Probit Normal Correlated Topic Model

Size: px
Start display at page:

Download "Probit Normal Correlated Topic Model"

Transcription

1 Open Journal of Statistics, 2014, 4, Publishe Online December 2014 in SciRes. Probit Normal Correlate Topic Moel Xingchen u, Ernest Fooué Center for Quality an Applie Statistics, Rochester Institute of Technology, Rochester, N, USA Receive 3 October 2014; revise 28 October 2014; accepte 15 November 2014 Copyright 2014 by authors an Scientific Research Publishing Inc. This wor is license uner the Creative Commons Attribution International License (CC B. Abstract The logistic normal istribution has recently been aapte via the transformation of multivariate Gaussian variables to moel the topical istribution of ocuments in the presence of correlations among topics. In this paper, we propose a probit normal alternative approach to moelling correlate topical structures. Our use of the probit moel in the context of topic iscovery is novel, as many authors have so far concentrate solely of the logistic moel partly ue to the formiable inefficiency of the multinomial probit moel even in the case of very small topical spaces. We herein circumvent the inefficiency of multinomial probit estimation by using an aaptation of the iagonal orthant multinomial probit in the topic moels context, resulting in the ability of our topic moeling scheme to hanle corpuses with a large number of latent topics. An aitional an very important benefit of our metho lies in the fact that unlie with the logistic normal moel whose non-conugacy leas to the nee for sophisticate sampling schemes, our approach exploits the natural conugacy inherent in the auxiliary formulation of the probit moel to achieve greater simplicity. The application of our propose scheme to a well-nown Associate Press corpus not only helps iscover a large number of meaningful topics but also reveals the capturing of compellingly intuitive correlations among certain topics. Besies, our propose approach lens itself to even further scalability thans to various existing high performance algorithms an architectures capable of hanling millions of ocuments. eywors Topic Moel, Bayesian, Gibbs Sampler, Cumulative Distribution Function, Probit, Logit, Diagonal Orthant, Efficient Sampling, Auxiliary Variable, Correlation Structure, Topic, Vocabulary, Conugate, Dirichlet, Gaussian 1. Introuction The tas of recovering the latent topics unerlying a given corpus of DD ocuments has been in the forefront of active research in statistical machine learning for more than a ecae, an continues to receive the eicate How to cite this paper: u, X.C. an Fooué, E. (2014 Probit Normal Correlate Topic Moel. Open Journal of Statistics, 4,

2 X. C. u, E. Fooué contributions from many researchers from aroun the worl. Since the introuction of Latent Dirichlet Allocation (LDA [1] an then the extension to correlate topic moels (CTM [2], a series of excellent contributions have been mae to this exciting fiel, ranging from slight extension in the moelling structure to the evelopment of scalable topic moeling algorithms capable of hanling extremely large collections of ocuments, as well as selecting an optimal moel among a collection of competing moels or using the output of topic moeling as entry points (inputs to other machine learning or ata mining tass such as image analysis an sentiment extraction, ust to name a few. As far as correlate topic moels are concerne, virtually all the contributors to the fiel have so far concentrate solely on the use of the logistic normal topic moel. The seminal paper on correlate topic moel [2] aopts a variational approximation approach to moel fitting while subsequent authors lie [3] propose a Gibbs sampling scheme with ata augmentation of uniform ranom variables. More recently, [4] presente an exact an scalable Gibbs sampling algorithm with Polya-Gamma istribute auxiliary variables which is a recent evelopment of efficient sampling of logistic moel. Despite the inseparable relationship between logistic an probit moel in statistical moelling, the probit moel has not yet been propose, probably ue to its computational inefficiency for multiclass classification problem an high posterior epenence between auxiliary variables an parameters. As for practical application where topic moels are commonly employe, having multiple topics is extremely prevalent. In some cases, more than 1000 topics will be fitte to large atasets such as Wiipeia an Pubme ata. Therefore, using MCMC probit moel in topic moeling application will be impractical an inconceivable ue to its computational inefficiency. Nonetheless, a recent wor on iagonal orthant probit moel [5] substantially improve the sampling efficiency while maintaining the preictive performance, which motivate us to buil an alternative correlate topic moeling with probit normal topic istribution. On the other han, probit moels inherently capture a better epenency structure between topics an co-occurrence of wors within a topic as it oesn t assume the IIA (inepenence of irrelevant alternatives restriction of logistic moels. The rest of this paper is organize as follows: in Section 2, we present a conventional formulation of topic moeling along with our general notation an the correlate topic moels extension. Section 3 introuces our aaptation of the iagonal orthant probit moel to topic iscovery in the presence correlations among topics, along with the corresponing auxiliary variable sampling scheme for upating the probit moel parameters an the remainer of all the posterior istributions of the parameters of the moel. Unlie with the logistic normal formulation where the non-conugacy leas to the nee for sophisticate sampling scheme, in this section we clearly reveal the simplicity of our propose metho resulting from the natural conugacy inherent in the auxiliary formulation of the upating of the parameters. We also show compelling computational emonstrations of the efficiency of the iagonal orthant approach compare to the traitional multinomial probit for on both the auxiliary variable sampling an the estimation of the topic istribution. Section 4 presents the performance of our propose approach on the Associate Press ata set, featuring the intuitively appealing topics iscovere, along with the correlation structure among topics an the loglielihoo as a function of topical space imension. Section 5 eals with our conclusion, iscussion an elements of our future wor. 2. General Aspects of Topic Moels In a given corpus, one coul imagine that each ocument eals with one or more topics. For instance, one of the collection consiere in this paper is provie by the Associate Press an covers topics as varie as aviation, eucation, weather, broacasting, air force, navy, national security, international treaties, investing, international trae, war, courts, entertainment inustry, politics, an etc. From a statistical perspective, a topic is often moele as a probability istribution over wors, an as a result a given ocument is treate as a mixture of probabilistic topics [1]. We consier a setting where we have a total of V unique wors in the reference vocabulary an topics unerlying the D ocuments provie. Let w n enote the n-th wor in the -th ocument, an let z n refer to the label of the topic assigne to the n -th wor of that -th ocument. Then the probability of w is given by where Pr ( z n ( w = ( w z = ( z = Pr Pr Pr, (1 n n n n = 1 n = is the probability that the n-th wor in the -th ocument is assigne to topic. This quantity plays an important role in the analysis of correlate topic moels. In the seminal article on correlate topic moels [2], Pr ( zn = is moele for each ocument as a function of a -imensional vector η 880

