Near-optimal sensor placements in Gaussian processes

Guestrin, Carlos; Krause, Andreas; Singh, Ajit P.

doi:10.1145/1102351.1102385

Cited by 353 publications

(349 citation statements)

References 15 publications

Supporting

Mentioning

345

Contrasting

Order By: Relevance

“…This is referred to as experimental design and can significantly reduce the number of experiments required to train a statistical model. There is a welldeveloped theory for designing informative experiments using Gaussian process models, which has been applied to a number of problems including environmental monitoring and traffic prediction (18)(19)(20). Experimental design can be posed as a combinatorial optimization problem, where the objective quantifies the informativeness of a set of observations, typically as a function of their covariance matrix.…”

Section: Resultsmentioning

confidence: 99%

Navigating the protein fitness landscape with Gaussian processes

Romero

Krause

Arnold

2012

Proc. Natl. Acad. Sci. U.S.A.

Self Cite

295

311

View full text Add to dashboard Cite

Knowing how protein sequence maps to function (the "fitness landscape") is critical for understanding protein evolution as well as for engineering proteins with new and useful properties. We demonstrate that the protein fitness landscape can be inferred from experimental data, using Gaussian processes, a Bayesian learning technique. Gaussian process landscapes can model various protein sequence properties, including functional status, thermostability, enzyme activity, and ligand binding affinity. Trained on experimental data, these models achieve unrivaled quantitative accuracy. Furthermore, the explicit representation of model uncertainty allows for efficient searches through the vast space of possible sequences. We develop and test two protein sequence design algorithms motivated by Bayesian decision theory. The first one identifies small sets of sequences that are informative about the landscape; the second one identifies optimized sequences by iteratively improving the Gaussian process model in regions of the landscape that are predicted to be optimized. We demonstrate the ability of Gaussian processes to guide the search through protein sequence space by designing, constructing, and testing chimeric cytochrome P450s. These algorithms allowed us to engineer active P450 enzymes that are more thermostable than any previously made by chimeragenesis, rational design, or directed evolution.protein engineering | recombination | machine learning | experimental design | active learning I n the mapping of protein sequence to protein behavior, the phenotype can be envisioned as a surface, or landscape, over the high-dimensional space of possible sequences (1). This "fitness landscape" could describe how the protein contributes to organismal fitness, or it may represent a biophysical property, such as stability, enzyme activity, or ligand binding affinity. The structure of this surface describes the spectrum of possible phenotypes as well as the mutational accessibility among them and therefore strongly influences protein evolution. This surface is also the objective function for protein engineering, which seeks to identify protein sequences that are highly optimized for a given property or set of properties.Identifying such optimized sequences is extremely challenging for several reasons. First, the space of possible protein sequences is incomprehensibly large and will never be searched exhaustively by any means, naturally, in the laboratory, or computationally (2, 3). Second, within this vast space, functional proteins are extremely scarce, with estimates that range from a high of 1 in 10 11 to as little as 1 in 10 77 (4,5). Of the sequences that are functional, most have poor fitness and their numbers decrease exponentially with higher levels of fitness (6, 7). Thus, highly fit sequences are vanishingly rare and overwhelmed by nonfunctional and mediocre sequences.Computational protein engineering uses models of protein function to guide a search for optimized sequences. These models typically contain an atomic struc...

show abstract

Section: Resultsmentioning

confidence: 99%

Navigating the protein fitness landscape with Gaussian processes

Romero

Krause

Arnold

2012

Proc. Natl. Acad. Sci. U.S.A.

Self Cite

295

311

View full text Add to dashboard Cite

show abstract

“…We note there exist other submodular measures of information gain such as mutual information with respect to a reference set [5]. A formal treatment of the mutual information criterion is beyond our scope, suffice to say one can often optimize it using similar methods [4]. Sample selection is comparatively simple for a 1D time series.…”

Section: B Sample Selectionmentioning

confidence: 99%

“…This problem equates to active learning [2] in which the agent allocates future measurements to reduce uncertainty as measured by the Shannon entropy of the process at all time steps. Common measures of information gain are the entropy of the measurements themselves [3] or their mutual information with respect to unobserved time steps [4]. Submodular optimization algorithms can efficiently solve a broad class of these cost functions [5].…”

Section: Introductionmentioning

confidence: 99%

Adaptive sensing of time series with application to remote exploration

Thompson

Cabrol

Furlong

et al. 2013

2013 IEEE International Conference on Robotics and Automation

View full text Add to dashboard Cite

Abstract-We address the problem of adaptive informationoptimal data collection in time series. Here a remote sensor or explorer agent throttles its sampling rate in order to track anomalous events while obeying constraints on time and power. This problem is challenging because the agent has limited visibility -all collected datapoints lie in the past, but its resource allocation decisions require predicting far into the future. Our solution is to continually fit a Gaussian process model to the latest data and optimize the sampling plan on line to maximize information gain. We compare the performance characteristics of stationary and nonstationary Gaussian process models. We also describe an application based on geologic analysis during planetary rover exploration. Here adaptive sampling can improve coverage of localized anomalies and potentially benefit mission science yield of long autonomous traverses.

show abstract

“…cSamp-T provides more fine-grained flow coverage objectives and reduces duplicated flow reports. Sensor network monitoring: There has been recent work applying the theory of maximizing submodular functions in sensor networks [34,35]. The problem of placing sensors robust to adversarial objectives [11] is conceptually similar to maximizing the minimum fractional coverage.…”

Section: Other Related Workmentioning

confidence: 99%

Coordinated sampling sans Origin-Destination identifiers: Algorithms and analysis

Sekar

Gupta

Reiter

et al. 2010

2010 Second International Conference on COMmunication Systems and NETworks (COMSNETS 2010)

View full text Add to dashboard Cite

Abstract-Flow monitoring is used for a wide range of network management applications. Many such applications require that the monitoring infrastructure provide high flow coverage and support fine-grained network-wide objectives. Coordinated Sampling (cSamp) is a recent proposal that improves the monitoring capabilities of ISPs to address these demands.In this paper, we address a key deployment impediment for cSamp-like solutions-the need for routers to determine the Origin-Destination (OD) pair of each packet. In practice, however, this information is not available without expensive changes. We present a new framework called cSamp-T, in which each router uses only local information, instead of the OD-pair identifiers. Leveraging results from the theory of maximizing submodular set functions, cSamp-T provides near-ideal performance in maximizing the total flow coverage in the network. Further, with a small amount of targeted upgrades to a few routers, cSamp-T nearly optimally maximizes the minimum fractional coverage across all OD-pairs. We demonstrate these results on a range of real topologies.

show abstract

Near-optimal sensor placements in Gaussian processes

Cited by 353 publications

References 15 publications

Navigating the protein fitness landscape with Gaussian processes

Navigating the protein fitness landscape with Gaussian processes

Adaptive sensing of time series with application to remote exploration

Coordinated sampling sans Origin-Destination identifiers: Algorithms and analysis

Contact Info

Product

Resources

About