DOCSLIB.ORG

Generative Vs. Discriminative Classifiers Naïve Bayes Vs Logistic

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Generative Vs. Discriminative Classifiers Naïve Bayes Vs Logistic Generative vs. Discriminative Classifiers z Goal: Wish to learn f: X → Y, e.g., P(Y|X) z Generative classifiers (e.g., Naïve Bayes): Yi z Assume some functional form for P(X|Y), P(Y) This is a ‘generative’ model of the data! Xi z Estimate parameters of P(X|Y), P(Y) directly from training data z Use Bayes rule to calculate P(Y|X= x) z Discriminative classifiers: Yi z Directly assume some functional form for P(Y|X) This is a ‘discriminative’ model of the data! Xi z Estimate parameters of P(Y|X) directly from training data © Eric Xing @ CMU, 2006-2011 1 Naïve Bayes vs Logistic Regression z Consider Y boolean, X continuous, X=<X1 ... Xm> z Number of parameters to estimate: ⎪⎧ ⎛ 1 ⎞⎪⎫ π exp − ⎜ (x − µ )2 − logσ − C ⎟ k ⎨ ∑ j ⎜ 2 j k , j k , j ⎟⎬ ⎪ ⎝ 2σ k , j ⎠⎪ NB: p(y | x) = ⎩ ⎭ ** ⎧ ⎛ ⎞⎫ ⎪ ⎜ 1 2 ⎟⎪ π k ' exp⎨− (x j − µk ', j ) − logσ k ', j − C ⎬ ∑∑kj' ⎜ 2σ 2 ⎟ ⎩⎪ ⎝ k ', j ⎠⎭⎪ LR: 1 µ(x) = T 1 + e−θ x z Estimation method: z NB parameter estimates are uncoupled z LR parameter estimates are coupled © Eric Xing @ CMU, 2006-2011 2 1 Naïve Bayes vs Logistic Regression z Asymptotic comparison (# training examples → infinity) z when model assumptions correct z NB, LR produce identical classifiers z when model assumptions incorrect z LR is less biased – does not assume conditional independence z therefore expected to outperform NB © Eric Xing @ CMU, 2006-2011 3 Naïve Bayes vs Logistic Regression z Non-asymptotic analysis (see [Ng & Jordan, 2002] ) z convergence rate of parameter estimates – how many training examples needed to assure good estimates? NB order log m (where m = # of attributes in X) LR order m z NB converges more quickly to its (perhaps less helpful) asymptotic estimates © Eric Xing @ CMU, 2006-2011 4 2 Rate of convergence: logistic regression z Let hDis,m be logistic regression trained on n examples in m dimensions. Then with higgph probability : z Implication: if we want for some small constant ε0, it suffices to pick order m examples Æ Convergences to its asymptotic classifier, in order m examples z result follows from Vapnik’s structural risk bound, plus fact that the "VC Dimension" of an m-dimensional linear separators is m © Eric Xing @ CMU, 2006-2011 5 Rate of convergence: naïve Bayes parameters z Let any ε1, δ>0, and any n ≥ 0 be fixed. Assume that for some fixed ρ0 >0> 0, we have that z Let z Then with ppyrobability at least 1-δ, after n examples: 1. For discrete input, for all i and b 2. For continuous inputs, for all i and b © Eric Xing @ CMU, 2006-2011 6 3 Some experiments from UCI data sets © Eric Xing @ CMU, 2006-2011 7 Summary z Naïve Bayes classifier z What’s the assumption z Why we use it z How do we learn it z Logistic regression z Functional form follows from Naïve Bayes assumptions z For Gaussian Naïve Bayes assuming variance z For discrete-valued Naïve Bayes too z But training procedure picks parameters without the conditional independence assumption z Gradient ascent/descent z – General approach when closed-form solutions unavailable z Generative vs. Discriminative classifiers z – Bias vs. variance tradeoff © Eric Xing @ CMU, 2006-2011 8 4 Machine Learning 1010--701/15701/15--781,781, Fall 2011 Linear Regression and Sparsity Eric Xing Lecture 4, September 21, 2011 Reading: © Eric Xing @ CMU, 2006-2011 9 Machine learning for apartment hunting z Now you've moved to Pittsburgh!! And you want to find the most reasonably priced apartment satisfying your needs: square-ft., # of bedroom, distance to campus … Living area (ft2) # bedroom Rent ($) 230 1 600 506 2 1000 433 2 1100 109 1 500 … 150 1 ? 