UMass Machine Learning and Friends Lunch | Main / Learning Optimal Projections For Manifold Alignment

In this talk we introduce a novel approach to manifold alignment. Our approach differs from the state of the art approaches in that it automatically results in a mapping function that projects the data instances (from two different spaces) to a lower dimensional space simultaneously preserving the neighborhood relationship inside of each of the two sets and minimizing the difference between corresponding instances. We describe and evaluate our approach both theoretically and experimentally, providing results showing useful knowledge transfer from one domain to another. Real-world applications include automatic machine translation, cross-lingual information retrieval, topic discovery, representation and control transfer in Markov decision processes, bioinformatics and image comparison.