Cross-campus collaboration fuels research and innovation, and is a critical and valued endeavor for the University of Wisconsin–Madison. We invite you to explore the many opportunities for collaborating on campus and discover the breadth of institutional-wide research.
Secure Sharing of Clinical History and Genetic Data
This project develops differential privacy versions of leading machine learning algorithms, and tests their preservation of utility when learning predictive models for personalized medicine. It also develops and tests secure local environments for ensuring end-to-end security, a secure virtual environment for privacy-preserving data analysis “in the cloud,” and anonymizing data publishing algorithms.
Funding
National Institutes of Health
Safe Semi-supervised Learning
To use unlabeled data or not, that is the question. It is known that semi-supervised learning can be inferior to supervised learning if its model assumption is violated. Can we design semi-supervised algorithms which are provably robust to such failure? The challenge is to detect model assumption violation from limited labeled data, where semi-supervised learning is most useful.
Funding
National Science Foundation
A multimodal risk prediction approach for breast cancer using MRI and genomic data
Breast magnetic resonance imaging is a new technology that offers high sensitivity for diagnosis and treatment of breast cancer. In particular, breast MRI allows visual assessment of breast density, which relates strongly to breast cancer risk. The goal of this project is to develop new biomarkers for breast cancer using breast MR images concurrently with genomic data. Combining texture analysis methods to describe the appearance of breast images simultaneously with features derived from the genetic profiles of individuals, we will develop new methods for characterizing outcomes such as response to treatment and predicting future malignancy.
Fighting Bullying with Machine Learning
Bullying is a serious national health issue. Social science study of bullying traditionally has used personal surveys in schools, suffering from small sample size and low temporal resolution. We are developing novel machine learning models to study bullying. Our model aims to reconstruct a bullying event — who the bullies, victims, and witnesses are, and what happened to them — from publicly available social media posts. Our model and data can improve the scientific study, intervention and policy-making of bullying.
Funding
National Science Foundation
Human iPS/ES Cell-based Models for Predictive Neural Toxicity and Teratogenicity
This project applies supervised machine learning to RNAseq gene expression data from pluripotent cells exposed to neurotoxins and non-neurotoxins. Supervised learning is used to construct models to predict the neurotoxicity of new compounds from the gene expression patterns they induce in the same cells.
Funding
National Institutes of Health
From Social Media to Knowledge
It is one thing to collect tons of tweets; it is another to turn them into knowledge. We develop machine learning models for this purpose. For example, our Socioscope model can robustly estimate spatio-temporal distributions of any target phenomenon using inhomogeneous Poisson point process (ECML-PKDD 2012 Best Paper).
Funding
University of Wisconsin
Fish and Wildlife Service: Connectivity Optimization
Details about this project are found in the story “Bigger bang for your buck: Restoring fish habitat by removing barriers.”
Collaborators
Enhancing Human Learning using Computational Learning Theory
The project goal is to develop computational learning models and theory to predict and influence human learning behaviors.
Funding
National Science Foundation
Accelerated Renewable Energy
Economic forces continue to drive long-term trends increasing the size of agricultural operations. These trends support expansion of CAFO (confined animal feeding operations) livestock systems as adding more animals means more milk/meat output to generate more cash flow/revenues; and to spread fixed costs and management over more units of output. Increasing CAFO size increases manure volumes and Clean Water Act (CWA) regulatory issues that arise when manure volumes exceed the land base (crop nutrient demand) required to efficiently use manure nutrients without significant environmental losses. We have built a farm-scale, infinite time-horizon optimization model, to aid farmers’ decisions on crop/rotation planning and nutrient sources applications. The objective is to minimize annual operational cost while respecting environmental constraints such as Phosphorus and Soil Loss. Our model can be used to evaluate the impacts of new manure separation techniques in the long term, or to recommend new such technological development/research. It can be easily modified to be a short-term planning for 1-2 rotational time period (8-16 years). We capture the environmental factors by interfacing a sophisticated environmental model, SNAP+ which uses the RUSLE2 for computing soil loss, in a way that new knowledge/updates in these models can automatically translate into our optimization models. The model is flexible that can be extendable by incorporating more specific details; for example, location-specific crop rotation requirement or crop yield.
Funding
U.S. Department of Agriculture (Biomass Research & Development Initiative)
Multifaceted Mathematics Center
The production, distribution, storage, and use of electrical energy are undergoing significant changes. Demand and production patterns are being altered radically by the advent of “smart grids,” renewable generation, hybrid electric vehicles and storage technologies, and by new regulatory constraints. The uncertainties and challenges such as discrete nature, dynamics, and multiple scales—raise a critical need in the energy sector for new mathematical and computational tools. The M2AC2S Center is a locus for a multifaceted, integrated applied mathematics program for the DOE grand challenge of complex electrical energy systems and related infrastructure. Motivated by the applications to complex energy systems, we perform leading-edge research in fundamental mathematical challenges in predictive modeling, integrative mathematical analysis and abstraction frameworks, mathematics of decision, and scalable algorithms. We aim to make integrated advances across these areas and bring them to bear on subchallenges in complex electrical energy systems and related infrastructure: integrated grid and infrastructure planning under sustainability considerations; next-generation architectures for electricity generation, storage and distribution; real-time interconnect-wide system model calibration and prediction; and predictive control of cascading blackouts and real-time contingency analysis. The research program intends to foster integration of mathematical techniques and cross-fertilization across math disciplines.
Funding
Department of Education
Center for Predictive Computational Phenotyping (CPCP)
This project develops improved algorithms for predictive computational phenotyping with applications to electronic health records, transcriptomics, epigenetics, medical imaging data and breast cancer. Research topics for all these applications include data management, dimensionality reduction, graphical models, value of information and high-throughput computation.
Funding
National Institutes of Health
Website
Collaborators
- Miron Livny, Computer Sciences
- Jignesh Patel, Computer Sciences
- AnHai Doan, Computer Sciences
- Mark Craven, Biostatistic & Medical Informatics
- Michael Newton, Biostatistics & Medical Informatics
- C David Page, Biostatistics & Medical Informatics and Computer Sciences
- Jerry Zhu, Computer Sciences
- Sunduz Keles, Biostatistics & Medical Informatics
- Paul Rathouz, Biostatistics & Medical Informatics