270 1.5 ? © Eric Xing @ CMU, 2006-2011 10 5 The learning problem z Features: t z Living area, distance to campus, # bedroom … ren z Denote as x=[x1, x2, … xk] z Target: z Rent Living area z Denoted as y z Training set: 1 2 k ⎡− − x1 − −⎤ ⎡x1 x1 K x1 ⎤ ⎢ ⎥ ⎢ 1 2 k ⎥ − − x − − x x x X = ⎢ 2 ⎥ = ⎢ 2 2 K 2 ⎥ rent ⎢ M M M ⎥ ⎢ M M M M ⎥ ⎢ ⎥ ⎢ 1 2 k ⎥ ⎣− − xn − −⎦ ⎣⎢xn xn K xn ⎦⎥ ⎡ y1 ⎤ ⎡− y1 −⎤ ⎢y ⎥ ⎢− y −⎥ Y = ⎢ 2 ⎥ or ⎢ 2 ⎥ Location ⎢ M ⎥ ⎢M M M ⎥ ⎢ ⎥ ⎢ ⎥ ⎣yn ⎦ ⎣− y n −⎦ Living area © Eric Xing @ CMU, 2006-2011 11 Linear Regression z Assume that Y (target) is a linear function of X (features): z e.g.: 1 2 yˆ = θ0 +θ1x +θ2 x z let's assume a vacuous "feature" X0=1 (this is the intercept term, why?), and define the feature vector to be: z then we have the following general representation of the linear function: z Our goal is to pick the optimal θ . How! z We seek θ that minimize the following cost function: 1 n ˆ v 2 J (θ ) = ∑(yi (xi ) − yi ) 2 i=1 © Eric Xing @ CMU, 2006-2011 12 6 The Least-Mean-Square (LMS) method z The Cost Function: n 1 T 2 J (θ ) = ∑(xi θ − yi ) 2 i=1 z Consider a gradient descent algorithm: ∂ θ t+1 = θ t −α J (θ ) j j ∂θ j t © Eric Xing @ CMU, 2006-2011 13 The Least-Mean-Square (LMS) method z Now we have the following descent rule: n t+1 t v T t j θ j = θ j +α∑(yi − xi θ )xi i=1 z For a single training point, we have: z This is known as the LMS update rule, or the Widrow-Hoff learning rule z This is actually a "stochastic", "coordinate" descent algorithm z This can be used as a on-line algorithm © Eric Xing @ CMU, 2006-2011 14 7 Geometric and Convergence of LMS N=1 N=2 N=3 t+1 t v T t v θ = θ +α(yi − xi θ )xi Claim: when the step size α satisfies certain condition, and when certain other technical conditions are satisfied, LMS will converge to an “optimal region”. © Eric Xing @ CMU, 2006-2011 15 Steepest Descent and LMS z Steepest descent z Note that: T n ⎡ ∂ ∂ ⎤ T ∇θ J = ⎢ J,K, J ⎥ = −∑(yn − xn θ )xn ⎣∂θ1 ∂θk ⎦ i=1 n t+1 t T t θ = θ +α∑(yn − xn θ )xn i=1 z This is as a batch gradient descent algorithm © Eric Xing @ CMU, 2006-2011 16 8 The normal equations z Write the cost function in matrix form: ⎡− − x1 − −⎤ n 1 T 2 ⎢ ⎥ − − x2 − − J(θ ) = ∑(xi θ − yi ) X = ⎢ ⎥ 2 i=1 ⎢ M M M ⎥ ⎢ ⎥ 1 T − − x − − = ()()Xθ − yv Xθ − yv ⎣ n ⎦ 2 ⎡ y1 ⎤ ⎢ ⎥ 1 T T T T T T y = ()θ X Xθ −θ X yv − yv Xθ + yv yv yv = ⎢ 2 ⎥ 2 ⎢ M ⎥ ⎢ ⎥ ⎣yn ⎦ z To minimize J(θ), take derivative and set to zero: 1 T T T T v vT vT v ∇θ J = ∇θ tr()θ X Xθ −θ X y − y Xθ + y y 2 ⇒ X T Xθ = X T yv 1 = ()∇ trθ T X T Xθ −2∇ tryvT Xθ + ∇ tryvT yv 2 θ θ θ The normal equations 1 = ()X T Xθ + X T Xθ −2X T yv 2 ⇓ −1 = X T Xθ − X T yv = 0 θ * = (X T X ) X T yv © Eric Xing @ CMU, 2006-2011 17 Some matrix derivatives m×n z For f : R a R , define: ⎡ ∂ ∂ ⎤ f L f ⎢ ∂A ∂A ⎥ ⎢ 11 1n ⎥ ⎢ ⎥ ∇ A f (A) = M O M ⎢ ∂ ∂ ⎥ ⎢ f L f ⎥ ⎢∂A1m ∂Amn ⎥ ⎣⎢ ⎦⎥ z Trace: n tra = a , trA = ∑ Aii , trABC = trCAB = trBCA i=1 z Some fact of matrix derivatives (without proof) T T T T −1 T ∇ AtrAB = B , ∇ AtrABA C = CAB + C AB , ∇ A A = A(A ) © Eric Xing @ CMU, 2006-2011 18 9 Comments on the normal equation z In most situations of practical interest, the number of data points