AT&T Labs Research — Leading Invention, Driving Innovation

Steven Phillips speaks about our unique research environment in this video:

The definitive version was published in AAAI conference. , 2011-08-07

{In order to better protect and conserve biodiversity, ecologists are making increasing use of machine learning and statistical modeling to understand how species respond to their environment and to predict how they will respond to future climate change, habitat loss and other threats. A fundamental modeling task is to estimate the conditional probability that a given species is present in (or uses) a site, conditional on environmental variables such as precipitation and temperature. For a limited number of species, survey data consisting of both presence and absence records are available, and can be used to fit a variety of conventional classification and regression models. For most species, however, the available data consists only of occurrence data�locations where the species has been observed. In two closely-related but separate bodies of ecological literature, a diversity of specialpurpose models have been developed that contrast occurrence data with a random sample of available environmental conditions. The most widespread statistical approaches involve either fitting an exponential model of species� conditional probability of presence, or fitting a naive logistic model in which the random sample of available conditions is treated as absence data; both approaches have wellknown problems, and in particular, do not necessarily produce valid probabilities. In this paper, after summarizing existing methods and their drawbacks, we overcome those drawbacks by introducing a new scaled binomial loss function that is straightforward to integrate into existing methods such as GLM, GAM, and boosted regression trees, in order to estimate an underlying logistic model of species presence/ absence. Our approach is simpler than the Expectation- Maximization approach of Ward et al., which has not yet been used by ecologists despite giving valid probabilities. Following Ward et al., our method requires an estimate of population prevalence, since prevalence is not identifiable from occurrence data alone. We demonstrate that recent approaches (such as the weighted distribution method of Lele and Keim) that try to avoid the identifiability issue by making parametric data assumptions do not typically produce valid probability estimates. Lastly, we introduce two additional new methods based on maximum entropy and a Chernoff bound that both also estimate the underlying logistic model given an estimate of prevalence.}

"The definitive version is available at onlinelibrary.wiley.com." , Volume 17, 2011-01-01

{MaxEnt is a program for modelling species distributions from presence-only species records. This paper is written for ecologists and describes the MaxEnt model from a statistical perspective, making explicit links between the structure of the model, decisions required in producing a modelled distribution, and knowledge about the species and the data that might affect those decisions. To begin we discuss the characteristics of presence-only data, highlighting implications for modelling distributions. We particularly focus on the problems of sample bias and lack of information on species prevalence. The keystone of the paper is a new statistical explanation of MaxEnt which shows that the model minimizes the relative entropy between two probability densities (one estimated from the presence data and one, from the landscape) defined in covariate space. For many users, this viewpoint is likely to be a more accessible way to understand the model than previous ones that rely on machine learning concepts. We then step through a detailed explanation of MaxEnt describing key components (e.g. covariates and features, and definition of the landscape extent), the mechanics of model fitting (e.g. feature selection, constraints and regularization) and outputs. Using case studies for a Banksia species native to south-west Australia and a riverine fish, we fit models and interpret them, exploring why certain choices affect the result and what this means. The fish example illustrates use of the model with vector data for linear river segments rather than raster (gridded) data. Appropriate treatments for survey bias, unprojected data, locally restricted species, and predicting to environments outside the range of the training data are demonstrated, and new capabilities discussed. Online appendices include additional details of the model and the mathematical links between previous explanations and this one, example code and data, and further information on the case studies. }

The definitive version was published in Biological Conservation (Elsevier). , Volume 143, 2010-07-01

{Indices for site prioritization are widely 2 used to address the question: which sites are most important for conservation of biodiversity? We investigate the theoretical underpinnings of target-based prioritization, which measures sites' contribution to achieving predetermined conservation targets. We show a strong connection between site prioritization and the mathematical theory of voting power. Current site prioritization indices are afflicted by well-known paradoxes of voting power: a site can have zero priority despite having non-zero habitat (the paradox of dummies) and discovery of habitat in a new site can raise the priority of existing sites (the paradox of new members). These paradoxes arise because of the razor's edge nature of voting, and therefore we seek a new index that is not strictly based on voting. By negating such paradoxes, we develop a set of intuitive axioms that an index should obey. We introduce a simple new index, "fraction-of-spare," that satisfies all the axioms. For ingle-species site prioritization, the fraction-of-spare(s) of a site s equals zero if s has no habitat for the species and one if s is essential for meeting the target area for the species. In-between those limits it is linearly interpolated, and equals area(s) / (total area - target). In an evaluation involving multi-year scheduling of site acquisitions for conservation of forest types in New South Wales under specified clearing rates, fraction-of-spare outperforms 58 existing prioritization indices. We also compute the optimal schedule of acquisitions for each of three evaluation measures (under the assumed clearing rates) using integer programming, which indicates that there is still potential for improvement in site prioritization for conservation scheduling. }

(c) ACM, 2010. This is the author's version of the work. It is posted here by permission of ACM for your personal use. Not for redistribution.

The definitive version was published in ACM GreenMetrics Workshop , Volume 38, Issue 3, 2010-10-10.

{Information and communications technology accounts for a significant fraction of worldwide electricity consumption. Given the relentless growth of demand for communications services, telecommunications providers will need to transition to more energy-efficient technology in order to limit their environmental impact. Here we focus on priority-setting for the transition process. In particular, we introduce a method for statistically inferring the electricity consumption of different components of the installed base of telecommunications equipment, while avoiding the high cost of performing direct measurements. Our method relies only on databases of installed equipment in central offices, together with aggregate electricity consumption per office. It takes advantage of inter-office variation in installed equipment to partition per-office electricity consumption by major equipment type. When applied to a collection of 3,918 central offices of a major U.S. telecommunications provider, our approach reveals the (previously unknown) network-wide energy consumption of each major type of equipment. In particular, we find that electricity consumption is dominated by Class-5 telephone switches, which account for 43% of aggregate consumption, and which should therefore be a primary target of central office electricity conservation efforts.}

Awards
AT&T Science & Technology Medal, 2009. For invention, leadership and advocacy of maximum entropy methods and Maxent software in conservation biology.