Abstract

The engineering and scientific study of large-scale systems is now a major focus in technology, biology, sociology, and cognitive science. Deciding where, when, what and how to `sense’ or `measure’ is a crucial question in the scientific study of such systems. For example, to understand the nature of the Internet or an ecosystem one would ideally like to monitor it everywhere and all the time. This is impossible, and consequently, engineers and scientists must select a relatively small number of times, places and features for measurement. Similar problems arise in systems biology. Biological systems are not defined by the independent functions of individual genes, but rather they depend on the complex interactions of thousands of genes, proteins, and small molecules. The dynamic interplay of these elements is precisely coordinated by signaling networks that orchestrate their interactions. High-throughput experimental techniques now provide biologists with incredibly rich and potentially revealing datasets. But it is impossible to exhaustively explore the full experimental space, and so scientists must judiciously choose which experiments to perform. The complexity and high-dimensionality of these systems makes it incredibly difficult for humans alone to manage and optimize sensing processes. A grand challenge for engineering and scientific discovery in the 21st century is to devise machines that automatically control and adapt sensing operations in large-scale systems.

Traditional approaches to sensing and information processing are non-adaptive in the sense that all data are collected prior to analysis and processing. One can envision, however, adaptive strategies in which information gleaned from previously collected data is used to guide the selection of new data. This talk focuses on the emerging theory and practice of adaptive sensing systems. I will show that automatic feedback from data analysis to data collection can be crucial for effective learning and inference. To illustrate the role of feedback, I will describe two adaptive sensing systems. First, I will consider binary-valued prediction (classification) problems arising, for example, in wireless sensing applications. Under a fixed sensing or communication budget, the prediction errors of adaptive sensor networks can be exponentially smaller than those of non-adaptive sensing systems. Second, motivated by problems in systems biology, I will discuss the role of adaptive sensing in the recovery of sparse signals in noise. I will show that certain weak, sparse patterns are imperceptible in non-adaptive measurements, but can be recovered perfectly using adaptive sensing and experimental design.