See also: Presentations and Invited Talks, Google Scholar Citations List, Resume.
Skip to: Computer Science, Physics, Religion.
"Real life tends to cancel out." (Carroll's Razor)
James L. Carroll and Christopher D. Tomkins, "Physics-Based Constraints in the Forward Modeling Analysis of Time-Correlated Image Data," in 11th International Conference on Machine Learning and Applications (ICMLA 2012), Boca Raton, Florida, USA, December 12-15, 2012.
Abstract: The forward-model approach has been shown to produce accurate model reconstructions of scientific measurements for single-time image data. Here we extend the approach to a series of images that are correlated in time using the physics-based constraints that are often available with scientific imaging. The constraints are implemented through a representational bias in the model and, owing to the smooth nature of the physics evolution in the specified model, provide an effective temporal regularization. Unlike more general temporal regularization techniques, this restricts the space of solutions to those that are physically realizable. We explore the performance of this approach on a simple radiographic imaging problem of a simulated object evolving in time. We demonstrate that the constrained simultaneous analysis of the image sequence outperforms the independent forward modeling analysis over a range of degrees of freedom in the physics constraints, including when the physics model is under-constrained. Further, this approach outperforms the independent analysis over a large range of signal-to-noise ratios.
Full long version tech report: pdf,
Short IEEE published paper: ieeexplore,
AMS presentation: pptx,
James L. Carroll, "Epiphenomenalism, the Problem with Property Dualism," in The Proceedings of the 2012 Conference of the Mormon Transhumanist Association, Salt Lake City, Utah, April, 2012.
Abstract: We show that theories of qualia based upon “property dualism” (sometimes called “natural dualism” and “dual aspect theories”) lead to a form of epiphenomenalism, the situation where our behavior does not causally flow from our subjective experiences. That would mean that our claims, beliefs, and memories about our subjective experiences do not directly arise from our real subjective experiences. We will show that this ultimately leads to what I call “zombie solipsism” and to cognitive instability. This argument should allow us to reject “property dualism” in all its forms. If we reject property dualism, we must continue to search for some other explanation for subjective experience.
Kristine Monteith, James L. Carroll, Kevin Seppi, and Tony Martinez, "Turning Bayesian Model Averaging Into Bayesian Model Combination," in Proceedings of the 2011 International Joint Conference on Neural Networks (IJCNN 2011), San Jose, California July 31 - August 5, 2011.
Abstract: Bayesian methods are theoretically optimal in many situations. Bayesian model averaging is generally considered the standard model for creating ensembles of learners using Bayesian methods, but this technique is often outperformed by more ad hoc methods in empirical studies. The reason for this failure has important theoretical implications for our understanding of why ensembles work. It has been proposed that Bayesian model averaging struggles in practice because it accounts for uncertainty about which model is correct but still operates under the assumption that only one of them is. In order to more effectively access the benefits inherent in ensembles, Bayesian strategies should therefore be directed more towards model combination rather than the model selection implicit in Bayesian model averaging. This work provides empirical verification for this hypothesis using several different Bayesian model combination approaches tested on a wide variety of classification problems. We show that even the most simplistic of Bayesian model combination strategies outperforms the traditional ad hoc techniques of bagging and boosting, as well as outperforming BMA over a wide variety of cases. This suggests that the power of ensembles does not come from their ability to account for model uncertainty, but instead comes from the changes in representational and preferential bias inherent in the process of combining several different models.
James L. Carroll, "A Bayesian Decision Theoretical Approach to Supervised Learning, Selective Sampling, and Empirical Function Optimization," BYU PhD. Dissertation, 2010.
Abstract: Many have used the principles of statistics and Bayesian decision theory to model speciﬁc learning problems. It is less common to see models of the processes of learning in general. One exception is the model of the supervised learning process known as the “Extended Bayesian Formalism” or EBF. This model is descriptive, in that it can describe and compare learning algorithms. Thus the EBF is capable of modeling both eﬀective and ineﬀective learning algorithms.
We extend the EBF to model un-supervised learning, semi-supervised learning, supervised learning, and empirical function optimization. We also generalize the utility model of the EBF to deal with non-deterministic outcomes, and with utility functions other than 0-1 loss. Finally, we modify the EBF to create a “prescriptive” learning model, meaning that, instead of describing existing algorithms, our model deﬁnes how learning should optimally take place. We call the resulting model the Uniﬁed Bayesian Decision Theoretical Model, or the UBDTM. We show that this model can serve as a cohesive theory and framework in which a broad range of questions can be analyzed and studied. Such a broadly applicable uniﬁed theoretical framework is one of the major missing ingredients of machine learning theory.
