Systems | Information | Learning | Optimization

## Optimal Recovery under Approximability Models, with Applications

For functions acquired through point evaluations, is there an optimal way to estimate a quantity of interest or even to approximate the functions in full? We give an affirmative answer to this question under the novel assumption that the functions belong to a model set defined by approximation capabilities. In …

## Using mobile device data to generate population estimates of fertility parameters

Pregnancy loss is a primary limit on human reproduction and a key driver of population dynamics. It is also very difficult to study. 85% of pregnancy loss happens prior to clinical interaction. We use information from over a million users of a fertility tracking application to construct estimates of pregnancy …

## Entity Matching Meets Data Science: A Progress Report from the Magellan Project

Entity matching (EM) finds data instances that refer to the same real-world entity. In 2015, we started the Magellan project at UW-Madison to build EM systems. Most current EM systems are stand-alone monoliths. In contrast, Magellan borrows ideas from the field of data science (DS), to build a novel kind …

## Large sample asymptotics of spectra of Laplacians and semilinear elliptic PDEs on random geometric graphs.

Given a data set $\mathcal{X}=\{x_1, \dots, x_n\}$ and a weighted graph structure $\Gamma= (\mathcal{X},W)$ on $\mathcal{X}$, graph based methods for learning use analytical notions like graph Laplacians, graph cuts, and Sobolev semi-norms to formulate optimization problems whose solutions serve as sensible approaches to machine learning tasks. When the data set …

## Safety and Robustness Guarantees with Learning in the Loop

In this talk, we present recent progress towards developing learning-based control strategies for the design of safe and robust autonomous systems. Our approach is to recognize that machine learning algorithms produce inherently uncertain estimates or predictions, and that this uncertainty must be explicitly quantified (e.g., using non-asymptotic guarantees of contemporary …

## Towards a science of social interactions

The bulk of research in psychology and neuroscience has focused on studying processes within single individuals. However, outside of the laboratory people rarely operate independently but rather in the context of a complex constellation of social relationships. Social interactions form a critical aspect of the human experience from romantic relationships, …

## Adaptive Sampling for False Discovery Control

In many limited budget experimental settings, such as A/B testing or Protein design, there is a need for adaptive sampling to guide the discovery of as many true positives as possible subject to a low rate of false discoveries (i.e. false alarms). Like active learning for binary classification, this experimental …

## Hardware Accelerators for Deep Learning: A Proving Ground for Specialized Computing

The computing industry has a power problem: the days of ideal transistor scaling are over, and chips now have more devices than can be fully powered simultaneously, limiting performance. New architecture-level solutions are needed to continue scaling performance, and specialized hardware accelerators are one such solution. While accelerators promise to …

## Learning From Sub-Optimal Data

Learning algorithms typically assume their input data is good natured. If one takes this input data and trains an agent with it, then the agent should, given enough time and compute, eventually learn how to solve the intended task. But this is not always a realistic expectation. Sometimes, the data …

## Network Inference from Graph Dynamic Processes

We address the problem of identifying the structure of an undirected graph from the observation of signals defined on its nodes. Fundamentally, the unknown graph encodes direct relationships between signal elements, which we aim to recover from observable indirect relationships generated by a diffusion process on the graph. Our approach …