Amy McGovern

Exploiting relational structure to understand publication patterns in high-energy physics

We analyze publication patterns in theoretical high-energy physics using a relational learning approach. We focus on four related areas: understanding and identifying patterns of citations, examining publication patterns at the author level, predicting whether a paper will be accepted by specific journals, and identifying research communities from the citation patterns and paper text. Each of these analyses contributes to an overall understanding of theoretical high-energy physics.

Identifying predictive multi-dimensional time series motifs: an application to severe weather prediction

We introduce an efficient approach to mining multi-dimensional temporal streams of real-world data for ordered temporal motifs that can be used for prediction. Since many of the dimensions of the data are known or suspected to be irrelevant, our approach first identifies the salient dimensions of the data, then the key temporal motifs within each dimension, and finally the temporal ordering of the motifs necessary for prediction. For the prediction element, the data are assumed to be labeled. We tested the approach on two real-world data sets. To verify the generality of the approach, we validated the application on several subjects from the CMU Motion Capture database. Our main application uses several hundred numerically simulated supercell thunderstorms where the goal is to identify the most important features and feature interrelationships which herald the development of strong rotation in the lowest altitudes of a storm. We identified sets of precursors, in the form of meteorological quantities reaching extreme values in a particular temporal sequence, unique to storms producing strong low-altitude rotation. The eventual goal is to use this knowledge for future severe weather detection and prediction algorithms.

Building a Basic Block Instruction Scheduler with Reinforcement Learning and Rollouts

The execution order of a block of computer instructions on a pipelined machine can make a difference in running time by a factor of two or more. Compilers use heuristic schedulers appropriate to each specific architecture implementation to achieve the best possible program speed. However, these heuristic schedulers are time-consuming and expensive to build. We present empirical results using both rollouts and reinforcement learning to construct heuristics for scheduling basic blocks. In simulation, the rollout scheduler outperformed a commercial scheduler on all benchmarks tested, and the reinforcement learning scheduler outperformed the commercial scheduler on several benchmarks and performed well on the others. The combined reinforcement learning and rollout approach was also very successful. We present results of running the schedules on Compaq Alpha machines and show that the results from the simulator correspond well to the actual run-time results.

Mobile agents on the digital battlefield

The Domain Adaptive Information System (DAIS) is a mobile intebigent agent system for information discovery and dissemination in a military intelligence network DAIS is used reliably over low bandwidth miIitary radio networks to facihtate Counter InteIligcnce/Human Intelligence (CWHUMINT) operations. DAIS tested and validated two hypotheses: that agent technology could improve the efficiency of military tactical operations; and that mobile agents would outperform static agents in the unrcliablc, low bandwidth networks used in military operations. DAIS increased by two orders of magnitude military intelligence analysts’ access to distributed information. This paper presents the design of the system, the cxpcrimcnts, and the results, as well as the future directions of DAIS.

Spatiotemporal Relational Probability Trees: An Introduction

We introduce spatiotemporal relational probability trees (SRPTs), probability estimation trees for relational data that can vary in both space and time. The SRPT algorithm addresses the exponential increase in search complexity through sampling. We validate the SRPT using a simulated data set and we empirically demonstrate the SRPT algorithm on two real-world data sets.

Classification of Convective Areas Using Decision Trees

Abstract This paper presents an automated approach for classifying storm type from weather radar reflectivity using decision trees. Recent research indicates a strong relationship between storm type (morphology) and severe weather, and such information can aid in the warning process. Furthermore, new adaptive sensing tools, such as the Center for Collaborative Adaptive Sensing of the Atmosphere’s (CASA’s) weather radar, can make use of storm-type information in real time. Given the volume of weather radar data from those tools, manual classification of storms is not possible when dealing with real-time data streams. An automated system can more quickly and efficiently sort through real-time data streams and return value-added output in a form that can be more easily manipulated and understood. The method of storm classification in this paper combines two machine learning techniques: K-means clustering and decision trees. K-means segments the reflectivity data into clusters, and decision trees classify eac...

Using spatiotemporal relational random forests to improve our understanding of severe weather processes

Major severe weather events can cause a significant loss of life and property. We seek to revolutionize our understanding of and our ability to predict such events through the mining of severe weather data. Because weather is inherently a spatiotemporal phenomenon, mining such data requires a model capable of representing and reasoning about complex spatiotemporal dynamics, including temporally and spatially varying attributes and relationships. We introduce an augmented version of the Spatiotemporal Relational Random Forest, which is a random forest that learns with spatiotemporally varying relational data. Our algorithm maintains the strength and performance of random forests but extends their applicability, including the estimation of variable importance, to complex spatiotemporal relational domains. We apply the augmented Spatiotemporal Relational Random Forest to three severe weather data sets. These are: predicting atmospheric turbulence across the continental United States, examining the formation of tornadoes near strong frontal boundaries, and understanding the spatial evolution of drought across the southern plains of the United States. The results on such a wide variety of real-world domains demonstrate the extensive applicability of the Spatiotemporal Relational Random Forest. Our long-term goal is to significantly improve the ability to predict and warn about severe weather events. We expect that the tools and techniques we develop will be applicable to a wide range of complex spatiotemporal phenomena.