Using the UBDTM, we concentrate on supervised learning and empirical function optimization. We then use the UBDTM to reanalyze many important theoretical issues in Machine Learning, including No-Free-Lunch, utility implications, and active learning. We also point forward to future directions for using the UBDTM to model learnability, sample complexity, and ensembles. We also provide practical applications of the UBDTM by using the model to train a Bayesian variation to the CMAC supervised learner in closed form, to perform a practical empirical function optimization task, and as part of the guiding principles behind an ongoing project to create an electronic and print corpus of tagged ancient Syriac texts using active learning.
James L. Carroll, Neil Toronto, Robbie Haertel, and Kevin D. Seppi, "Explicit Utility in Supervised Learning," in Proceedings of the NIPS 2008 Workshop on Cost-Sensitive Machine Learning, Whistler, British Columbia, Canada.
Abstract: We use a graphical model of the supervised learning problem to explore the theoretical effect of utility on supervised learning, No-Free-Lunch, sample complexity, and active learning. There are two sources of utility that can affect the above problems: utility that comes from end use and utility that comes from sample costs. We explore which parts of these problems depend on utility, and which parts are utility free. Further, we propose a novel interpretation of the No-Free-Lunch theorems that is independent of utility. We propose that sample complexity should be redefined in terms of expected sample costs to achieve a given threshold on expected end use effectiveness (which would be defined in terms of end use utility). Finally, we explore the effects of the sample cost function and the end use utility function on active learning techniques both theoretically (through the optimal active learning equations) and through several examples including a synthetic data set and a real life part of speech tagging scenario.
Presentation poster: pdf.
Robbie Haertel, Kevin Seppi, Eric Ringger, James Carroll, "Return on Investment for Active Learning," in Proceedings of the NIPS 2008 Workshop on Cost-Sensitive Machine Learning, Whistler, British Columbia, Canada.
Abstract: Active Learning (AL) can be defined as a selectively supervised learning protocol intended to present those data to an oracle for labeling which will be most enlightening for machine learning. While AL traditionally accounts for the value of the information obtained, it often ignores the cost of obtaining the information thus causing it to perform sub-optimally with respect to total cost. We present a framework for AL that accounts for this cost and discuss optimality and tractability in this framework. Using this framework we motivate Return On Investment (ROI), a practical, cost-sensitive heuristic that can be used to convert existing algorithms into cost-conscious active learners. We demonstrate the validity of ROI in a simulated AL part-of-speech tagging task on the Penn Treebank in which ROI achieves as high as a 73% reduction in hourly cost over random selection.
Robbie Haertel, Eric Ringger, Kevin Seppi, James Carroll, Peter McClanahan. "Assessing the Costs of Sampling Methods in Active Learning for Annotation," in the Proceedings of the Conference of the Association of Computational Linguistics (ACL-NAACL: HLT 2008), Columbus, Ohio.
Abstract: Traditional Active Learning (AL) techniques assume that the annotation of each datumcosts the same. This is not the case when annotating sequences; some sequences will take longer than others. We show that the AL technique which performs best depends on how cost is measured. Applying an hourly cost model based on the results of an annotation user study, we approximate the amount of time necessary to annotate a given sentence. This model allows us to evaluate the effectiveness of AL sampling methods in terms of time spent in annotation. We acheive a 77% reduction in hours from a random baseline to achieve 96.5% tag accuracy on the Penn Treebank. More significantly, we make the case for measuring cost in assessing AL methods.
Eric Ringger, Marc Carmen, Robbie Haertel, Noel Ellison, Kevin Seppi, Deryle Lonsdale, Peter McClanahan, James Carroll, "Assessing the Costs of Machine-Assisted Corpus Annotation through a User Study," in the Proceedings of the Sixth Language Resources and Evaluation Conference (LREC 2008). Marrakech, Morocco.
