue Reliability and Effective Turbulence Models in Wind Farms

Fatigue Reliability and Effective Turbulence Models in Wind Farms Downloaded from vbn.aau.dk on: marts 28, 2019 Aalborg Universitet Fatigue Reliability and Effective Turbulence Models in Wind Farms Sørensen, John Dalsgaard; Frandsen, S.; Tarp-Johansen, N.J. Published

More information

Aalborg Universitet. CLIMA proceedings of the 12th REHVA World Congress Heiselberg, Per Kvols. Publication date: 2016

Aalborg Universitet. CLIMA proceedings of the 12th REHVA World Congress Heiselberg, Per Kvols. Publication date: 2016 Aalborg Universitet CLIMA 2016 - proceedings of the 12th REHVA World Congress Heiselberg, Per Kvols Publication date: 2016 Document Version Final published version Link to publication from Aalborg University

More information

Curve Fitting Re-visited, Bishop1.2.5

Curve Fitting Re-visited, Bishop1.2.5 Curve Fitting Re-visited, Bishop1.2.5 Maximum Likelihood Bishop 1.2.5 Model Likelihood differentiation p(t x, w, β) = Maximum Likelihood N N ( t n y(x n, w), β 1). (1.61) n=1 As we did in the case of the

More information

Dynamic Asset Allocation - Identifying Regime Shifts in Financial Time Series to Build Robust Portfolios

Dynamic Asset Allocation - Identifying Regime Shifts in Financial Time Series to Build Robust Portfolios Downloaded from orbit.dtu.dk on: Jan 22, 2019 Dynamic Asset Allocation - Identifying Regime Shifts in Financial Time Series to Build Robust Portfolios Nystrup, Peter Publication date: 2018 Document Version

More information

Published in: Proceedings of the Twentieth (2010) International Offshore and Polar Engineering Conference

Published in: Proceedings of the Twentieth (2010) International Offshore and Polar Engineering Conference Aalborg Universitet Performance Evaluation of an Axysimmetric Floating OWC Alves, M. A.; Costa, I. R.; Sarmento, A. J.; Chozas, Julia Fernandez Published in: Proceedings of the Twentieth (010) International

More information

Clustering with k-means and Gaussian mixture distributions

Clustering with k-means and Gaussian mixture distributions Clustering with k-means and Gaussian mixture distributions Machine Learning and Object Recognition 2017-2018 Jakob Verbeek Clustering Finding a group structure in the data Data in one cluster similar to

More information

Accelerating interior point methods with GPUs for smart grid systems

Accelerating interior point methods with GPUs for smart grid systems Downloaded from orbit.dtu.dk on: Dec 18, 2017 Accelerating interior point methods with GPUs for smart grid systems Gade-Nielsen, Nicolai Fog Publication date: 2011 Document Version Publisher's PDF, also

More information

An optimization problem related to the storage of solar energy van der Wal, J.

An optimization problem related to the storage of solar energy van der Wal, J. An optimization problem related to the storage of solar energy van der Wal, J. Published: 01/01/1987 Document Version Publisher s PDF, also known as Version of Record (includes final page, issue and volume

More information

Multivariate Bayesian Linear Regression MLAI Lecture 11

Multivariate Bayesian Linear Regression MLAI Lecture 11 Multivariate Bayesian Linear Regression MLAI Lecture 11 Neil D. Lawrence Department of Computer Science Sheffield University 21st October 2012 Outline Univariate Bayesian Linear Regression Multivariate

More information

Polydiagnostic study on a surfatron plasma at atmospheric pressure

Polydiagnostic study on a surfatron plasma at atmospheric pressure Polydiagnostic study on a surfatron plasma at atmospheric pressure Citation for published version (APA): Palomares, J. M., Iordanova, E. I., Gamero, A., Sola, A., & Mullen, van der, J. J. A. M. (2009).

