Browsing by Author "Schrater, Paul"

Now showing 1 - 9 of 9
  • Thumbnail Image
    Item
    Detecting and Forecasting Economic Regimes in Automated Exchanges
    (2007-03-05) Ketter, Wolfgang; Collins, John; Gini, Maria; Gupta, Alok; Schrater, Paul
    We present basic building blocks of an agent that can use observable market conditions to characterize the microeconomic conditions of the market and predict future market trends. The agent can use this information to make both tactical decisions such as pricing and strategic decisions such as product mix and production planning. We develop methods that can learn dominant market conditions, such as over-supply or scarcity, from historical data using computational methods to construct price density functions. We discuss how this knowledge can be used, together with real-time observable information, to identify the current dominant market condition and to forecast market changes over a planning horizon. We validate our methods by presenting experimental results in a case study, the Trading Agent Competition for Supply Chain Management.
  • Thumbnail Image
    Item
    Effects of Shape and Contact Location Uncertainty on Grasp Quality
    (2006-06-08) Christopoulos, Vassilios N.; Schrater, Paul
    This paper addresses a new approach to the problem of selecting contact locations for grasping in the presence of shape and contact location uncertainty. Focusing on two-dimensional planar objects and two-finger grasps for simplicity, we present a principled approach for selecting contact points by minimizing the risk of force-closure failure. The key contribution of the paper is the risk measure defined on grasp plans. The risk measure successfully distinguishes grasps that are equivalent without uncertainty, and we illustrate the properties of the measure with simulation experiments in two classes of objects.
  • Thumbnail Image
    Item
    Finding What the Driver Does
    (2005-05-01) Veeraraghavan, Harini; Atev, Stefan; Bird, Nathaniel; Schrater, Paul; Papanikolopoulos, Nikolaos P
    Most research depends on detection of driver alertness through monitoring the eyes, face, head or facial expression. This research presents methods for recognizing and summarizing the activities of drivers using the appearance of the driver's position, and changes in position, as fundamental cues, based on the assumption that periods of safe driving are periods of limited motion in the driver's body. The system uses a side-mounted camera and utilizes silhouettes obtained from skin color segmentation for detecting activities. The unsupervised method uses agglomerative clustering to represent driver activity throughout a sequence, while the supervised learning method uses a Bayesian eigen image classifier to distinguish between activities. The results validate the advantages of using driver appearance obtained from skin color segmentation for classification and clustering purposes. Advantages include increased robustness to illumination variations and elimination of the need for tracking and pose determination.
  • Thumbnail Image
    Item
    Internal models for object manipulation: Determining optimal contact locations
    (2006-02-13) Schrater, Paul; Schlicht, Erik J.
    Although there are an infinite number of ways that humans can lift an object, they tend to reach in a predictable manner. This suggests that people are aware of, and optimizing, some sort of loss function. This paper outlines a natural loss function that may be used to predict people's actions in everyday reaching tasks. The loss function is based on the physics of object manipulation and the assumption that people are planning for the intended motion of the object. Using this framework, we are able to make predictions about how people should reach if they are minimizing their expected risk. To test the model, we required people to reach to objects at varying orientations. Our experimental results indicate that people are reaching in a manner that minimizes their expected risk for the task. These findings suggest that people planning for the intended motion of the object and that our brain is aware of the physics involved with object manipulation.
  • Thumbnail Image
    Item
    Measuring spontaneous devaluations in user preferences
    (2013-04-09) Kapoor, Komal; Srivastava, Nisheeth; Srivastava, Jaideep; Schrater, Paul
    Spontaneous devaluation in preferences is ubiquitous, where yesterday's hit is today's affliction. Despite technological advances facilitating access to a wide range of media commodities, finding engaging content is a major enterprise with few principled solutions. Systems tracking spontaneous devaluation in user preferences can allow prediction of the onset of boredom in users potentially catering to their changed needs. In this work, we study the music listening histories of Last.fm users focusing on the changes in their preferences based on their choices for different artists at different points in time. A hazard function, commonly used in statistics for survival analysis, is used to capture the rate at which a user returns to an artist as a function of exposure to the artist. The analysis provides the first evidence of spontaneous devaluation in preferences of music listeners. Better understanding of the temporal dynamics of this phenomenon can inform solutions to the similarity-diversity dilemma of recommender systems.
