【李宏毅机器学习课件】23.结构化预测：结构化SVM.pdf

Structured Support Vector Machine Hung-yi Lee Structured Learning • We need a more powerful function f • Input and output are both objects with structures • Object: sequence, list, tree, bounding box … f : X  Y X is the space of Y is the space of one kind of object another kind of object Unified Framework Step 1: Training • Find a function F F : X Y R • F(x,y): evaluate how compatible the objects x and y is Step 2: Inference (Testing) • Given an object x ~   y arg max F x , y y Y Three Problems Problem 1: Evaluation • What does F(x,y) look like? Problem 2: Inference • How to solve the “arg max” problem   y arg max F x , y y Y Problem 3: Training • Given training data, how to find F(x,y) Example Task: Object Detection Example Task Keep in mind that what you will learn today can be applied to other tasks. Source of image: /viewdoc/download?doi=95.6007rep=rep1type=pdf http://www.vision.ee.ethz.ch/~hpedemo/gallery.php Problem 1: Evaluation • F(x,y) is linear → → = ∙ Open question: What if F(x,y) is not linear? Problem 2: Inference ෤ = argmax∙(,) ∈