Learning Data Representations with Convex Optimization
Event details
| Date | 06.04.2009 |
| Hour | 16:15 |
| Speaker | Dr. Kilian Weinberger, Yahoo, Mountain View, USA |
| Location |
INM202
|
| Category | Conferences - Seminars |
One of the fundamental challenges of machine learning and artificial intelligence is the learning of suitable representations of data. Many learning algorithms assume that input data is presented in low-dimensional vectorial form, where Euclidean distances reflect dissimilarities. However, different tasks can have very different definitions of similarity. Ideally, one should use a "hand-tailored" representation of each particular data set for any given task.
In this talk, I will give an overview of three algorithms for learning compact representations that give rise to semantically meaningful similarity metrics. Each of the algorithms involves, at its core, a convex optimization problem that learns the new representation under meaningful constraints. This framework provides perfect reproducibility and theoretical guarantees.
The three methods are suited to different data settings: Taxonomy Embedding learns low-dimensional representations of text documents in a Euclidean space such that distances reflect dissimilarities according to a given topic hierarchy. Maximum Variance Unfolding reduces the dimensionality of data sets with underlying manifold structure and can be viewed as a non-linear extension of Principal Component Analysis (PCA). Large Margin Nearest Neighbor learns a relaxation of the Euclidean metric tailored specifically for k-nearest neighbor classification.
I present state-of-the-art classification results on several real world applications, including face recognition and document categorization on the OHSUMED medical journal data base.
Bio:
Kilian Weinberger is a Research Scientist at Yahoo Research in Santa Clara, California. He works on next-generation spam filtering algorithms, multimedia search, high-dimensional data analysis, and machine learning with convex optimization. He received his Ph.D. in Computer Science at the University of Pennsylvania in 2007 under the supervision of Professor Lawrence Saul. His work on supervised and unsupervised metric learning has won several outstanding paper awards at CVPR, AISTATS and ICML. Prior to his doctoral studies he earned a first class honor BA in Mathematics and Computer Science from the University of Oxford. Kilian is originally from Germany.
K. Weinberger's homepage
Practical information
- General public
- Free