N is larggyer than the dimensionality k of the inppput space and the matrix X is of full column rank. If this condition holds, then it is easy to verify that XTX is necessarily invertible. z The assumption that XTX is invertible implies that it is positive definite, thus the critical point we have found is a minimum. z What if X has less than full column rank? Æ regularization (later). © Eric Xing @ CMU, 2006-2011 19 Direct and Iterative methods z Direct methods: we can achieve the solution in a single step byyg solving the normal eq uation z Using Gaussian elimination or QR decomposition, we converge in a finite number of steps z It can be infeasible when data are streaming in in real time, or of very large amount z Iterative methods: stochastic or steepest gradient z Converging in a limiting sense z But more attractive in large practical problems z Caution is needed for deciding the learning rate α © Eric Xing @ CMU, 2006-2011 20 10 Convergence rate z Theorem: the steepest descent equation algorithm converggye to the minimum of the cost characterized by normal equation: If z A formal analysis of LMS need more math-mussels; in practice, one can use a small α, or gradually decrease α. © Eric Xing @ CMU, 2006-2011 21 A Summary: z LMS update rule t+1 t T t θ j = θ j +α(yn − xn θ )xn,i z Pros: on-line, low per-step cost, fast convergence and perhaps less prone to local optimum z Cons: convergence to optimum not always guaranteed z Steepest descent n t+1 t T t θ = θ +α∑(yn − xn θ )xn i=1 z Pros: easy to implement, conceptually clean, guaranteed convergence z Cons: batch, often slow converging z Normal equations − θ * = (X T X ) 1 X T yv z Pros: a single-shot algorithm! Easiest to implement.
Load more

Recommended publications
  • One-Network Adversarial Fairness
    One-network Adversarial Fairness Tameem Adel Isabel Valera Zoubin Ghahramani Adrian Weller University of Cambridge, UK MPI-IS, Germany University of Cambridge, UK University of Cambridge, UK [email protected] [email protected] Uber AI Labs, USA The Alan Turing Institute, UK [email protected] [email protected] Abstract is general, in that it may be applied to any differentiable dis- criminative model. We establish a fairness paradigm where There is currently a great expansion of the impact of machine the architecture of a deep discriminative model, optimized learning algorithms on our lives, prompting the need for ob- for accuracy, is modified such that fairness is imposed (the jectives other than pure performance, including fairness. Fair- ness here means that the outcome of an automated decision- same paradigm could be applied to a deep generative model making system should not discriminate between subgroups in future work). Beginning with an ordinary neural network characterized by sensitive attributes such as gender or race. optimized for prediction accuracy of the class labels in a Given any existing differentiable classifier, we make only classification task, we propose an adversarial fairness frame- slight adjustments to the architecture including adding a new work performing a change to the network architecture, lead- hidden layer, in order to enable the concurrent adversarial op- ing to a neural network that is maximally uninformative timization for fairness and accuracy. Our framework provides about the sensitive attributes of the data as well as predic- one way to quantify the tradeoff between fairness and accu- tive of the class labels.