Abstract: Fixed, limited budgets often constrain the amount of expert annotation that can go into the construction of annotated corpora. Estimating the cost of annotation is the first step toward using annotation resources wisely. We present here a study of the cost of annotation. This study includes the participation of annotators at various skill levels and with varying backgrounds. Conducted over the web, the study consists of tests that simulate machine-assisted pre-annotation, requiring correction by the annotator rather than annotation from scratch. The study also includes tests representative of an annotation scenario involving Active Learning as it progresses from a naïve model to a knowledgeable model; in particular, annotators encounter pre-annotation of varying degrees of accuracy. The annotation interface lists tags considered likely by the annotation model in preference to other tags. We present the experimental parameters of the study and report both descriptive and inferential statistics on the results of the study. We conclude with a model for estimating the hourly cost of annotation for annotators of various skill levels. We also present models for two granularities of annotation: sentence at a time and word at a time.
James L. Carroll, Robbie Haertel, Peter McClanahan, Eric Ringger, and Kevin Seppi,"Modeling the Annotation Process for Ancient Corpus Creation," in Chatressar 2007, Proceedings of the International Conference of Electronic Corpora of Ancient Languages (ECAL), Prague, November 2007.
Abstract: In corpus creation human annotation is expensive. Annotation costs can be minimized through machine learning and active learning, however there are many complex interactions among the machine learner, the active learning technique, the annotation cost, human annotation accuracy, the annotator user interface, and several other elements of the process. For example, we show that changing the way in which annotators are paid can drastically change the performance of active learning techniques. To date these interactions have been poorly understood. We introduce a decision-theoretic model of the annotation process suitable for ancient corpus annotation that clarifies these interactions and can guide the development of a corpus creation project.
Eric Ringger, Peter McClanahan, Robbie Haertel, George Busby, Marc Carmen, James Carroll, Kevin Seppi, and Deryle Lonsdale, "Active Learning for Part-of-Speech Tagging: Accelerating Corpus Annotation," in Proceedings of the ACL Linguistic Annotation Workshop, Association for Computational Linguistics, Prague, Czech Republic, June 2007, p. 101-108.
Abstract: In the construction of a part-of-speech annotated corpus, we are constrained by a fixed budget. A fully annotated corpus is required, but we can afford to label only a subset. We train a Maximum Entropy Markov Model tagger from a labeled subset and automatically tag the remainder. This paper addresses the question of where to focus our manual tagging efforts in order to deliver an annotation of highest quality. In this context, we find that active learning is always helpful. We focus on Query by Uncertainty (QBU) and Query by Committee (QBC) and report on experiments with several baselines and new variations of QBC and QBU, inspired by weaknesses particular to their use in this application. Experiments on English prose and poetry test these approaches and evaluate their robustness. The results allow us to make recommendations for both types of text and raise questions that will lead to further inquiry.
James L. Carroll and Kevin D. Seppi, "No-Free-Lunch and Bayesian Optimality," in Meta-Learning IJCNN Workshop 2007.
Abstract: We take a Bayesian approach to the issues of bias, meta bias, transfer, overfit, and No-Free-Lunch in the context of supervised learning. If we accept certain relationships between the function class, on training set data, and off training set data, then a graphical model can be created that represents the supervised learning problem. This graphical model dictates a specific algorithm which will be the “optimal” approach to learning the parameters of any given function representation given the variable relationships. Thus, there is an optimal technique for supervised learning. We reconcile this idea of an optimal technique with the ideas of No-Free-Lunch and show how these ideas relate to the concepts of meta and transfer learning through hierarchical versions of the graphical model.
James L. Carroll, Christopher K. Monson, and Kevin D. Seppi, "A Bayesian CMAC for High Assurance Supervised Learning," in Applications of Neural Networks in High-Assurance Systems, IJCNN Workshop 2007.
Abstract: We analyze the drawbacks to using ANNs in high assurance systems and propose a solution based upon a Bayesian approach with a specific network topology that can be solved in closed form. The Bayesian approach leads to better answers in the traditional sense, while also allowing us to quantify risk and deal with it in a reasonable manner. We demonstrate this approach on several synthetic functions and the Abalone data set.
Christopher K. Monson, Kevin D. Seppi, James L. Carroll, "A Utile Function Optimizer," in The 2007 IEEE Congress of Evolutionary Computation (CEC 2007).