3 1 2 of parameters. Specifically, the logistic-normal efines η ( η, η,, η typically set to zero for ientifiability an assumes with η ~ MVN (, η X. C. u, E. Fooué = where the last element η is µ Σ with e 1 θ = Pr zn 1 η = = f ( η =, = 1, 2,, 1 an θ, = η η e e = 1 = 1 Also, n { N } an z ~ Mult ( θ, an w ~ ( 1, 2,, n n Mult β. With all these moel components efine, the estimation tas in correlate topic moeling from a Bayesian perspective can be summarize in the following posterior D N p( η, ZW, µ, Σ p( W Z p( zn p( η µ, Σ = 1 n= 1 N δ ( C D β 1 ( = δ β = 1 n= 1 where δ ( is efine using the Gamma function ( + z n = θ N ( η µ, Σ, ( u Gamma so for a -imension vector u, = 1 δ ( u Γ =. Γ u = 1 (3 provies the ingreients for estimating the parameter vectors η that help capture the correlations among topics, an the matrix Z that contains the topical assignments. Uner the logistic normal moel, sampling from the full posterior of η erive from the oint posterior in (3 requires the use of sophisticate sampling schemes lie the one use in [4]. Although these authors manage to achieve great performances on large corpuses of ocuments, we thought it useful to contribute to correlate topic moeling by way of the multinomial probit. Clearly, as inicate earlier, most authors concentrate on logistic-normal even espite non-conugacy, an the lac of probit topic moeling can be easily attribute to the inefficiency of the corresponing sampling scheme. In the most raw formulation of the multinomial probit that intens to capture the full exten of all the correlations among the topics, the topic assignment probability is efine by (3. ( n = = θ = φ ( η Pr z u;, R u (3 The practical evaluation of (3 involves a complicate high imensional integral which is typically computationally intractable when the number of categories is greater than 4. A relaxe version of (3, one that still captures more correlation than the logit an that is also very commonly use in practice, efines θ as + θ = ( v η η ϕ( v v ( ( V η η, Φ + =Εϕ v Φ + (4 = 1, = 1, v v where 2 φ ( v = e is the stanar normal ensity, an Φ ( v = φ ( uu is the stanar normal is- 2π tribution function. Despite this relaxation, the multinomial probit in this formulation still has maor rawbacs namely: 1 Even when one is given the vector η, the calculation of θ remains computationally prohibitive even for moerate values of. In practice, one may consier using a monte carlo approximation to that integral in (4. However, such an approach in the context of a large corpus with many unerlying latent topics reners the probit formulation almost unusable. 2 As far as the estimation of η is concerne, a natural approach to sampling from the posterior of η in this context woul be to use the Metropolis-Hastings upating scheme, since the full posterior in this case is not available. Unfortunately, the Metropolis in this case is excruciatingly slow with poor mixing rates an high sensitivity to the proposal istribution. It turns out that an apparently appealing solution in this case coul come from the auxiliary variable formulation as escribe in [6]. Un- (2 881

4 X. C. u, E. Fooué fortunately, even this promising formulation fails catastrophically for moerate values as we will emonstrate in the subsequent section, ue to the high epenency structure between auxiliary variables an parameters. Essentially, the nee for Metropolis is avoie by efining an auxiliary vector of imension. For n = 1,, N, we consier the vector z n containing the current topic allocation an we repeately sample n from a -imensional multivariate Gaussian until the component of n that correspons to the non-zero inex in z n is the largest of all the components of n, i.e. zn n = = 1,, { n} max. The conition in (5 typically fails to be fulfille even when is moerately large. In fact, we emonstrate later that in some cases, it becomes impossible to fin a vector n satisfying that conition. Besies, the epenency of n on the current value of η further complicates the sampling scheme especially in the case of large topical space. In the next section, we remey these inefficiencies by proposing an eveloping our aaptation of the iagonal orthant multinomial probit. 3. Diagonal Orthant Probit for Correlate Topic Moels In a recent wor, [5] evelope the iagonal orthant probit approach to multicategorical classification. Their approach circumvents the bottlenecs mentione earlier an substantially improves the sampling efficiency while maintaining the preictive performance. Essentially, the iagonal orthant probit approach successfully maes the most of the benefits of binary classification, thereby substantially reucing the high epenency that mae the conition (5 computationally unattainable. Inee, with the iagonal orthant multinomial moel, we achieve three main benefits A more tractable an easily computatble efinition of topic istribution θ = Pr ( zn = η A clear an very straightforwar an aaptable auxiliary variable sampling scheme Thecapacity to hanle a very large number of topics ue to the efficiency an low epenency. Uner the iagonal orthant probit moel, we have θ = ( 1 Φ( η Φ( η ( 1 Φ( η Φ( η = 1 The generative process of our probit normal topic moels is essentially ientical to logistic topic moels except that the topic istribution for each ocument now is obtaine by a probit transformation of a multivariate Gaussian variable (6. As such, the generating process of a ocument of length N is as follows: 1 Draw η ~ MVN ( µ, Σ an transform η into topic istribution θ where each element of θ is compute as follows: θ = 2 For each wor position n ( 1,, N a Draw a topic assignment Z ~ Mult ( θ n n b Draw a wor W ~ ( z n Mult φ ( 1 Φ( η Φ( η ( 1 Φ( η Φ( η = 1 Where ( η ~ N µ, Σ. Throughout this paper, we ll use φ ( Φ represents the cumulative istribution of the stanar normal. We specify a Gaussian prior for η, namely ( ( to enote the -imensional multivariate Gaussian ensity function, 1 1 Τ 1 φ ( η ; µ, Σ = exp ( η µ ( η µ. Σ 2π 2 Σ ( To complete the Bayesian analysis of our probit normal topic moel, we nee to sample from the oint posterior.. (5 (6 (7 882

5 ( η Z W µ ( η µ ( Z η ( W Z X. C. u, E. Fooué p,,, Σ p, Σ p p. (8 As note earlier, the secon benefit of the iagonal orthant probit moel lies in its clear, simple, straightforwar yet powerful auxiliary variable sampling scheme. We tae avantage of that iagonal orthant property when ealing with the full posterior for η given by ( η W Z µ ( η µ ( Z η p,,, Σ p, Σ p. (9 While sampling irectly from (9 is impractical, efining a collection of auxiliary variables allows a p η, Z, W, µ, Σ using the following: For each ocument, the matrix N contains all the values of the auxiliary variables, scheme that samples from the oint posterior ( Each row ( 1,, n n n,, n = 1 2, N 1, N 1, N 1, N 1 1 2, N, N, N, N Τ = of has components, an the iagonal orthant upates them reaily using the following straightforwar sampling scheme: Let be the current topic allocation for the nth wor. For the component of n whose inex correspons to the label of current topic assignment of wor n sample from a truncate normal istribution with variance 1 restricte to positive outcomes ( η ( η ~ N,1 z 1 n + n = For all components of n whose inices o correspon to the label of current topic assignment of wor n sample from a truncate normal istribution with variance 1 restricte to negative outcomes Once the matrix rawing where ( η ( η ~ N,1 z 1 n n is obtaine, the sampling scheme upates the parameter vector ( η A µ Σ MVN ( µ η Σ η,,, ~,, ( 1 1 ( ( 1 1 Τ Τ µ 1 η = Σ η Σ µ + X A ec Ση = Σ + X A X 1 I an v ( v an. η by conveniently with X = N ec representing the row-wise vectorization of the matrix. Aopting the fully Bayesian treatment of our probit normal correlate topic moel, we a an extra layer to the hierarchy in orer to capture the variation in the mean vector an the variance-covariance matrix of the parameter vector η. Taing avantage of conugacy, we specify a normal-inverse-wishart prior for ( µ, Σ, namely, p µ, Σ = NIW µ, κ, Ψ, ν, meaning that ν, Ψ ~ IW ( Ψ, ν ( ( Σ an ( µµ, Σ, κ ~ MVN ( µ, Σ κ normal-inverse-wishart, so that we can write where κ = κ0 + D, ν = ν0 + D, p ( µ Σ W Z η = NIW ( µ κ Ψ ν,,,,,,, D = + κ 0 µ η µ 0 D+ κ0 D+ κ0, an κ Ψ = Ψ + Q + ( η µ ( η µ 0 Τ κ0 + D,. The corresponing posterior is 883