Enhancing understanding and improving prediction of severe weather through spatiotemporal relational learning

Severe weather, including tornadoes, thunderstorms, wind, and hail annually cause significant loss of life and property. We are developing spatiotemporal machine learning techniques that will enable meteorologists to improve the prediction of these events by improving their understanding of the fundamental causes of the phenomena and by building skillful empirical predictive models. In this paper, we present significant enhancements of our Spatiotemporal Relational Probability Trees that enable autonomous discovery of spatiotemporal relationships as well as learning with arbitrary shapes. We focus our evaluation on two real-world case studies using our technique: predicting tornadoes in Oklahoma and predicting aircraft turbulence in the United States. We also discuss how to evaluate success for a machine learning algorithm in the severe weather domain, which will enable new methods such as ours to transfer from research to operations, provide a set of lessons learned for embedded machine learning applications, and discuss how to field our technique.

Solar Energy Prediction: An International Contest to Initiate Interdisciplinary Research on Compelling Meteorological Problems

15 As meteorological observing systems and models grow in complexity and number, the size of 16 the data becomes overwhelming for humans to analyze using traditional techniques. Com17 puter scientists, and specifically machine learning and data mining researchers, are develop18 ing frameworks for analyzing big data. The AMS Committee on Artificial Intelligence and 19 its Applications to Environmental Science aims to bring AI researchers and environmental 20 scientists together to increase the synergy between the two. The AI committee has spon21 sored 4 previous contests on a variety of meteorological problems including wind energy, 22 storm classification, winter hydrometeor classification, and air pollution, with the goal of 23 bringing together the two fields of research. Although these were successful, the audience 24 was limited to existing environmental science researchers (usually 10-20 teams of people 25 primarily within the AMS community). For the 2013/14 contest, we expanded to a global 26 audience by focusing on the compelling problem of solar energy prediction and by having 27 the established forum Kaggle host our contest. Using this forum, we had over 160 teams 28 from all around the world participate. Improved solar energy forecasting is a necessary com29 ponent of making solar energy a viable alternative power source. This paper summarizes 30 our experiences in the 2013/14 contest, discusses the data in detail, and presents the win31 ning prediction methods. The contest data come from the NOAA/ESRL Global Ensemble 32 Forecasting System Reforecast Version 2 and the Oklahoma Mesonet with sponsorship from 33 EarthRisk Technologies. All winning methods utilized gradient boosted regression trees but 34 differed in parameter choices and interpolation methods. 35

Machine learning enhancement of storm scale ensemble precipitation forecasts

Precipitation forecasts provide both a crucial service for the general populace and a challenging forecasting problem due to the complex, multi-scale interactions required for precipitation formation. The Center for the Analysis and Prediction of Storms (CAPS) Storm Scale Ensemble Forecast (SSEF) system is a promising method of providing high-resolution forecasts of the intensity and uncertainty in precipitation forecasts. The SSEF incorporates multiple models with varied parameterization scheme combinations and produces forecasts every 4 km over the continental US. The SSEF precipitation forecasts exhibit significant negative biases and placement errors. In order to correct these issues, multiple machine learning algorithms have been applied to the SSEF precipitation forecasts to correct the forecasts using the NSSL National Mosaic and Multisensor QPE (NMQ) grid as verification. The 2010 SSEF was used for training. Two levels of post-processing are performed. In the first, probabilities of any precipitation are determined and used to find optimal thresholds for the precipitation areas. Then, three types of forecasts are produced in those areas. First, the probability of the 1-hour accumulated precipitation exceeding a threshold is predicted with random forests, logistic regression, and multivariate adaptive regression splines (MARS). Second, deterministic forecasts based on a correction from the ensemble mean are made with linear regression, random forests, and MARS. Third, fixed probability interval forecasts are made with quantile regressions and quantile regression forests. Models are generated from points sampled from the western, central, and eastern sections of the domain. Verification statistics and case study results show improvements in the reliability and skill of the forecasts compared to the original ensemble while controlling for the over-prediction of the precipitation areas and without sacrificing smaller scale details from the model runs.