Theory and Methods for Accurate and Scalable Learning Machines
Swiss National Science Foundation (407540-167319/1)
Dates: 2017-2021 [PhD student: Teresa Yeo; Postdoc: Kamalaruban Parameswaran]
Large-Scale Sparse Bayesian Modeling, Inference, and Design
Team: Volkan Cevher [PI], Andreas Krause @ ETHZ [co-PI], Jarvis Haupt @ UMN [co-PI]
Scalable and Accurate Quantum Tomography
Swiss National Science Foundation (200021-146750)
Dates: 2013-2016 [PhD student: Yen-Huan Li]
Theory and methods for compressive sensing with graphical models [GM-CS]
Swiss National Science Foundation (200021-132620)
Dates: 2011-2013 [PhD student: Anastasios Kyrillidis]
Compressive sensing (CS) is an emerging alternative to Shannon/Nyquist sampling paradigm for simultaneous sensing and compression of signals having a sparse representation. By sparse representation, we mean the signal can be well approximated by a few nonzero coefficients in some basis. According to CS, the number of compressive measurements for stable recovery is proportional to the signal sparsity, rather than to its Fourier bandwidth. For this reason, CS permits revolutionary new sensing designs and algorithms for dimensionality reduction.
This project exploits probabilistic graphical models in CS as realistic signal models that can capture underlying, application dependent order of sparse coefficients, and as sampling matrices with information preserving properties that can be implemented in practical systems.