6 X. C. u, E. Fooué where D ( ( Q = η η η η Τ. = 1 As far as sampling from the full posterior istribution of Z n is concerne, we use the expression w C n, n + βw n n = Z n n W n ( n n = W n Z n θ θ V V C, n + β = 1 = 1 Pr z 1, w, p w z 1,,. where the use of C, n is use to inicate that the n-th wor is not inclue in the topic or ocument uner consieration. 4. Computational Results on the Associate Press Data In this section, we use a famous Associate Press ata set from [7] in R to uncover the wor topic istribution, the correlation structure between various topics as well as selecting optimal moels. The Associate Press corpus consists of 2244 ocuments an 10,473 wors. After preprocessing the corpus by picing frequent an common terms, we reuce the size of the wors from 10,473 to 2643 for efficient sampling. In our first experimentation, we built a correlate topic moeling structure base on the traitional multinomial probit an then teste the computational spee for ey sampling tass. The high posterior epenency structure between auxiliary variables an parameters mae multinormal probit essentially unscalable for situations where it is impossible for the sampler to yiel a ranom variate of the auxiliary variable corresponing the current topic allocation label that is also the maximum (5. For a ranom initialization of topic assignment, the sampling of auxiliary variable cannot even complete one single iteration. In the case of goo initialization of topical prior η which leas to smooth sampling of auxiliary variables, the computational efficiency is still unesirable an we observe that for larger topical space such as = 40, the auxiliary variable stumble again after some amount of iterations, inicating even goo initialization will not ease the troublesome epenency relationship between the auxiliary variables an parameters in larger topical space. Unlie with the traitional probit moel for which the computation of θ is virtually impractical for large, the iagonal orthant approach maes this computation substantially faster ever for large. The comparison of the computational spee of two essential sampling tass between the multinomial probit moel an igonal orthant probit moel are shown as below in Table 1. Table 1. All the numbers in this table represent the processing time (in secons, an are compute in Ron PC using a parallel algorithm acting on 4 CPU cores. NA here represents situations where it is impossible for the sampler to yiel a ranom variate of the auxiliary variable corresponing the current topic allocation label that is also the maximum. Sampling Tas ( = 10 MNP DO Probit Topic Distribution θ Auxiliary variable (108 to NA 3.09 Sampling Tas ( = 20 MNP DO Probit Topic Distribution θ Auxiliary variable (334 to NA 3.39 Sampling Tas ( = 30 MNP DO Probit Topic Distribution θ Auxiliary variable (528 to NA 3.49 Sampling Tas ( = 40 MNP DO Probit Topic Distribution θ Auxiliary variable (1785 to NA

7 X. C. u, E. Fooué In aition to the rastic improvement of the overall sampling efficiency, we notice that the computational complexity for sampling the auxiliary variable an topic istribution is close to O(1 an O( respectively, suggesting that probit normal topic moel now becomes an attainable an feasible tool of the traitional correlate topic moel. Central to topic moeling is the nee to etermine for a given corpus the optimal number of latent topics. As it is the case for most latent variable moels, this tas can be formiable at times, an there is no consensus among machine learning researchers as to which of the existing methos is the best. Figure 1 shows the loglielihoo as a function of the number of topics iscovere in the moel. Apart from the loglielihoo, many other techniques are commonly use such as perplexity, harmonic mean metho an so on. As we see, the optimal number of topics in this case is 30. In Table 2, we show a subset of the 30 topics uncovere where each topic is represente by the 10 most frequent wors. It can be seen that our probit normal topic moel is able to capture the co-occurrence of wors within topics successfully. In Figure 2, we also show the correlation structure between various topics which is the essential purpose of employing the correlate topic moel. Eviently, the correlation capture intuitively reflect the natural relationship between similar topics. 5. Conclusion an Discussion In the context of topic moeling where many other researchers seem to have avoie it. By aapting the iagonal orthant probit moel, we propose a probit alternative to the logit approach to the topic moeling. Compare to the multinomial probit moel we constructe, our topic iscovery scheme using iagonal orthant probit moel enoye several esirable properties; First,we gaine the efficiency in computing the topic istribution θ ; Secon, we achieve a clear an very straightforwar an aaptable auxiliary variable sampling scheme that substantially reuce the strength of the epenence structure between auxiliary variables an moel parameters, responsible for absorbing state in the Marov chain; Thirly, as a consequence of goo mixing, our approach mae the probit moel a viable an competitive alternatives to its logistic counterpart. In aition to all these benefits, our propose metho offers a straightforwar an inherent conugacy, which helps avoi those complicate sampling schemes employe in the logistics normal probit moel. In the Associate Press example explore in the previous section, not only oes our metho prouce a better lielihoo than the logistic normal topic moel with variational EM, but also iscovers meaningful topics along with unerlying correlation structure between topics. Overall, the metho we evelope in this paper offers another feasible alternatives in the context of correlate topic moel that we hope will be further explore an extene by many other researchers. Figure 1. Log lielihoo as a function of the number of topics. 885

8 X. C. u, E. Fooué Table 2. Representation of topics iscovere by our metho. Topic 25 Topic 18 Topic 23 Topic 11 Topic 1 Topic 24 Topic 27 Wor 1 court company bush stuents tax fire air Wor 2 trial billion senate school buget water plane Wor 3 uge inc vote meese billion rain flight Wor 4 prison corp uais stuent bill northern airlines Wor 5 convicte percent percent schools percent southern pilots Wor 6 ury stoc bill teachers senate inches aircraft Wor 7 rug worers enney boar income fair planes Wor 8 guilty contract sales eucation legislation egrees airline Wor 9 fbi companies bentsen teacher taxes snow eastern Wor 10 sentence offer ticet tax bush temperatures airport Topic 6 Topic 12 Topic 20 Topic 2 Topic 22 Topic 16 Topic 15 Wor 1 percent space military soviet ai police ollar Wor 2 stoc shuttle china gorbachev rebels arreste yen Wor 3 inex soviet chinese bush contras shot rates Wor 4 billion nasa soliers reagan nicaragua shooting bi Wor 5 prices launch troops moscow contra inure prices Wor 6 rose mission saui summit saninista car price Wor 7 stocs earth trae soviets military officers lonon Wor 8 average north rebels treaty ortega bus gol Wor 9 points orean hong europe saninistas illing percent Wor 10 shares south army germany rebel arrest traing Topic 19 Topic 14 Topic 7 Topic 4 Topic 30 Topic 8 Topic 17 Wor 1 iraq trae israel navy percent south film Wor 2 uwait percent israeli ship oil africa movie Wor 3 iraqi farmers ewish coast prices african music Wor 4 german farm palestinian islan price blac theater Wor 5 gulf billion arab boat cents church actor Wor 6 germany apan palestinians ships gasoline pope actress Wor 7 saui agriculture army earthquae average manela awar Wor 8 iran apanese occupie sea offers blacs ban Wor 9 bush tons stuents scale gol aparthei boo Wor 10 military rought gaza guar crue catholic films Base on the promising results we have seen in this paper, the probit normal topic moel opens the oor for various future wors. For instance, [8] propose a multi-fiel correlate topic moel by relaxing the assumption of using common set of topics globally among all ocuments, which can also be applie to the probit moel to enrich the comprehensiveness of structural relationships between topics. Another potential irection woul be to 886

9 X. C. u, E. Fooué Figure 2. Graphical representation of the correlation among topics. enhance the scalability of the moel. Currently we use a simple istribute algorithm propose by [9] an [10] for efficient Gibbs sampling. The architecture for topic moels presente by [11] can be further utilize to reuce the computational complexity substantially while elivering comparable performance. Furthermore, a novel sampling metho involving the Gibbs Max-Margin Topic [12] will further improve the computational efficiency. Acnowlegements We want to express our sincere gratitue to our reviewer for comprehensive an constructive avice. We also wish to express our heartfelt gratitue an infinite thans to Our Lay of Perpetual Help for Her ever-present support an guiance, especially for the uninterrupte flow of inspiration receive through Her most powerful intercession. References [1] Blei, D.M. an Ng, A.., Joran, M.I. an Lafferty, J. (2003 Latent Dirichlet Allocation. Journal of Machine Learning Research, 3. [2] Blei, D.M. an Lafferty, J.D. (2006 Correlate Topic Moels. Proceeings of the 23r International Conference on Machine Learning, MIT Press, Cambrige, Massachusetts, [3] Mimno, D., Wallach, H.M. an Mccallum, A. (2008 Gibbs Sampling for Logistic Normal Topic Moels with Graph- Base Priors. Proceeings of NIPS Worshop on Analyzing Graphs, [4] Chen, J.F., Zhu, J., Wang, Z., Zheng, X. an Zhang, B. (2013 Scalable Inference for Logistic-Normal Topic Moels. In Burges, C.J.C., Bottou, L., Welling, M., Ghahramani, Z. an Weinberger,.Q., Es., Avances in Neural Information Processing Systems 26, Curran Associates, Inc., [5] Johnrow, J., Lum,. an Dunson, D.B. (2013 Diagonal Orthant Multinomial Probit Moels. JMLR Proceeings, Volume 31 of AISTATS, [6] Albert, J.H. an Chib, S. (1993 Bayesian Analysis of Binary an Polychotomous Response Data. Journal of the American Statistical Association, 88, [7] Grun, B. an Horni,. (2011 Topicmoels: An R Pacage for Fitting Topic Moels. Journal of Statistical Software, 40,

