Robert Nowak ECE Dept UWMadison nowakengr wisc edu

Robert Nowak ECE Dept. , UW-Madison nowak@engr. wisc. edu www. ece. wisc. edu/~nowak Research Interests: statistical signal processing, machine learning, imaging and network science, and applications in communications, bio/medical imaging, and in silico genomics. Network Science, National Academies Press, 2006 The study of complex networked systems. Key Challenges : “Characterization of the dynamics and information flow in networked systems, modeling, analysis, and acquisition of experimental data for extremely large networks. ” My take: In many large-scale problems we have limited prior knowledge, but a wealth of data. How much can we learn from data? Adaptivity to unknown system behavior is key.

Challenge 1: Inferring Networks from Experimental Data Network Tomography: Infer network behavior and structure from indirect and incomplete data Challenges: • ill-posed problem • errors and noise • calibration Internet routing behavior/structure MAP Kinase Regulation Network

Challenge 2: Detecting Weak Non-Local Signals Network Detection: Xi = data at each node Test: H 0 : Xi ~ N(0, 1) for all i vs. H 1 : Xi ~ N(m, 1), m > 0, at handful of nodes Challenge: • m > 0 may be so small, that individual testing at each node is unreliable (e. g. , biohazard or Internet virus detection) • plug-in schemes (e. g. , the GLRT) are suboptimal in high dimensional settings • Data fusion (aggregation) can enhance detection capabilities, but typically requires strong prior knowledge Detection must be adaptive to unknown network behavior and/or structure