More information

Week 3: The EM algorithm

Week 3: The EM algorithm Week 3: The EM algorithm Maneesh Sahani maneesh@gatsby.ucl.ac.uk Gatsby Computational Neuroscience Unit University College London Term 1, Autumn 2005 Mixtures of Gaussians Data: Y = {y 1... y N } Latent

More information

Clustering with k-means and Gaussian mixture distributions

Clustering with k-means and Gaussian mixture distributions Clustering with k-means and Gaussian mixture distributions Machine Learning and Category Representation 2014-2015 Jakob Verbeek, ovember 21, 2014 Course website: http://lear.inrialpes.fr/~verbeek/mlcr.14.15

More information

Aalborg Universitet. Publication date: Document Version Publisher's PDF, also known as Version of record

Aalborg Universitet. Publication date: Document Version Publisher's PDF, also known as Version of record Aalborg Universitet Experimental Investigation of the Influence of Different Flooring Emissivity on Night- Time Cooling using Displacement Ventilation Dreau, Jerome Le; Karlsen, Line Røseth; Litewnicki,

More information

Analytical Studies of the Influence of Regional Groundwater Flow by on the Performance of Borehole Heat Exchangers

Analytical Studies of the Influence of Regional Groundwater Flow by on the Performance of Borehole Heat Exchangers Analytical Studies of the Influence of Regional Groundwater Flow by on the Performance of Borehole Heat Exchangers Claesson, Johan; Hellström, Göran Published in: [Host publication title missing] Published:

More information

Aalborg Universitet. Selection of Air Terminal Device. Nielsen, Peter V. Publication date: 1988

Aalborg Universitet. Selection of Air Terminal Device. Nielsen, Peter V. Publication date: 1988 Downloaded from vbn.aau.dk on: March 11, 2019 Aalborg Universitet Selection of Air Terminal Device Nielsen, Peter V. Publication date: 1988 Document Version Publisher's PDF, also known as Version of record

More information

Graphs with given diameter maximizing the spectral radius van Dam, Edwin

Graphs with given diameter maximizing the spectral radius van Dam, Edwin Tilburg University Graphs with given diameter maximizing the spectral radius van Dam, Edwin Published in: Linear Algebra and its Applications Publication date: 2007 Link to publication Citation for published

More information

Aalborg Universitet. On Perceptual Distortion Measures and Parametric Modeling Christensen, Mads Græsbøll. Published in: Proceedings of Acoustics'08

Aalborg Universitet. On Perceptual Distortion Measures and Parametric Modeling Christensen, Mads Græsbøll. Published in: Proceedings of Acoustics'08 Aalborg Universitet On Perceptual Distortion Measures and Parametric Modeling Christensen, Mads Græsbøll Published in: Proceedings of Acoustics'08 Publication date: 2008 Document Version Publisher's PDF,

More information

Pattern Recognition and Machine Learning. Bishop Chapter 6: Kernel Methods

Pattern Recognition and Machine Learning. Bishop Chapter 6: Kernel Methods Pattern Recognition and Machine Learning Chapter 6: Kernel Methods Vasil Khalidov Alex Kläser December 13, 2007 Training Data: Keep or Discard? Parametric methods (linear/nonlinear) so far: learn parameter

More information

Statistical Machine Learning Lecture 8: Markov Chain Monte Carlo Sampling

Statistical Machine Learning Lecture 8: Markov Chain Monte Carlo Sampling 1 / 27 Statistical Machine Learning Lecture 8: Markov Chain Monte Carlo Sampling Melih Kandemir Özyeğin University, İstanbul, Turkey 2 / 27 Monte Carlo Integration The big question : Evaluate E p(z) [f(z)]

More information

Optimal acid digestion for multi-element analysis of different waste matrices

Optimal acid digestion for multi-element analysis of different waste matrices Downloaded from orbit.dtu.dk on: Sep 06, 2018 Optimal acid digestion for multi-element analysis of different waste matrices Götze, Ramona; Astrup, Thomas Fruergaard Publication date: 2014 Document Version

More information