10 X. C. u, E. Fooué [8] Salomatin,., ang,.m. an La, A. (2009 Multi-Fiel Correlate Topic Moeling. Proceeings of the SIAM International Conference on Data Mining, SDM 2009, April 30-May 2, 2009, Spars, [9] ao, L.M., Mimno, D. an McCallum, A. (2009 Efficient Methos for Topic Moel Inference on Streaming Document Collections. DD 2009: Proceeings of 15th ACM SIGDD int l Conference on nowlege Discovery an Data Mining, [10] Newman, D., Asuncion, A., Smyth, P. an Welling, M. (2009 Distribute Algorithms for Topic Moels. Journal of Machine Learning Research, 10, [11] Smola, A. an Narayanamurthy, S. (2010 An Architecture for Parallel Topic Moels. Proc. VLDB Enow., 3, [12] Zhu, J., Chen, N., Perins, H. an Zhang, B. (2013 Gibbs Max-Margin Topic Moels with Data Augmentation. CoRR, abs/

11

Collapsed Variational Inference for HDP

Collapsed Variational Inference for HDP Collapse Variational Inference for HDP Yee W. Teh Davi Newman an Max Welling Publishe on NIPS 2007 Discussion le by Iulian Pruteanu Outline Introuction Hierarchical Bayesian moel for LDA Collapse VB inference

More information

Topic Modeling: Beyond Bag-of-Words

Topic Modeling: Beyond Bag-of-Words Hanna M. Wallach Cavenish Laboratory, University of Cambrige, Cambrige CB3 0HE, UK hmw26@cam.ac.u Abstract Some moels of textual corpora employ text generation methos involving n-gram statistics, while

More information

Topic Uncovering and Image Annotation via Scalable Probit Normal Correlated Topic Models

Topic Uncovering and Image Annotation via Scalable Probit Normal Correlated Topic Models Rochester Institute of Technology RIT Scholar Wors Theses Thesis/Dissertation Collections 5-2015 Topic Uncovering an Image Annotation via Scalable Probit Normal Correlate Topic Moels Xingchen Yu Follow

More information

Collapsed Gibbs and Variational Methods for LDA. Example Collapsed MoG Sampling

Collapsed Gibbs and Variational Methods for LDA. Example Collapsed MoG Sampling Case Stuy : Document Retrieval Collapse Gibbs an Variational Methos for LDA Machine Learning/Statistics for Big Data CSE599C/STAT59, University of Washington Emily Fox 0 Emily Fox February 7 th, 0 Example

More information

Lecture Introduction. 2 Examples of Measure Concentration. 3 The Johnson-Lindenstrauss Lemma. CS-621 Theory Gems November 28, 2012

Lecture Introduction. 2 Examples of Measure Concentration. 3 The Johnson-Lindenstrauss Lemma. CS-621 Theory Gems November 28, 2012 CS-6 Theory Gems November 8, 0 Lecture Lecturer: Alesaner Mąry Scribes: Alhussein Fawzi, Dorina Thanou Introuction Toay, we will briefly iscuss an important technique in probability theory measure concentration

More information

Least-Squares Regression on Sparse Spaces

Least-Squares Regression on Sparse Spaces Least-Squares Regression on Sparse Spaces Yuri Grinberg, Mahi Milani Far, Joelle Pineau School of Computer Science McGill University Montreal, Canaa {ygrinb,mmilan1,jpineau}@cs.mcgill.ca 1 Introuction

More information

7.1 Support Vector Machine

7.1 Support Vector Machine 67577 Intro. to Machine Learning Fall semester, 006/7 Lecture 7: Support Vector Machines an Kernel Functions II Lecturer: Amnon Shashua Scribe: Amnon Shashua 7. Support Vector Machine We return now to

More information

Survey Sampling. 1 Design-based Inference. Kosuke Imai Department of Politics, Princeton University. February 19, 2013

Survey Sampling. 1 Design-based Inference. Kosuke Imai Department of Politics, Princeton University. February 19, 2013 Survey Sampling Kosuke Imai Department of Politics, Princeton University February 19, 2013 Survey sampling is one of the most commonly use ata collection methos for social scientists. We begin by escribing

More information

. Using a multinomial model gives us the following equation for P d. , with respect to same length term sequences.

. Using a multinomial model gives us the following equation for P d. , with respect to same length term sequences. S 63 Lecture 8 2/2/26 Lecturer Lillian Lee Scribes Peter Babinski, Davi Lin Basic Language Moeling Approach I. Special ase of LM-base Approach a. Recap of Formulas an Terms b. Fixing θ? c. About that Multinomial

More information

A Review of Multiple Try MCMC algorithms for Signal Processing

A Review of Multiple Try MCMC algorithms for Signal Processing A Review of Multiple Try MCMC algorithms for Signal Processing Luca Martino Image Processing Lab., Universitat e València (Spain) Universia Carlos III e Mari, Leganes (Spain) Abstract Many applications

More information

Schrödinger s equation.

Schrödinger s equation. Physics 342 Lecture 5 Schröinger s Equation Lecture 5 Physics 342 Quantum Mechanics I Wenesay, February 3r, 2010 Toay we iscuss Schröinger s equation an show that it supports the basic interpretation of

More information

Lecture 2: Correlated Topic Model

Lecture 2: Correlated Topic Model Probabilistic Moels for Unsupervise Learning Spring 203 Lecture 2: Correlate Topic Moel Inference for Correlate Topic Moel Yuan Yuan First of all, let us make some claims about the parameters an variables

More information

'HVLJQ &RQVLGHUDWLRQ LQ 0DWHULDO 6HOHFWLRQ 'HVLJQ 6HQVLWLYLW\,1752'8&7,21

'HVLJQ &RQVLGHUDWLRQ LQ 0DWHULDO 6HOHFWLRQ 'HVLJQ 6HQVLWLYLW\,1752'8&7,21 Large amping in a structural material may be either esirable or unesirable, epening on the engineering application at han. For example, amping is a esirable property to the esigner concerne with limiting

More information

LDA Collapsed Gibbs Sampler, VariaNonal Inference. Task 3: Mixed Membership Models. Case Study 5: Mixed Membership Modeling

LDA Collapsed Gibbs Sampler, VariaNonal Inference. Task 3: Mixed Membership Models. Case Study 5: Mixed Membership Modeling Case Stuy 5: Mixe Membership Moeling LDA Collapse Gibbs Sampler, VariaNonal Inference Machine Learning for Big Data CSE547/STAT548, University of Washington Emily Fox May 8 th, 05 Emily Fox 05 Task : Mixe

More information

Factorized Multi-Modal Topic Model

Factorized Multi-Modal Topic Model Factorize Multi-Moal Topic Moel Seppo Virtanen 1, Yangqing Jia 2, Arto Klami 1, Trevor Darrell 2 1 Helsini Institute for Information Technology HIIT Department of Information an Compute Science, Aalto

More information

THE VAN KAMPEN EXPANSION FOR LINKED DUFFING LINEAR OSCILLATORS EXCITED BY COLORED NOISE

THE VAN KAMPEN EXPANSION FOR LINKED DUFFING LINEAR OSCILLATORS EXCITED BY COLORED NOISE Journal of Soun an Vibration (1996) 191(3), 397 414 THE VAN KAMPEN EXPANSION FOR LINKED DUFFING LINEAR OSCILLATORS EXCITED BY COLORED NOISE E. M. WEINSTEIN Galaxy Scientific Corporation, 2500 English Creek

