Shape Matching and Object Recognition using Low Distortion 形状匹配和使用低畸变目标识别.ppt

What is task, shape modeling List limitations, then do approximations Mention sampling here Eliminate bottom Move figure to left, list pros with cons Lower case, say based on pah Font? Font, mention f_data possibilities Take out equations, just say theta is word counts, and we regularize Include acronyms Say hyperprior doesn’t help Bring earlier, mention PCA, say shrinkage, likelihood perform very poorly Bring up red, black first, label them on image, then show rest Change order, write not all parameters deserve equal sharing Popup for how to estimate so many parameters Popupbigger, more prominent, change order of animation Popupbigger, more prominent, change order of animation Bring up red, black first, label them on image, then show rest Mention shrinkage, maxlike Fix widths Convex Point Estimation using Undirected Bayesian Transfer Hierarchies Gal Elidan Ben Packer Geremy Heitz Daphne Koller Stanford AI Lab Motivation Problem: With few instances, learned models aren’t robust MEAN Principal Components Task: Shape modeling std +1 std -1 std +1 std -1 MEAN Transfer Learning MEAN Principal Components Shape is stabilized, but doesn’t look like an elephant Can we use rhinos to help elephants? std +1 std -1 std +1 std -1 MEAN Hierarchical Bayes qroot qElephant qRhino P(qElephant|qroot) P(qRhino|qroot) P(DataElephant|qElephant) P(DataRhino|qRhino) P(qroot) Goals Transfer between related classes Range of settings, tasks Probabilistic motivation Multilevel, complex hierarchies Simple, efficient computation Automatically learn what to transfer Hierarchical Bayes qroot qElephant qRhino Compute full posterior P(￡|D) P(￡c|￡root) must be conjugate to P(D|￡c) Problem: Often can’t perform full Bayesian computations Approx.: Point estimation qroot qElephant qRhino Empirical Bayes Point estimation Other approximations: Posterior as normal, sampling, etc. Best parameters are good enough; don’t need full distribution More Issues: Multiple Levels Conjuga