Naval forces frequently operate in unpredictable environments. Currently, typical fielded systems are rule-based and operate well in the specific, controlled environments for which they are designed. These systems however fail when encountering variations in the scenarios that fall outside their model of the world. It is, therefore, imperative that autonomous systems have the capability to adapt to the operating scenarios as they learn from experience and errors. Although significant progress has been been made in component technologies (e.g., signal processing and knowledge management), what is lacking are systems that can successfully act in open world with Uncertain, Incomplete, Imprecise, and Contradictory (UIIC) data. To address the challenge, we propose to develop a computational framework for learning and inference with UIIC data based on the theory of matrix completion and random matrix theory. We will focus on three key research questions: kernel learning that aims to represent UIIC data by multiple partial kernel matrices and develop theories and computational algorithms for learning from the partial kernel matrices, (ii) transfer learning that aims to improve the learning of UIIC data by effectively exploiting the prediction models learned in the related tasks, and (iii) online learning that aims to iteratively refine and improve the learned statistical model for UIIC data based on user feedbacks.

Students

1. Mehrdad Mahdavi
2. Lijun Zhang

Publications

1. M. Xu, R. Jin, and Z.-H. Zhou, Speedup Matrix Completion with Side Information: Application to Multi-Label Learning, Advance in Neural Information Processing Systems (NIPS) 2013
2. J. Yi, L. Zhang, R. Jin, and A. Jain, Semi-supervised Clustering by Input Pattern Assisted Pairwise Similarity Matrix Completion, International Conference on Machine Learning (ICML 2013), 2013
3. L. Zhang, J. Yi, and R. Jin, Online Kernel Learning with a Near Optimal Sparsity Bound, International Conference on Machine Learning (ICML), 2013
4. L. Zhang, M. Mahdavi, and R. Jin, Recovering the Optimal Solution by Dual Random Projection, Conference one Learning Theory (COLT), 2013
5. L. Zhang, M. Mahdavi, and R. Jin, Linear Convergence with Condition Number Independent Access of Full Gradients, Advance in Neural Information Processing Systems(NIPS), 2013
6. M. Mahdavi, L. Zhang, and R. Jin, Mixed Optimization for Smooth Functions, Advance in Neural Information Processing Systems (NIPS), 2013
7. M. Mahdavi, T. Yang, and R. Jin, Stochastic Convex Optimization with Multiple Objectives, Advance in Neural Information Processing Systems(NIPS), 2013