More information

Homework 2 Solutions EM, Mixture Models, PCA, Dualitys

Homework 2 Solutions EM, Mixture Models, PCA, Dualitys Homewor Solutions EM, Mixture Moels, PCA, Dualitys CMU 0-75: Machine Learning Fall 05 http://www.cs.cmu.eu/~bapoczos/classes/ml075_05fall/ OUT: Oct 5, 05 DUE: Oct 9, 05, 0:0 AM An EM algorithm for a Mixture

More information

CUSTOMER REVIEW FEATURE EXTRACTION Heng Ren, Jingye Wang, and Tony Wu

CUSTOMER REVIEW FEATURE EXTRACTION Heng Ren, Jingye Wang, and Tony Wu CUSTOMER REVIEW FEATURE EXTRACTION Heng Ren, Jingye Wang, an Tony Wu Abstract Popular proucts often have thousans of reviews that contain far too much information for customers to igest. Our goal for the

More information

Lower Bounds for the Smoothed Number of Pareto optimal Solutions

Lower Bounds for the Smoothed Number of Pareto optimal Solutions Lower Bouns for the Smoothe Number of Pareto optimal Solutions Tobias Brunsch an Heiko Röglin Department of Computer Science, University of Bonn, Germany brunsch@cs.uni-bonn.e, heiko@roeglin.org Abstract.

More information

An Optimal Algorithm for Bandit and Zero-Order Convex Optimization with Two-Point Feedback

An Optimal Algorithm for Bandit and Zero-Order Convex Optimization with Two-Point Feedback Journal of Machine Learning Research 8 07) - Submitte /6; Publishe 5/7 An Optimal Algorithm for Banit an Zero-Orer Convex Optimization with wo-point Feeback Oha Shamir Department of Computer Science an

More information

IN the evolution of the Internet, there have been

IN the evolution of the Internet, there have been 1 Tag-Weighte Topic Moel For Large-scale Semi-Structure Documents Shuangyin Li, Jiefei Li, Guan Huang, Ruiyang Tan, an Rong Pan arxiv:1507.08396v1 [cs.cl] 30 Jul 2015 Abstract To ate, there have been massive

More information

Linear First-Order Equations

Linear First-Order Equations 5 Linear First-Orer Equations Linear first-orer ifferential equations make up another important class of ifferential equations that commonly arise in applications an are relatively easy to solve (in theory)

More information

Introduction to Machine Learning

Introduction to Machine Learning How o you estimate p(y x)? Outline Contents Introuction to Machine Learning Logistic Regression Varun Chanola April 9, 207 Generative vs. Discriminative Classifiers 2 Logistic Regression 2 3 Logistic Regression

More information

Robust Forward Algorithms via PAC-Bayes and Laplace Distributions. ω Q. Pr (y(ω x) < 0) = Pr A k

Robust Forward Algorithms via PAC-Bayes and Laplace Distributions. ω Q. Pr (y(ω x) < 0) = Pr A k A Proof of Lemma 2 B Proof of Lemma 3 Proof: Since the support of LL istributions is R, two such istributions are equivalent absolutely continuous with respect to each other an the ivergence is well-efine

More information

Expected Value of Partial Perfect Information

Expected Value of Partial Perfect Information Expecte Value of Partial Perfect Information Mike Giles 1, Takashi Goa 2, Howar Thom 3 Wei Fang 1, Zhenru Wang 1 1 Mathematical Institute, University of Oxfor 2 School of Engineering, University of Tokyo

More information

WEIGHTING A RESAMPLED PARTICLE IN SEQUENTIAL MONTE CARLO. L. Martino, V. Elvira, F. Louzada

WEIGHTING A RESAMPLED PARTICLE IN SEQUENTIAL MONTE CARLO. L. Martino, V. Elvira, F. Louzada WEIGHTIG A RESAMPLED PARTICLE I SEQUETIAL MOTE CARLO L. Martino, V. Elvira, F. Louzaa Dep. of Signal Theory an Communic., Universia Carlos III e Mari, Leganés (Spain). Institute of Mathematical Sciences

More information

Time-of-Arrival Estimation in Non-Line-Of-Sight Environments

Time-of-Arrival Estimation in Non-Line-Of-Sight Environments 2 Conference on Information Sciences an Systems, The Johns Hopkins University, March 2, 2 Time-of-Arrival Estimation in Non-Line-Of-Sight Environments Sinan Gezici, Hisashi Kobayashi an H. Vincent Poor

More information

Introduction. A Dirichlet Form approach to MCMC Optimal Scaling. MCMC idea

Introduction. A Dirichlet Form approach to MCMC Optimal Scaling. MCMC idea Introuction A Dirichlet Form approach to MCMC Optimal Scaling Markov chain Monte Carlo (MCMC quotes: Metropolis et al. (1953, running coe on the Los Alamos MANIAC: a feasible approach to statistical mechanics

More information

KNN Particle Filters for Dynamic Hybrid Bayesian Networks

KNN Particle Filters for Dynamic Hybrid Bayesian Networks KNN Particle Filters for Dynamic Hybri Bayesian Networs H. D. Chen an K. C. Chang Dept. of Systems Engineering an Operations Research George Mason University MS 4A6, 4400 University Dr. Fairfax, VA 22030

More information

Transmission Line Matrix (TLM) network analogues of reversible trapping processes Part B: scaling and consistency

Transmission Line Matrix (TLM) network analogues of reversible trapping processes Part B: scaling and consistency Transmission Line Matrix (TLM network analogues of reversible trapping processes Part B: scaling an consistency Donar e Cogan * ANC Eucation, 308-310.A. De Mel Mawatha, Colombo 3, Sri Lanka * onarecogan@gmail.com

More information

Optimization of Geometries by Energy Minimization

Optimization of Geometries by Energy Minimization Optimization of Geometries by Energy Minimization by Tracy P. Hamilton Department of Chemistry University of Alabama at Birmingham Birmingham, AL 3594-140 hamilton@uab.eu Copyright Tracy P. Hamilton, 1997.

More information

Conservation Laws. Chapter Conservation of Energy

Conservation Laws. Chapter Conservation of Energy 20 Chapter 3 Conservation Laws In orer to check the physical consistency of the above set of equations governing Maxwell-Lorentz electroynamics [(2.10) an (2.12) or (1.65) an (1.68)], we examine the action

More information

Chapter 6: Energy-Momentum Tensors

Chapter 6: Energy-Momentum Tensors 49 Chapter 6: Energy-Momentum Tensors This chapter outlines the general theory of energy an momentum conservation in terms of energy-momentum tensors, then applies these ieas to the case of Bohm's moel.

More information

Construction of the Electronic Radial Wave Functions and Probability Distributions of Hydrogen-like Systems

Construction of the Electronic Radial Wave Functions and Probability Distributions of Hydrogen-like Systems Construction of the Electronic Raial Wave Functions an Probability Distributions of Hyrogen-like Systems Thomas S. Kuntzleman, Department of Chemistry Spring Arbor University, Spring Arbor MI 498 tkuntzle@arbor.eu

More information

A Course in Machine Learning

A Course in Machine Learning A Course in Machine Learning Hal Daumé III 12 EFFICIENT LEARNING So far, our focus has been on moels of learning an basic algorithms for those moels. We have not place much emphasis on how to learn quickly.

