Georgios Giannakis
University of Minnesota, MN·Engineering
Modern networks, from short-range wireless links to cellular systems and connected sensors, generate data that can be noisy, incomplete, distorted or even deliberately corrupted. This work combines signal processing, wireless communications and statistical machine learning to build models and algorithms that stay reliable under these conditions.
- Fields
Engineering
Computer Science
Mathematics & Statistics
- Topics
wireless communications
optical wireless communication
resonant beam
- Grants
- 3 known grants, through May 2027
What they work on
- Model the resonant-beam optical wireless channel, an infrared beam formed in an open-air optical cavity, using nonlinear time-varying Volterra series and kernel-based learning for both downlink and uplink.
- Design transceivers for resonant-beam links that tolerate the channel's nonlinearity and large delay spread, starting from on-off keying and orthogonal frequency-division multiplexing and extending to AI-designed end-to-end signals.
- Develop optical designs with a spatial light modulator, a device that reshapes light, to serve multiple users and compensate for disturbances. Validate the algorithms on an experimental testbed.
- Identify anomalous nodes in processes over network graphs by combining random sampling and consensus with graph signal processing tools, so outlying or corrupted data can be pinpointed.
- Make graph convolutional networks robust to errors in graph links using dithering-inspired modules, and design graph-aware ensemble and crowdsourcing algorithms that rate the reliability of information sources.
- Model 5G and next-generation cellular network measurements with ensembles of Gaussian processes that capture cross-layer features and changing environment, user and application conditions.
- Develop Bayesian active sampling to guide which network measurements to take, plus tools to de-bias, de-noise and merge incomplete, conflicting or adversarial crowd-sourced measurements.
More about the research
Three projects share this thread: a new kind of optical wireless link, robust machine learning on network-structured data, and learning-driven measurement of next-generation cellular networks.
Why it matters
High-speed, low-latency wireless links that do not interfere with radio devices could support augmented and virtual reality, industrial Internet of Things, and smart homes, hospitals and factories. Learning methods that withstand noisy data, faulty network structure and malicious sources could make inferences more trustworthy in social analytics, health informatics and Internet-of-Things security. Better measurement models could also help operators monitor, manage and deploy next-generation cellular networks.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Shuai Zhang
New Jersey Institute of Technology, NJ·Engineering
Develop topology sampling and graph neural network pruning methods for single-layer graph models, trimming network structure and model size so spatial-temporal AIoT data can be analyzed more efficiently.
Feng Chen
University of Texas at Dallas, TX·Computer Science
Develop online fairness-aware deep learning models for changing data streams, using causal representation learning, differentiable architecture search and online meta-learning, and release open software.
Robert Heath
University of California-San Diego, CA·Engineering
Build MIMO input-output models of dynamic metasurface antennas (low-power panels of tunable elements) that capture waveguide loss, coupling between elements and reconfiguration effects, using existing circuit and antenna models.
The school
University of Minnesota: 88 labs on PI Pages, $1.4B a year on research, 722 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.