3M Non-Tenured Faculty Award

I am honored to have received 3M’s Non-Tenured Faculty Award, which recognizes outstanding junior faculty nominated by 3M researchers on the basis of their demonstrated record of research, experience, and academic leadership. I look forward to working with 3M Machine Learning researchers to advance data science research!

Publications Update

We have had many exciting publications in the last several months.

My student Dejiao Zhang and I worked with Mario Figueiredo and two other Michigan students on applying OWL regularization in deep networks. The intuition is that since OWL can tie correlated regressors, it should be able to do the same in deep nets that experience a high degree of co-adaptation (and correlation) of nodes in the network. Dejiao presented our paper Learning to Share: Simultaneous Parameter Tying and Sparsification for Deep Learning at ICLR last month and we will present Simultaneous Sparsity and Parameter Tying for Deep Learning using Ordered Weighted L1 Regularization at SSP next month.

With my colleague Johanna Mathieu and her student Greg Ledva, we published a paper in Transactions on Power Systems studying Real-Time Energy Disaggregation of a Distribution Feeder’s Demand Using Online Learning. The work leverages recent results in dynamic online learning where classes of dynamical models are used to apply online learning to the time-varying signal setting. This work can leverage existing sensing structure to improve prediction of distributed energy resources, demand-responsive electric loads and residential solar generation. We also have a book chapter in Energy Markets and Responsive Grids that was written also with my student Zhe Du.

Greg Ongie, David Hong, Dejiao Zhang, and I have been working on adaptive sampling for subspace estimation. If one has a matrix in memory that is large and difficult to access, but you want to compute a low-rank approximation of that matrix, one way is to sketch it by reading only parts of the matrix and computing an approximation. Our paper Enhanced Online Subspace Estimation Via Adaptive Sensing describes an adaptive sampling scheme to do exactly that, and using that scheme along with the GROUSE subspace estimation algorithm, we gave global convergence guarantees to the true underlying low-rank matrix. We will also present Online Estimation of Coherent Subspaces with Adaptive Sampling at SSP next month, which constrains the adaptive samples to be entry-wise and sees similar improvements.

Rounding it out, Zhe Du will be presenting our work with Necmiye Ozay on A Robust Algorithm for Online Switched System Identification at the SYS ID conference in July, and Bob Malinas and David Hong will present our work with Jeff Fessler on Learning Dictionary-Based Unions of Subspaces for Image Denoising at EUSIPCO in September. This spring Amanda Bower presented our work with Lalit Jain on The Landscape of Nonconvex Quadratic Feasibility, studying the minimizers for a non-convex formulation of the preference learning problem; and next week Naveen Murthy presents our work with Greg Ongie and Jeff Fessler on Memory-efficient Splitting Algorithms for Large-Scale Sparsity Regularized Optimization at the CT Meeting. Last fall Greg Ongie, Saket Dewangan, Jeff Fessler and I had a paper Online Dynamic MRI Reconstruction via Robust Subspace Tracking at GlobalSIP, pursuing the interesting idea of online subspace tracking for time-varying signals.

So many exciting research directions that we will continue to pursue!

Monotonic Matrix Completion

Ravi Ganti and Rebecca Willett and I had a paper in NIPS 2015 called “Matrix Completion under Monotonic Single Index Models.” We studied a matrix completion problem where a low-rank matrix is observed through a monotonic function applied to each entry. We developed a calibrated loss function that allowed a neat implementation and analysis. Now the code is available for public usage at this bitbucket link.

Variety Matrix Completion code

The code for our matrix completion algorithm from the ICML paper “Algebraic Variety Models for High-rank Matrix Completion” can be found here in Greg Ongie’s github repository. Using the algebraic variety as a low-dimensional model for data, Greg’s algorithm is a kernel method for doing matrix completion in the space associated with a polynomial kernel. It allows us to do matrix completion even when the matrix does not have low linear rank — but instead when it has low dimension in the form of a nonlinear variety. Start with the README for example scripts.