More information

Quantum mechanical approaches to the virial

Quantum mechanical approaches to the virial Quantum mechanical approaches to the virial S.LeBohec Department of Physics an Astronomy, University of Utah, Salt Lae City, UT 84112, USA Date: June 30 th 2015 In this note, we approach the virial from

More information

d dx But have you ever seen a derivation of these results? We ll prove the first result below. cos h 1

d dx But have you ever seen a derivation of these results? We ll prove the first result below. cos h 1 Lecture 5 Some ifferentiation rules Trigonometric functions (Relevant section from Stewart, Seventh Eition: Section 3.3) You all know that sin = cos cos = sin. () But have you ever seen a erivation of

More information

Gaussian processes with monotonicity information

Gaussian processes with monotonicity information Gaussian processes with monotonicity information Anonymous Author Anonymous Author Unknown Institution Unknown Institution Abstract A metho for using monotonicity information in multivariate Gaussian process

More information

Inter-domain Gaussian Processes for Sparse Inference using Inducing Features

Inter-domain Gaussian Processes for Sparse Inference using Inducing Features Inter-omain Gaussian Processes for Sparse Inference using Inucing Features Miguel Lázaro-Greilla an Aníbal R. Figueiras-Vial Dep. Signal Processing & Communications Universia Carlos III e Mari, SPAIN {miguel,arfv}@tsc.uc3m.es

More information

The Exact Form and General Integrating Factors

The Exact Form and General Integrating Factors 7 The Exact Form an General Integrating Factors In the previous chapters, we ve seen how separable an linear ifferential equations can be solve using methos for converting them to forms that can be easily

More information

Math Notes on differentials, the Chain Rule, gradients, directional derivative, and normal vectors

Math Notes on differentials, the Chain Rule, gradients, directional derivative, and normal vectors Math 18.02 Notes on ifferentials, the Chain Rule, graients, irectional erivative, an normal vectors Tangent plane an linear approximation We efine the partial erivatives of f( xy, ) as follows: f f( x+

More information

TMA 4195 Matematisk modellering Exam Tuesday December 16, :00 13:00 Problems and solution with additional comments

TMA 4195 Matematisk modellering Exam Tuesday December 16, :00 13:00 Problems and solution with additional comments Problem F U L W D g m 3 2 s 2 0 0 0 0 2 kg 0 0 0 0 0 0 Table : Dimension matrix TMA 495 Matematisk moellering Exam Tuesay December 6, 2008 09:00 3:00 Problems an solution with aitional comments The necessary

More information

A. Exclusive KL View of the MLE

A. Exclusive KL View of the MLE A. Exclusive KL View of the MLE Lets assume a change-of-variable moel p Z z on the ranom variable Z R m, such as the one use in Dinh et al. 2017: z 0 p 0 z 0 an z = ψz 0, where ψ is an invertible function

More information

Capacity Analysis of MIMO Systems with Unknown Channel State Information

Capacity Analysis of MIMO Systems with Unknown Channel State Information Capacity Analysis of MIMO Systems with Unknown Channel State Information Jun Zheng an Bhaskar D. Rao Dept. of Electrical an Computer Engineering University of California at San Diego e-mail: juzheng@ucs.eu,

More information

Fast Inference and Learning for Modeling Documents with a Deep Boltzmann Machine

Fast Inference and Learning for Modeling Documents with a Deep Boltzmann Machine Fast Inference and Learning for Modeling Documents with a Deep Boltzmann Machine Nitish Srivastava nitish@cs.toronto.edu Ruslan Salahutdinov rsalahu@cs.toronto.edu Geoffrey Hinton hinton@cs.toronto.edu

More information

The Principle of Least Action

The Principle of Least Action Chapter 7. The Principle of Least Action 7.1 Force Methos vs. Energy Methos We have so far stuie two istinct ways of analyzing physics problems: force methos, basically consisting of the application of

More information

The Role of Models in Model-Assisted and Model- Dependent Estimation for Domains and Small Areas

The Role of Models in Model-Assisted and Model- Dependent Estimation for Domains and Small Areas The Role of Moels in Moel-Assiste an Moel- Depenent Estimation for Domains an Small Areas Risto Lehtonen University of Helsini Mio Myrsylä University of Pennsylvania Carl-Eri Särnal University of Montreal

More information

ensembles When working with density operators, we can use this connection to define a generalized Bloch vector: v x Tr x, v y Tr y

ensembles When working with density operators, we can use this connection to define a generalized Bloch vector: v x Tr x, v y Tr y Ph195a lecture notes, 1/3/01 Density operators for spin- 1 ensembles So far in our iscussion of spin- 1 systems, we have restricte our attention to the case of pure states an Hamiltonian evolution. Toay

More information

Introduction to Markov Processes

Introduction to Markov Processes Introuction to Markov Processes Connexions moule m44014 Zzis law Gustav) Meglicki, Jr Office of the VP for Information Technology Iniana University RCS: Section-2.tex,v 1.24 2012/12/21 18:03:08 gustav

More information

Robust Low Rank Kernel Embeddings of Multivariate Distributions

Robust Low Rank Kernel Embeddings of Multivariate Distributions Robust Low Rank Kernel Embeings of Multivariate Distributions Le Song, Bo Dai College of Computing, Georgia Institute of Technology lsong@cc.gatech.eu, boai@gatech.eu Abstract Kernel embeing of istributions

More information

Multi-View Clustering via Canonical Correlation Analysis

Multi-View Clustering via Canonical Correlation Analysis Keywors: multi-view learning, clustering, canonical correlation analysis Abstract Clustering ata in high-imensions is believe to be a har problem in general. A number of efficient clustering algorithms

More information

Agmon Kolmogorov Inequalities on l 2 (Z d )

Agmon Kolmogorov Inequalities on l 2 (Z d ) Journal of Mathematics Research; Vol. 6, No. ; 04 ISSN 96-9795 E-ISSN 96-9809 Publishe by Canaian Center of Science an Eucation Agmon Kolmogorov Inequalities on l (Z ) Arman Sahovic Mathematics Department,

More information

Homework 2 EM, Mixture Models, PCA, Dualitys

Homework 2 EM, Mixture Models, PCA, Dualitys Homework 2 EM, Mixture Moels, PCA, Dualitys CMU 10-715: Machine Learning (Fall 2015) http://www.cs.cmu.eu/~bapoczos/classes/ml10715_2015fall/ OUT: Oct 5, 2015 DUE: Oct 19, 2015, 10:20 AM Guielines The

More information

Thermal conductivity of graded composites: Numerical simulations and an effective medium approximation

Thermal conductivity of graded composites: Numerical simulations and an effective medium approximation JOURNAL OF MATERIALS SCIENCE 34 (999)5497 5503 Thermal conuctivity of grae composites: Numerical simulations an an effective meium approximation P. M. HUI Department of Physics, The Chinese University

More information

ELEC3114 Control Systems 1

ELEC3114 Control Systems 1 ELEC34 Control Systems Linear Systems - Moelling - Some Issues Session 2, 2007 Introuction Linear systems may be represente in a number of ifferent ways. Figure shows the relationship between various representations.

More information

The total derivative. Chapter Lagrangian and Eulerian approaches

The total derivative. Chapter Lagrangian and Eulerian approaches Chapter 5 The total erivative 51 Lagrangian an Eulerian approaches The representation of a flui through scalar or vector fiels means that each physical quantity uner consieration is escribe as a function

More information

Survey-weighted Unit-Level Small Area Estimation

Survey-weighted Unit-Level Small Area Estimation Survey-weighte Unit-Level Small Area Estimation Jan Pablo Burgar an Patricia Dörr Abstract For evience-base regional policy making, geographically ifferentiate estimates of socio-economic inicators are

More information

Estimating Causal Direction and Confounding Of Two Discrete Variables

Estimating Causal Direction and Confounding Of Two Discrete Variables Estimating Causal Direction an Confouning Of Two Discrete Variables This inspire further work on the so calle aitive noise moels. Hoyer et al. (2009) extene Shimizu s ientifiaarxiv:1611.01504v1 [stat.ml]

More information

Nonlinear Adaptive Ship Course Tracking Control Based on Backstepping and Nussbaum Gain

Nonlinear Adaptive Ship Course Tracking Control Based on Backstepping and Nussbaum Gain Nonlinear Aaptive Ship Course Tracking Control Base on Backstepping an Nussbaum Gain Jialu Du, Chen Guo Abstract A nonlinear aaptive controller combining aaptive Backstepping algorithm with Nussbaum gain