  • Thumbnail Image
    Item
    Reaching to multiple potential targets: An optimal control perspective
    (2011-06-08) Christopoulos, Vassilios N.; Schrater, Paul
    Living in a dynamic environment, we must be able to make flexible plans that can handle ambiguity and changes in goals while acting. Recent studies suggest that brain builds multiple competing plans related to potential goals and use perceptual information to drive this competition, until a single policy is selected. We propose an extended optimal control framework to model human behavior in tasks with multiple goals and show that goal competition is a natural by-product of handling goal uncertainty. We show how an agent's optimal policy in the presence of goal ambiguity, can be expressed as a weighted mixture of multiple control policies, each of which produces a sequence of actions associated with a specific target. At any instant, weighting factor is an inference of which goal's policy is best to follow, starting from the current state. Simulations of our multiple-goal optimal control model replicated reaching strategies observed in several human studies. Finally, we made novel predictions about the effects of the spatial probability distributions of the candidate targets and their expected pay-off values on the optimal policy.
  • Thumbnail Image
    Item
    Spatial Contextual Classification and Prediction Models for Mining Geospatial Data
    (2002-02-14) Shekhar, Shashi; Schrater, Paul; Vatsavai, Ranga R.; WuLi, Wei; Chawla, Sanjay
    Modeling spatial context (e.g., autocorrelation) is a key challenge in classification problems that arise in geospatial domains. Markov Random Fields (MRFs) is a popular model for incorporating spatial context into image segmentation and land-use classification problems. The spatial autoregression model (SAR), which is an extension of the classical regression model for incorporating spatial dependence, is popular for prediction and classification of spatial data in regional economics, natural resources, and ecological studies. There is little literature comparing these alternative approaches to facilitate the exchange of ideas (e.g., solution procedures). We argue that the SAR model makes more restrictive assumptions about the distribution of feature values and class boundaries than MRF. The relationship between SAR and MRF is analogousto the relationship between regression and Bayesian classifiers. This paper provides comparisons between the two models using a probabilistic and an experimental framework. Keywords: Spatial Context, Spatial Data Mining, Markov Random Fields, Spatial Autoregression.
  • Thumbnail Image
    Item
    Tactical and Strategic Sales Management for Intelligent Agents Guided By Economic Regimes
    (2008-10-20) Ketter, Wolfgang; Collins, John; Gini, Maria; Gupta, Alok; Schrater, Paul
    We present a computational approach that autonomous software agents can adopt to make tactical decisions, such as product pricing, and strategic decisions, such as product mix and production planning, to maximize profit in markets with supply and demand uncertainties. Using a combination of machine learning and optimization techniques, the agent is able to characterize economic regimes, which are historical microeconomic conditions reflecting situations such as over-supply and scarcity. We assume an agent is capable of using real-time observable information to identify the current dominant market condition and we show how it can forecast regime changes over a planning horizon. We demonstrate how the agent can then use regime characterization to predict prices, price trends, and the probability of receiving a customer order in a dynamic supply chain environment. We validate our methods by presenting experimental results from a testbed derived from the Trading Agent Competition for Supply Chain Management (TAC SCM). The results show that our agent outperforms traditional short- and long-term predictive methodologies (such as exponential smoothing) significantly, resulting in accurate prediction of customer order probabilities, and competitive market prices. This, in turn, has the potential to produce higher profits. We also demonstrate the versatility of our computational approach by applying the methodology to prediction of stock price trends.
  • Thumbnail Image
    Item
    Temporal Sequence Prediction Using an Actively Pruned Hypothesis Space
    (2004-04-14) Jensen, Steven; Boley, Daniel; Gini, Maria; Schrater, Paul
    We propose a novel time/space efficient method for learning temporal sequences that operates on-line, is rapid (requiring few exemplars), and adapts easily to changes in the underlying stochastic world model. This work is motivated by humans' remarkable ability to learnspatio-temporal patterns and make short-term predictions much faster than most existing machine learning methods. Using a short-term memory of recent observations, our method maintains a dynamic space of candidate hypotheses in which the growth of the space is systematically and dynamically pruned using an entropy measure over the observed predictive quality of each candidate hypothesis. We further demonstrate the application of this method in the domain of ``matching pennies'' and ``rock-paper-scissors'' games.