    [Show full text]
  • Remote Sensing
    remote sensing Article A Method for the Estimation of Finely-Grained Temporal Spatial Human Population Density Distributions Based on Cell Phone Call Detail Records Guangyuan Zhang 1, Xiaoping Rui 2,* , Stefan Poslad 1, Xianfeng Song 3 , Yonglei Fan 3 and Bang Wu 1 1 School of Electronic Engineering and Computer Science, Queen Mary University of London, London E1 4NS, UK; [email protected] (G.Z.); [email protected] (S.P.); [email protected] (B.W.) 2 School of Earth Sciences and Engineering, Hohai University, Nanjing 211000, China 3 College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China; [email protected] (X.S.); [email protected] (Y.F.) * Correspondence: [email protected] Received: 10 June 2020; Accepted: 8 August 2020; Published: 10 August 2020 Abstract: Estimating and mapping population distributions dynamically at a city-wide spatial scale, including those covering suburban areas, has profound, practical, applications such as urban and transportation planning, public safety warning, disaster impact assessment and epidemiological modelling, which benefits governments, merchants and citizens. More recently, call detail record (CDR) of mobile phone data has been used to estimate human population distributions. However, there is a key challenge that the accuracy of such a method is difficult to validate because there is no ground truth data for the dynamic population density distribution in time scales such as hourly. In this study, we present a simple and accurate method to generate more finely grained temporal-spatial population density distributions based upon CDR data. We designed an experiment to test our method based upon the use of a deep convolutional generative adversarial network (DCGAN).
    [Show full text]
  • On Construction of Hybrid Logistic Regression-Naıve Bayes Model For
    JMLR: Workshop and Conference Proceedings vol 52, 523-534, 2016 PGM 2016 On Construction of Hybrid Logistic Regression-Na¨ıve Bayes Model for Classification Yi Tan & Prakash P. Shenoy [email protected], [email protected] University of Kansas School of Business 1654 Naismith Drive Lawrence, KS 66045 Moses W. Chan & Paul M. Romberg fMOSES.W.CHAN, [email protected] Advanced Technology Center Lockheed Martin Space Systems Sunnyvale, CA 94089 Abstract In recent years, several authors have described a hybrid discriminative-generative model for clas- sification. In this paper we examine construction of such hybrid models from data where we use logistic regression (LR) as a discriminative component, and na¨ıve Bayes (NB) as a generative com- ponent. First, we estimate a Markov blanket of the class variable to reduce the set of features. Next, we use a heuristic to partition the set of features in the Markov blanket into those that are assigned to the LR part, and those that are assigned to the NB part of the hybrid model. The heuristic is based on reducing the conditional dependence of the features in NB part of the hybrid model given the class variable. We implement our method on 21 different classification datasets, and we compare the prediction accuracy of hybrid models with those of pure LR and pure NB models. Keywords: logistic regression, na¨ıve Bayes, hybrid logistic regression-na¨ıve Bayes 1. Introduction Data classification is a very common task in machine learning and statistics. It is considered an in- stance of supervised learning. For a standard supervised learning problem, we implement a learning algorithm on a set of training examples, which are pairs consisting of an input object, a vector of explanatory variables F1;:::;Fn, called features, and a desired output object, a response variable C, called class variable.
    [Show full text]
  • Data-Driven Topo-Climatic Mapping with Machine Learning Methods
    View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by RERO DOC Digital Library Nat Hazards (2009) 50:497–518 DOI 10.1007/s11069-008-9339-y ORIGINAL PAPER Data-driven topo-climatic mapping with machine learning methods A. Pozdnoukhov Æ L. Foresti Æ M. Kanevski Received: 31 October 2007 / Accepted: 19 December 2008 / Published online: 16 January 2009 Ó Springer Science+Business Media B.V. 2009 Abstract Automatic environmental monitoring networks enforced by wireless commu- nication technologies provide large and ever increasing volumes of data nowadays. The use of this information in natural hazard research is an important issue. Particularly useful for risk assessment and decision making are the spatial maps of hazard-related parameters produced from point observations and available auxiliary information. The purpose of this article is to present and explore the appropriate tools to process large amounts of available data and produce predictions at fine spatial scales. These are the algorithms of machine learning, which are aimed at non-parametric robust modelling of non-linear dependencies from empirical data. The computational efficiency of the data-driven methods allows producing the prediction maps in real time which makes them superior to physical models for the operational use in risk assessment and mitigation. Particularly, this situation encounters in spatial prediction of climatic variables (topo-climatic mapping). In complex topographies of the mountainous regions, the meteorological processes are highly influ- enced by the relief. The article shows how these relations, possibly regionalized and non- linear, can be modelled from data using the information from digital elevation models.