More information

Lower bounds on Locality Sensitive Hashing

Lower bounds on Locality Sensitive Hashing Lower bouns on Locality Sensitive Hashing Rajeev Motwani Assaf Naor Rina Panigrahy Abstract Given a metric space (X, X ), c 1, r > 0, an p, q [0, 1], a istribution over mappings H : X N is calle a (r,

More information

Harmonic Modelling of Thyristor Bridges using a Simplified Time Domain Method

Harmonic Modelling of Thyristor Bridges using a Simplified Time Domain Method 1 Harmonic Moelling of Thyristor Briges using a Simplifie Time Domain Metho P. W. Lehn, Senior Member IEEE, an G. Ebner Abstract The paper presents time omain methos for harmonic analysis of a 6-pulse

More information

Problem Sheet 2: Eigenvalues and eigenvectors and their use in solving linear ODEs

Problem Sheet 2: Eigenvalues and eigenvectors and their use in solving linear ODEs Problem Sheet 2: Eigenvalues an eigenvectors an their use in solving linear ODEs If you fin any typos/errors in this problem sheet please email jk28@icacuk The material in this problem sheet is not examinable

More information

arxiv: v1 [math.co] 29 May 2009

arxiv: v1 [math.co] 29 May 2009 arxiv:0905.4913v1 [math.co] 29 May 2009 simple Havel-Hakimi type algorithm to realize graphical egree sequences of irecte graphs Péter L. Erős an István Miklós. Rényi Institute of Mathematics, Hungarian

More information

Topic 7: Convergence of Random Variables

Topic 7: Convergence of Random Variables Topic 7: Convergence of Ranom Variables Course 003, 2016 Page 0 The Inference Problem So far, our starting point has been a given probability space (S, F, P). We now look at how to generate information

More information

Integrated Data Reconciliation with Generic Model Control for the Steel Pickling Process

Integrated Data Reconciliation with Generic Model Control for the Steel Pickling Process Korean J. Chem. Eng., (6), 985-99 (3) Integrate Data Reconciliation with Generic Moel Control for the Steel Picling Process Paisan Kittisupaorn an Pornsiri Kaewprait Department of Chemical Engineering,

More information

Tutorial on Maximum Likelyhood Estimation: Parametric Density Estimation

Tutorial on Maximum Likelyhood Estimation: Parametric Density Estimation Tutorial on Maximum Likelyhoo Estimation: Parametric Density Estimation Suhir B Kylasa 03/13/2014 1 Motivation Suppose one wishes to etermine just how biase an unfair coin is. Call the probability of tossing

More information

27 : Distributed Monte Carlo Markov Chain. 1 Recap of MCMC and Naive Parallel Gibbs Sampling

27 : Distributed Monte Carlo Markov Chain. 1 Recap of MCMC and Naive Parallel Gibbs Sampling 10-708: Probabilistic Graphical Models 10-708, Spring 2014 27 : Distributed Monte Carlo Markov Chain Lecturer: Eric P. Xing Scribes: Pengtao Xie, Khoa Luu In this scribe, we are going to review the Parallel

More information

Examining Geometric Integration for Propagating Orbit Trajectories with Non-Conservative Forcing

Examining Geometric Integration for Propagating Orbit Trajectories with Non-Conservative Forcing Examining Geometric Integration for Propagating Orbit Trajectories with Non-Conservative Forcing Course Project for CDS 05 - Geometric Mechanics John M. Carson III California Institute of Technology June

More information

3.7 Implicit Differentiation -- A Brief Introduction -- Student Notes

3.7 Implicit Differentiation -- A Brief Introduction -- Student Notes Fin these erivatives of these functions: y.7 Implicit Differentiation -- A Brief Introuction -- Stuent Notes tan y sin tan = sin y e = e = Write the inverses of these functions: y tan y sin How woul we

More information

arxiv: v1 [physics.flu-dyn] 8 May 2014

arxiv: v1 [physics.flu-dyn] 8 May 2014 Energetics of a flui uner the Boussinesq approximation arxiv:1405.1921v1 [physics.flu-yn] 8 May 2014 Kiyoshi Maruyama Department of Earth an Ocean Sciences, National Defense Acaemy, Yokosuka, Kanagawa

More information

LATTICE-BASED D-OPTIMUM DESIGN FOR FOURIER REGRESSION

LATTICE-BASED D-OPTIMUM DESIGN FOR FOURIER REGRESSION The Annals of Statistics 1997, Vol. 25, No. 6, 2313 2327 LATTICE-BASED D-OPTIMUM DESIGN FOR FOURIER REGRESSION By Eva Riccomagno, 1 Rainer Schwabe 2 an Henry P. Wynn 1 University of Warwick, Technische

More information

A simple model for the small-strain behaviour of soils

A simple model for the small-strain behaviour of soils A simple moel for the small-strain behaviour of soils José Jorge Naer Department of Structural an Geotechnical ngineering, Polytechnic School, University of São Paulo 05508-900, São Paulo, Brazil, e-mail:

More information

Cascaded redundancy reduction

Cascaded redundancy reduction Network: Comput. Neural Syst. 9 (1998) 73 84. Printe in the UK PII: S0954-898X(98)88342-5 Cascae reunancy reuction Virginia R e Sa an Geoffrey E Hinton Department of Computer Science, University of Toronto,

More information

On Topic Evolution. Eric P. Xing School of Computer Science Carnegie Mellon University Technical Report: CMU-CALD

On Topic Evolution. Eric P. Xing School of Computer Science Carnegie Mellon University Technical Report: CMU-CALD On Topic Evolution Eric P. Xing School of Computer Science Carnegie Mellon University epxing@cs.cmu.eu Technical Report: CMU-CALD-05-5 December 005 Abstract I introuce topic evolution moels for longituinal

More information

Some Examples. Uniform motion. Poisson processes on the real line

Some Examples. Uniform motion. Poisson processes on the real line Some Examples Our immeiate goal is to see some examples of Lévy processes, an/or infinitely-ivisible laws on. Uniform motion Choose an fix a nonranom an efine X := for all (1) Then, {X } is a [nonranom]

More information

Scalable Inference for Logistic-Normal Topic Models

Scalable Inference for Logistic-Normal Topic Models Scalable Inference for Logistic-Normal Topic Moels Jianfei Chen, Jun Zhu, Zi Wang, Xun Zheng an Bo Zhang State ey Lab of Intelligent Tech. & Systems; Tsinghua National TNList Lab; Department of Computer

More information

A simple tranformation of copulas

A simple tranformation of copulas A simple tranformation of copulas V. Durrleman, A. Nikeghbali & T. Roncalli Groupe e Recherche Opérationnelle Créit Lyonnais France July 31, 2000 Abstract We stuy how copulas properties are moifie after

More information

Calculus in the AP Physics C Course The Derivative

Calculus in the AP Physics C Course The Derivative Limits an Derivatives Calculus in the AP Physics C Course The Derivative In physics, the ieas of the rate change of a quantity (along with the slope of a tangent line) an the area uner a curve are essential.

More information

Code_Aster. Detection of the singularities and computation of a card of size of elements

Code_Aster. Detection of the singularities and computation of a card of size of elements Titre : Détection es singularités et calcul une carte [...] Date : 0/0/0 Page : /6 Responsable : Josselin DLMAS Clé : R4.0.04 Révision : 9755 Detection of the singularities an computation of a car of size

More information

Code_Aster. Detection of the singularities and calculation of a map of size of elements

Code_Aster. Detection of the singularities and calculation of a map of size of elements Titre : Détection es singularités et calcul une carte [...] Date : 0/0/0 Page : /6 Responsable : DLMAS Josselin Clé : R4.0.04 Révision : Detection of the singularities an calculation of a map of size of

More information

More from Lesson 6 The Limit Definition of the Derivative and Rules for Finding Derivatives.