Congratulations John!

Congratulations to Dr. John Lipor for successfully defending his PhD thesis in September! The title of his work is “Sensing Structured Signals with Active and Ensemble Methods.” In January he will start as Assistant Professor in the Portland State University ECE Department.

Congratulations David!

David Hong was awarded the Richard and Eleanor Towner Prize for Outstanding PhD Research at the Michigan Engineering Graduate Symposium. This is a prize awarded annually across the entire college of engineering to PhD students within about a year of graduation, and the criteria for selection are creativity, innovation, impact on society, and achievement. Congratulations David!

Data-Driven Discovery of Models

Jason Corso and I have been awarded a DARPA D3M grant. Our project is called SPIDER: Subspace Primitives that are Interpretable and DivERse. We will be contributing machine learning software primitives for a system that helps domain experts perform a wide variety of automated data analysis on their datasets. The project has 24 teams and is already off to a great start — We look forward to final system developed by this team over the next few years!

Distance-Penalized Active Learning Using Quantile Search

Active sampling — where one chooses what samples to collect based on data collected thus far — is an important approach for spatial environmental sampling, where resources are drastically limited when compared to the extent of the signals of interest. However, most active learning literature studies the case where each sample has equal cost. In spatial sampling, the sample cost is often proportional to distance between samples. John Lipor and I collaborated with our colleagues in the department of Civil and Environmental Engineering and the department of Natural Resources to develop active sampling techniques for lake sampling.

The code is available here. You can also find a video about the project here.

Lipor, J., B. P. Wong, D. Scavia, B. Kerkez, and L. Balzano. 2017. “Distance-Penalized Active Learning Using Quantile Search.” IEEE Transactions on Signal Processing 65 (20): 5453–65. https://doi.org/10.1109/TSP.2017.2731323.

Lipor, J., L. Balzano, B. Kerkez, and D. Scavia. 2015. “Quantile Search: A Distance-Penalized Active Learning Algorithm for Spatial Sampling.” In 2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton), 1241–48. https://doi.org/10.1109/ALLERTON.2015.7447150.

ICML Acceptances, SIAM OPT success

Congratulations to postdoc Greg Ongie, whose excellent work on Variety Models for Matrix Completion has been accepted to ICML. We’re very excited about the potential applications and open problems that we posed in this work. Congratulations also to John Lipor, whose work on Active Subspace Clustering has also been accepted to ICML — I’ve spoken about this work before at the Simons Institute workshop on Interactive Learning. It achieves state of the art clustering error on several benchmark datasets using very few pairwise cluster queries.

We also just finished a week at the SIAM Optimization conference, where our mini-symposium on Non-convex Optimization in Data Analysis was a huge hit. We had a full room for each session and 12 outstanding talks. Thanks to my co-organizers Stephen Wright, Rebecca Willett, and Rob Nowak, and thanks to all the speakers and participants.

MIDAS seminar and new results

Last Friday I gave the MIDAS weekly seminar. You can find the description here, along with the link directly to the recording. I talked about two recent problems I have been working on: First I talked about my work with Ravi Ganti and Rebecca Willett on learning a low-rank matrix that is observed through a monotonic function from partial measurements. This is common in calibration and quantization problems. Follow up work with Nikhil Rao and Rob Nowak in addition generalized this to learning structured single index models. Second, I talked about the work of my student David Hong, co-advised by Jeff Fessler, on the asymptotic performance of PCA with heteroscedastic data. This is common in problems like sensor networks or medical imaging, where different measurements of the same phenomenon are taken with different quality sensing (eg high or low radiation). David has recently posted his paper on arxiv showing predictions of the asymptotic performance; exploiting the structure of these expressions we also showed that asymptotic recovery for a fixed average noise variance is maximized when the noise variances are equal (i.e., when the noise is in fact homoscedastic). Average noise variance is often a practically convenient measure for the overall quality of data, but our results show that it gives an overly optimistic estimate of the performance of PCA for heteroscedastic data.