    [Show full text]
  • A Discriminative Model to Predict Comorbidities in ASD
    A Discriminative Model to Predict Comorbidities in ASD The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:38811457 Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA i Contents Contents i List of Figures ii List of Tables iii 1 Introduction 1 1.1 Overview . 2 1.2 Motivation . 2 1.3 Related Work . 3 1.4 Contribution . 4 2 Basic Theory 6 2.1 Discriminative Classifiers . 6 2.2 Topic Modeling . 10 3 Experimental Design and Methods 13 3.1 Data Collection . 13 3.2 Exploratory Analysis . 17 3.3 Experimental Setup . 18 4 Results and Conclusion 23 4.1 Evaluation . 23 4.2 Limitations . 28 4.3 Future Work . 28 4.4 Conclusion . 29 Bibliography 30 ii List of Figures 3.1 User Zardoz's post on AutismWeb in March 2006. 15 3.2 User Zardoz's only other post on AutismWeb, published in February 2006. 15 3.3 Number of CUIs extracted per post. 17 3.4 Distribution of associated ages extracted from posts. 17 3.5 Number of posts written per author . 17 3.6 Experimental Setup . 22 4.1 Distribution of AUCs for single-concept word predictions . 25 4.2 Distribution of AUCs for single-concept word persistence . 25 4.3 AUC of single-concept word vs.
    [Show full text]
  • Performance Comparison of Different Pre-Trained Deep Learning Models in Classifying Brain MRI Images
    ACTA INFOLOGICA dergipark.org.tr/acin DOI: 10.26650/acin.880918 RESEARCH ARTICLE Performance Comparison of Different Pre-Trained Deep Learning Models in Classifying Brain MRI Images Beyin MR Görüntülerini Sınıflandırmada Farklı Önceden Eğitilmiş Derin Öğrenme Modellerinin Performans Karşılaştırması Onur Sevli1 ABSTRACT A brain tumor is a collection of abnormal cells formed as a result of uncontrolled cell division. If tumors are not diagnosed in a timely and accurate manner, they can cause fatal consequences. One of the commonly used techniques to detect brain tumors is magnetic resonance imaging (MRI). MRI provides easy detection of abnormalities in the brain with its high resolution. MR images have traditionally been studied and interpreted by radiologists. However, with the development of technology, it becomes more difficult to interpret large amounts of data produced in reasonable periods. Therefore, the development of computerized semi-automatic or automatic methods has become an important research topic. Machine learning methods that can predict by learning from data are widely used in this field. However, the extraction of image features requires special engineering in the machine learning process. Deep learning, a sub-branch of machine learning, allows us to automatically discover the complex hierarchy in the data and eliminates the limitations of machine learning. Transfer learning is to transfer the knowledge of a pre-trained neural network to a similar model in case of limited training data or the goal of reducing the workload. In this study, the performance of the pre-trained Vgg-16, ResNet50, Inception v3 models in classifying 253 brain MR images were evaluated. The Vgg-16 model showed the highest success with 94.42% accuracy, 83.86% recall, 100% precision and 91.22% F1 score.
    [Show full text]
  • A Latent Model of Discriminative Aspect
    A Latent Model of Discriminative Aspect Ali Farhadi 1, Mostafa Kamali Tabrizi 2, Ian Endres 1, David Forsyth 1 1 Computer Science Department University of Illinois at Urbana-Champaign {afarhad2,iendres2,daf}@uiuc.edu 2 Institute for Research in Fundamental Sciences [email protected] Abstract example, the camera location in the room might be a rough proxy for aspect. However, we believe this notion of aspect Recognition using appearance features is confounded by is not well suited for recognition tasks. First, it imposes phenomena that cause images of the same object to look dif- the unnatural constraint that objects in different classes be ferent, or images of different objects to look the same. This aligned into corresponding pairs of views. This is unnatural may occur because the same object looks different from dif- because the front of a car looks different to the front of a ferent viewing directions, or because two generally differ- bicycle. Worse, the appearance change from the side to the ent objects have views from which they look similar. In this front of a car is not comparable with that from side to front paper, we introduce the idea of discriminative aspect, a set of a bicycle. Second, the geometrical notion of aspect does of latent variables that encode these phenomena. Changes not identify changes that affect discrimination. We prefer a in view direction are one cause of changes in discrimina- notion of aspect that is explicitly discriminative, even at the tive aspect, but others include changes in texture or light- cost of geometric precision. ing.