More from Lesson 6 The Limit Definition of the Derivative and Rules for Finding Derivatives. Math 1314 ONLINE More from Lesson 6 The Limit Definition of the Derivative an Rules for Fining Derivatives Eample 4: Use the Four-Step Process for fining the erivative of the function Then fin f (1) f(

More information

A NONLINEAR SOURCE SEPARATION APPROACH FOR THE NICOLSKY-EISENMAN MODEL

A NONLINEAR SOURCE SEPARATION APPROACH FOR THE NICOLSKY-EISENMAN MODEL 6th European Signal Processing Conference EUSIPCO 28, Lausanne, Switzerlan, August 25-29, 28, copyright by EURASIP A NONLINEAR SOURCE SEPARATION APPROACH FOR THE NICOLSKY-EISENMAN MODEL Leonaro Tomazeli

More information

Some properties of random staircase tableaux

Some properties of random staircase tableaux Some properties of ranom staircase tableaux Sanrine Dasse Hartaut Pawe l Hitczenko Downloae /4/7 to 744940 Reistribution subject to SIAM license or copyright; see http://wwwsiamorg/journals/ojsaphp Abstract

More information

Leaving Randomness to Nature: d-dimensional Product Codes through the lens of Generalized-LDPC codes

Leaving Randomness to Nature: d-dimensional Product Codes through the lens of Generalized-LDPC codes Leaving Ranomness to Nature: -Dimensional Prouct Coes through the lens of Generalize-LDPC coes Tavor Baharav, Kannan Ramchanran Dept. of Electrical Engineering an Computer Sciences, U.C. Berkeley {tavorb,

More information

Polynomial Inclusion Functions

Polynomial Inclusion Functions Polynomial Inclusion Functions E. e Weert, E. van Kampen, Q. P. Chu, an J. A. Muler Delft University of Technology, Faculty of Aerospace Engineering, Control an Simulation Division E.eWeert@TUDelft.nl

More information

IMAGE classification is a topic of significant interest within

IMAGE classification is a topic of significant interest within IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 35, NO. 11, NOVEMBER 2013 2665 Latent Dirichlet Allocation Moels for Image Classification Nihil Rasiwasia, Member, IEEE, an Nuno Vasconcelos,

More information

MODELLING DEPENDENCE IN INSURANCE CLAIMS PROCESSES WITH LÉVY COPULAS ABSTRACT KEYWORDS

MODELLING DEPENDENCE IN INSURANCE CLAIMS PROCESSES WITH LÉVY COPULAS ABSTRACT KEYWORDS MODELLING DEPENDENCE IN INSURANCE CLAIMS PROCESSES WITH LÉVY COPULAS BY BENJAMIN AVANZI, LUKE C. CASSAR AND BERNARD WONG ABSTRACT In this paper we investigate the potential of Lévy copulas as a tool for

More information

Generalizing Kronecker Graphs in order to Model Searchable Networks

Generalizing Kronecker Graphs in order to Model Searchable Networks Generalizing Kronecker Graphs in orer to Moel Searchable Networks Elizabeth Boine, Babak Hassibi, Aam Wierman California Institute of Technology Pasaena, CA 925 Email: {eaboine, hassibi, aamw}@caltecheu

More information

Table of Common Derivatives By David Abraham

Table of Common Derivatives By David Abraham Prouct an Quotient Rules: Table of Common Derivatives By Davi Abraham [ f ( g( ] = [ f ( ] g( + f ( [ g( ] f ( = g( [ f ( ] g( g( f ( [ g( ] Trigonometric Functions: sin( = cos( cos( = sin( tan( = sec

More information

A Note on Exact Solutions to Linear Differential Equations by the Matrix Exponential

A Note on Exact Solutions to Linear Differential Equations by the Matrix Exponential Avances in Applie Mathematics an Mechanics Av. Appl. Math. Mech. Vol. 1 No. 4 pp. 573-580 DOI: 10.4208/aamm.09-m0946 August 2009 A Note on Exact Solutions to Linear Differential Equations by the Matrix

More information

Local Linear ICA for Mutual Information Estimation in Feature Selection

Local Linear ICA for Mutual Information Estimation in Feature Selection Local Linear ICA for Mutual Information Estimation in Feature Selection Tian Lan, Deniz Erogmus Department of Biomeical Engineering, OGI, Oregon Health & Science University, Portlan, Oregon, USA E-mail:

More information

STATISTICAL LIKELIHOOD REPRESENTATIONS OF PRIOR KNOWLEDGE IN MACHINE LEARNING

STATISTICAL LIKELIHOOD REPRESENTATIONS OF PRIOR KNOWLEDGE IN MACHINE LEARNING STATISTICAL LIKELIHOOD REPRESENTATIONS OF PRIOR KNOWLEDGE IN MACHINE LEARNING Mark A. Kon Department of Mathematics an Statistics Boston University Boston, MA 02215 email: mkon@bu.eu Anrzej Przybyszewski

More information

Computing Exact Confidence Coefficients of Simultaneous Confidence Intervals for Multinomial Proportions and their Functions

Computing Exact Confidence Coefficients of Simultaneous Confidence Intervals for Multinomial Proportions and their Functions Working Paper 2013:5 Department of Statistics Computing Exact Confience Coefficients of Simultaneous Confience Intervals for Multinomial Proportions an their Functions Shaobo Jin Working Paper 2013:5

More information

Similarity Measures for Categorical Data A Comparative Study. Technical Report

Similarity Measures for Categorical Data A Comparative Study. Technical Report Similarity Measures for Categorical Data A Comparative Stuy Technical Report Department of Computer Science an Engineering University of Minnesota 4-92 EECS Builing 200 Union Street SE Minneapolis, MN

More information

Influence of weight initialization on multilayer perceptron performance

Influence of weight initialization on multilayer perceptron performance Influence of weight initialization on multilayer perceptron performance M. Karouia (1,2) T. Denœux (1) R. Lengellé (1) (1) Université e Compiègne U.R.A. CNRS 817 Heuiasyc BP 649 - F-66 Compiègne ceex -

More information

u!i = a T u = 0. Then S satisfies

u!i = a T u = 0. Then S satisfies Deterministic Conitions for Subspace Ientifiability from Incomplete Sampling Daniel L Pimentel-Alarcón, Nigel Boston, Robert D Nowak University of Wisconsin-Maison Abstract Consier an r-imensional subspace

More information

Lecture 6: Calculus. In Song Kim. September 7, 2011

Lecture 6: Calculus. In Song Kim. September 7, 2011 Lecture 6: Calculus In Song Kim September 7, 20 Introuction to Differential Calculus In our previous lecture we came up with several ways to analyze functions. We saw previously that the slope of a linear

More information

MULTIFRACTAL NETWORK GENERATORS

MULTIFRACTAL NETWORK GENERATORS MULTIFRACTAL NETWORK GENERATORS AUSTIN R. BENSON, CARLOS RIQUELME, SVEN P. SCHMIT (0) Abstract. Generating ranom graphs to moel networks has a rich history. In this paper, we explore a recent generative

More information

Modelling and simulation of dependence structures in nonlife insurance with Bernstein copulas

Modelling and simulation of dependence structures in nonlife insurance with Bernstein copulas Moelling an simulation of epenence structures in nonlife insurance with Bernstein copulas Prof. Dr. Dietmar Pfeifer Dept. of Mathematics, University of Olenburg an AON Benfiel, Hamburg Dr. Doreen Straßburger

More information

Part I: Web Structure Mining Chapter 1: Information Retrieval and Web Search

Part I: Web Structure Mining Chapter 1: Information Retrieval and Web Search Part I: Web Structure Mining Chapter : Information Retrieval an Web Search The Web Challenges Crawling the Web Inexing an Keywor Search Evaluating Search Quality Similarity Search The Web Challenges Tim

More information