Milenkovic: Franklin W. Woeltge Professor of Electrical and Computer Engineering

Professorship: Franklin W. Woeltge Professor of Electrical and Computer Engineering

The Franklin W. Woeltge Professorships were established in the Department of Electrical and Computer Engineering thanks to a $4 million gift from Franklin Woeltge (BSEE ’26), who died in 1998 at the age of 95. Born in St. Louis in 1903, Mr. Woeltge lived there his entire life except for his years at the University of Illinois, where he participated in the Electrical Engineering Society and American Institute of Electrical Engineering.

As an engineer, Mr. Woeltge worked in the Avionics and Space Division at Emerson Electric Co., retiring in 1963. He accumulated much of his wealth by investing in the stock market—a favorite passion of his, along with reading and ballroom dancing. Mr. Woeltge had no surviving relatives, so he made his bequest to the Department of Electrical and Computer Engineering because, he once said, that is where he spent some of the best days of his life.

 

Faculty: Olgica Milenkovic

Prof. Milenkovic's research focuses on several areas including DNA storage, bioinformatics, compressive sensing, low rank matrix completion, community detection, hypergraph clustering, and ordinal data processing. 

Prof. Milenkovic was among the first researchers to get involved in the emerging field of DNA-based storage system design. Her research group developed the first prototype of a random access and rewritable DNA storage system, and they implemented their method by storing a number of Wikipedia pages in DNA. Furthermore, her group was the first to demonstrate through practical implementation that portable storage is possible, with readouts performed on highly noisy nanopore sequencing platforms such as Oxford Nanopore, and they received a patent for this work. The area of DNA-based storage brings with itself unique data addressing and error-control coding challenges that were previously not considered. In particular, Olgica proposed coding schemes from DNA profiles, DNA storage systems with nanopore readout systems, and so called codes in the Damerau distance that can correct block deletions/insertions and adjacent transpositions. In addition, she introduced the concept of weakly mutually uncorrelated addresses that capture important properties need for selective strand addressing in the proposed storage mechanism. This line of work was featured in the New York Times, and a special program was created by an industry Consortium and IARPA to build a scalable and cost efficient storage system by the year 2024. Olgica was the invited keynote speaker at the symposium organized by the Consortium in April 2016, and has given several other keynote and plenary lectures on the topic since that time.

Prof. Milenkovic's work in the area of compressive sensing established interesting new connections between the fields of high-dimensional statistics, dimensionality reduction algorithms and information theory. In particular, she developed one of the state-of-the-art methods for greedy-like compressive sensing reconstruction, termed Subspace Pursuit, and proceeded to apply it in the area of gene regulatory network analysis for the purpose of discovering causal transcriptional interactions. She also analyzed the performance of quantized compressive sensing techniques, establishing what the best performing quantization scheme is for any given sensing matrix. She also connected error-tolerant, non-negative compressive sensing with a new form of Euclidean superimposed coding, and derived fundamental results on the smallest number of measurements needed for lossless recovery. She consequently used these results for the practical implementation and analysis of the first compressive sensing DNA microarray design, a sensing modality that exploits genomic sequence cross-hybridization and therefore significantly reduces the cost of DNA microarray production. In the same topic area, she proposed a new Boolean compressive sensing paradigm, termed semiquantitative group testing. This sensing modality was designed with the purpose of addressing the high cost of sample preparation in certain large-scale DNA sequencing experiments and quantization error inherent in the readout process. The goal of the approach is to replace linear projections with highly non-linear operations that better capture the real biological sensing process. In addition to DNA sequencing, semiquantitative group testing has applications in classical areas of multiple access channel sequence design. 

In the area of low-rank matrix completion,  Prof. Milenkovic and her collaborators proposed a new algorithmic approach termed Subspace Evolution and Transfer (SET) which formulates the completion problem as a non-convex objective optimization problems on the Grassmanian and solves it by a specialized gradient descent method that involves a transfer step that allows for avoiding local optima introduced by search "barriers".

More recently, Prof. Milenkovic has been  working on new higher-order clustering algorithms, also known as motif or hypergraph clustering techniques. Here, the goal is to measure the quality of clustering not by how many edges lie across clusters  but how many motif subgraphs lie across clusters. She and her research group developed new hypergraph projection methods, hypergraph Laplacians and local spectral hypergraph clustering techniques. On the theoretical side, her group was among the first to propose several new paradigms in correlation clustering regarding minimax and constrained correlation clustering. They used these methods to develop state-of-the art driver gene identification tools that rely on mutual exclusivity and coverage constraints. These results have been presented at top machine learning conferences, such as Neural Information Processing Systems (NeurIPS) and International Conference on Machine Learning (ICML), including a "Spotlight" presentation at the NeurIPS 2017 conference.