    [Show full text]
  • On Surrogate Supervision Multi-View Learning
    AN ABSTRACT OF THE THESIS OF Gaole Jin for the degree of Master of Science in Electrical and Computer Engineering presented on December 3, 2012. Title: On Surrogate Supervision Multi-view Learning Abstract approved: Raviv Raich Data can be represented in multiple views. Traditional multi-view learning meth- ods (i.e., co-training, multi-task learning) focus on improving learning performance using information from the auxiliary view, although information from the target view is sufficient for learning task. However, this work addresses a semi-supervised case of multi-view learning, the surrogate supervision multi-view learning, where labels are available on limited views and a classifier is obtained on the target view where labels are missing. In surrogate multi-view learning, one cannot obtain a classifier without information from the auxiliary view. To solve this challenging problem, we propose discriminative and generative approaches. c Copyright by Gaole Jin December 3, 2012 All Rights Reserved On Surrogate Supervision Multi-view Learning by Gaole Jin A THESIS submitted to Oregon State University in partial fulfillment of the requirements for the degree of Master of Science Presented December 3, 2012 Commencement June 2013 Master of Science thesis of Gaole Jin presented on December 3, 2012. APPROVED: Major Professor, representing Electrical and Computer Engineering Director of the School of Electrical Engineering and Computer Science Dean of the Graduate School I understand that my thesis will become part of the permanent collection of Oregon State University libraries. My signature below authorizes release of my thesis to any reader upon request. Gaole Jin, Author ACKNOWLEDGEMENTS My most sincere thanks go to my advisor, Dr.
    [Show full text]
  • Generative and Discriminative Models
    30 Jonathan Richard Shewchuk 6 Decision Theory; Generative and Discriminative Models DECISION THEORY aka Risk Minimization [Today I’m going to talk about a style of classifier very di↵erent from SVMs. The classifiers we’ll cover in the next few weeks are based on probability.] [One aspect of probabilistic data is that sometimes a point in feature space doesn’t have just one class. Suppose one borrower with income $30,000 and debt $15,000 defaults. another ” ” ” ” ” ” ” doesn’t default. So in your feature space, you have two feature vectors at the same point with di↵erent classes. Obviously, in that case, you can’t draw a decision boundary that classifies all points with 100% accuracy.] Multiple sample points with di↵erent classes could lie at same point: we want a probabilistic classifier. Suppose 10% of population has cancer, 90% doesn’t. Probability distributions for calorie intake, P(X Y): [caps here mean random variables, not matrices.] | calories (X) < 1,200 1,200–1,600 > 1,600 cancer (Y = 1) 20% 50% 30% no cancer (Y = 1) 1% 10% 89% − [I made these numbers up. Please don’t take them as medical advice.] Recall: P(X) = P(X Y = 1) P(Y = 1) + P(X Y = 1) P(Y = 1) | | − − P(1,200 X 1,600) = 0.5 0.1 + 0.1 0.9 = 0.14 ⇥ ⇥ You meet guy eating x = 1,400 calories/day. Guess whether he has cancer? [If you’re in a hurry, you might see that 50% of people with cancer eat 1,400 calories, but only 10% of people with no cancer do, and conclude that someone who eats 1,400 calories probably has cancer.
