1 / 1

Max-Margin Minimum Entropy Models

h. x. y. h. x. Max-Margin Minimum Entropy Models. Kevin Miller, M. Pawan Kumar, Ben Packer, Danny Goodman, and Daphne Koller . Generalized Distribution. Experiments. Aim: To learn an accurate set of parameters for latent variable models.

rea
Download Presentation

Max-Margin Minimum Entropy Models

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. h x y h x Max-Margin Minimum Entropy Models Kevin Miller, M. Pawan Kumar, Ben Packer, Danny Goodman, and Daphne Koller Generalized Distribution Experiments Aim: To learn an accurate set of parameters for latent variable models Object Classification – Mammals Dataset Log-linear model over features: Image label y is object class only, h is bounding box is HOG features in bounding box (offset by class) Conditional distribution: Overview Generalized distribution: Latent variable models predict output from input They typically (i) compute the most likely latent variable value or (ii) marginalize over latent variable values Ignore an important factor: How certain is the model about the latent variable value? Our approach: Learn a model to (a) maximize probability of data (b) minimize latent variable uncertainty This yields a novel family of discriminative LVMs Latent SVM is a special case of the M3E family Rényi Entropy Family of measures for uncertainty over generalized distribution sum of negative log likelihood and uncertainty over normalized conditional probability of latent variables • Shannon Entropy: Special Cases • Minimum Entropy: Overall test error: LSVM (5.7%), alpha=Inf (5.4%), best M3E (4.2%) Motif Finding – UniProbe Dataset Max-Margin Minimum Entropy Models (M3E) x is DNA sequence, h is motif position, y is binding affinity Predict output by minimizing Rényi Entropy: Explains data well while minimizing uncertainty over latent variables Learn an M3E by maximizing margin: Latent Variable Models Difference of Convex (DOCP) x : input or observed variables y : output or observed variables h : hidden/latent variables Learning M3E Models Special Case: Minimum entropy Iterate: y = “Deer” Objective equivalent to Latent SVM Procedure nearly equivalent h = bounding box : Loss function (0-1) : feature function (HOG) Standard Learning Algorithms General Case Iterate: Goal: Given , learn parameters Linearize problem (Taylor approximation) Expectation-Maximization for Maximum Likelihood Maximize log likelihood (marginalize hidden variables): Latent Struct SVM Solve structural SVM problem Minimize upper bound on risk (maximize hidden variables): Overall improvement of 2% on both train and test error Code available for download at: http://ai.stanford.edu/~pawan/publications/MKPGK-AISTATS2012.html

More Related