    [Show full text]
  • Adaptive Discriminative Generative Model and Its Applications
    Adaptive Discriminative Generative Model and Its Applications Ruei-Sung Lin† David Ross‡ Jongwoo Lim† Ming-Hsuan Yang∗ †University of Illinois ‡University of Toronto ∗Honda Research Institute [email protected] [email protected] [email protected] [email protected] Abstract This paper presents an adaptive discriminative generative model that gen- eralizes the conventional Fisher Linear Discriminant algorithm and ren- ders a proper probabilistic interpretation. Within the context of object tracking, we aim to find a discriminative generative model that best sep- arates the target from the background. We present a computationally efficient algorithm to constantly update this discriminative model as time progresses. While most tracking algorithms operate on the premise that the object appearance or ambient lighting condition does not significantly change as time progresses, our method adapts a discriminative genera- tive model to reflect appearance variation of the target and background, thereby facilitating the tracking task in ever-changing environments. Nu- merous experiments show that our method is able to learn a discrimina- tive generative model for tracking target objects undergoing large pose and lighting changes. 1 Introduction Tracking moving objects is an important and essential component of visual perception, and has been an active research topic in computer vision community for decades. Object tracking can be formulated as a continuous state estimation problem where the unobserv- able states encode the locations or motion parameters of the target objects, and the task is to infer the unobservable states from the observed images over time. At each time step, a tracker first predicts a few possible locations (i.e., hypotheses) of the target in the next frame based on its prior and current knowledge.
    [Show full text]
  • Logistic Regression, Generative and Discriminative Classifiers
    Logistic Regression, Generative and Discriminative Classifiers Recommended reading: • Ng and Jordan paper “On Discriminative vs. Generative classifiers: A comparison of logistic regression and naïve Bayes,” A. Ng and M. Jordan, NIPS 2002. Machine Learning 10-701 Tom M. Mitchell Carnegie Mellon University Thanks to Ziv Bar-Joseph, Andrew Moore for some slides Overview Last lecture: • Naïve Bayes classifier • Number of parameters to estimate • Conditional independence This lecture: • Logistic regression • Generative and discriminative classifiers • (if time) Bias and variance in learning Generative vs. Discriminative Classifiers Training classifiers involves estimating f: X Æ Y, or P(Y|X) Generative classifiers: • Assume some functional form for P(X|Y), P(X) • Estimate parameters of P(X|Y), P(X) directly from training data • Use Bayes rule to calculate P(Y|X= xi) Discriminative classifiers: 1. Assume some functional form for P(Y|X) 2. Estimate parameters of P(Y|X) directly from training data • Consider learning f: X Æ Y, where • X is a vector of real-valued features, < X1 …Xn > • Y is boolean • So we use a Gaussian Naïve Bayes classifier • assume all Xi are conditionally independent given Y • model P(Xi | Y = yk) as Gaussian N(µik,σ) • model P(Y) as binomial (p) • What does that imply about the form of P(Y|X)? • Consider learning f: X Æ Y, where • X is a vector of real-valued features, < X1 …Xn > • Y is boolean • assume all Xi are conditionally independent given Y • model P(Xi | Y = yk) as Gaussian N(µik,σ) • model P(Y) as binomial (p)
    [Show full text]
  • Discriminative and Generative Models for Clinical Risk Estimation: an Empirical Comparison
    Discriminative and Generative Models for Clinical Risk Estimation: an Empirical Comparison John Stamford1 and Chandra Kambhampati2 1 Department of Computer Science, The University of Hull, Kingston upon Hull, HU6 7RX, United Kingdom [email protected] 2 Department of Computer Science, The University of Hull, Kingston upon Hull, HU6 7RX, United Kingdom [email protected] Abstract Linear discriminative models, in the form of Logistic Regression, are a popular choice within the clinical domain in the development of risk models. Logistic regression is commonly used as it offers explanatory information in addition to its predictive capabilities. In some examples the coefficients from these models have been used to determine overly simplified clinical risk scores. Such models are constrained to modeling linear relationships between the variables and the class despite it known that this relationship is not always linear. This paper compares the conditions under which linear discriminative and linear generative models perform best. This is done through comparing logistic regression and naïve Bayes on real clinical data. The work shows that generative models perform best when the internal representation of the data is closer to the true distribution of the data and when there is a very small difference between the means of the classes. When looking at variables such as sodium it is shown that logistic regression can not model the observed risk as it is non-linear in its nature, whereas naïve Bayes gives a better estimation of risk. The work concludes that the risk estimations derived from discriminative models such as logistic regression need to be considered in the wider context of the true risk observed within the dataset.
    [Show full text]