Abstract: The Evolutionary Optimization DBN (EO-DBN)---a dynamic Bayesian model of optimization---provides a unique opportunity to create an algorithm that more directly addresses the goal of optimization: to carefully select function samples so as to obtain information about the location of its global optimum. As thus described, optimization is fundamentally a decision process and will therefore be addressed using the language and tools of decision theory. Having once cast the problem as a probabilistic network (the EO-DBN), it is possible to create a decision-making agent that uses explicit definitions of utility and cost to rationally select sample locations that maximize information. This work presents and develops this idea, producing a model of optimization and a corresponding algorithm that is optimal with respect to well-stated optimization goals. The algorithm uses naturally expressed domain knowledge to determine where a function should be sampled and when the sampling process should stop, displaying sophisticated behavior when provided with simple specifications.
Paper: pdf, also available through IEEEXplore.
James L. Carroll Task Localization, Similarity, and Transfer; Towards a Reinforcement Learning Task Library System, BYU Masters Thesis, 2005.
Abstract: This thesis develops methods of task localization, task similarity discovery, and task transfer for eventual use in a reinforcement learning task library system, which can effectively "learn to learn," improving its performance as it encounters various tasks over the lifetime of the learning system.
James L. Carroll, Kevin Seppi "Task Similarity Measures for Transfer in Reinforcement Learning Task Libraries," in Proceedings of the 2005 International Joint Conference on Neural Networks, (IJCNN 2005), 2005.
Abstract: Recent research in task transfer and task clustering has necessitated the need for task similarity measures in reinforcement learning. Determining task similarity is necessary for selective transfer where only information from relevant tasks and portions of a task are transferred. Which task similarity measure to use is not immediately obvious. It can be shown that no single task similarity measure is uniformly superior. The optimal task similarity measure is dependent upon the task transfer method being employed. We define similarity in terms of tasks, and propose several possible task similarity measures, dT , dP , dQ, and dR which are based on the transfer time, policy overlap, Q-values, and reward structure respectively. We evaluate their performance in three separate experimental situations.
James L. Carroll, Kevin Seppi "A Bayesian Technique for Task Localization in Multiple Goal Markov Decision Processes," in The 2004 International Conference on Machine Learning and Applications, (ICMLA 2004).
Abstract: In a reinforcement learning task library system for Multiple Goal Markov Decision Process (MGMDP), localization in the task space allows the agent to determine whether a given task is already in its library in order to exploit previously learned experience. Task localization in MGMDPs can be accomplished through a Bayesian approach, however a trivial approach fails when the rewards are not distributed normally. This can be overcome through our Bayesian Task Localization Technique (BTLT).
Paper: ps, pdf.
James L. Carroll, Todd Peterson, Kevin Seppi "Reinforcement Learning Task Clustering (RLTC)," in The 2003 International Conference on Machine Learning and Applications, (ICMLA 2003).
Abstract: This work represents the first step towards a task library system in the reinforcement learning domain. Task libraries could be useful in speeding up the learning of new tasks through task transfer. Related transfer can increase learning rate and can help prevent convergence to sub-optimal policies in reinforcement learning. Unrelated transfer can be extremely detrimental to the learning rate. Thus task transfer is useful in reinforcement learning if the source task and the target task are sufficiently related. Task similarity in reinforcement learning can be determined using many different similarity metrics, and simple clustering mechanisms can be applied to determine a set of related tasks. Invariants can be determined among the set of related tasks and then used in transfer. This paper uses information gathered from a set of simple grid world tasks to show that clustering of tasks based upon a similarity metric can be helpful in determining the set of source tasks which should be utilized in transfer.
Paper: ps, pdf.
James L. Carroll, Todd S. Peterson "Fixed vs. Dynamic Sub-transfer in Reinforcement Learning," in The 2002 International Conference on Machine Learning and Applications,(ICMLA 2002).
Abstract: We survey various task transfer methods in Q-learning and present a variation on fixed sub-transfer which we call dynamic sub-transfer. We discuss the benefits and drawbacks of dynamic sub-transfer as compared with the other transfer methods, and we describe qualitatively the situations where this method would be prefered over the fixed version of sub-transfer. We test this method against several other transfer methods in a simple three room grid world where portions of the source's policy are relevant to the target task and other portions are not. In this situation we found that dynamic sub-transfer converged to the optimal solution, avoiding the sub-optimality inherent in fixed sub-transfer, while also avoiding some of the convergence problems often experienced by fixed sub-transfer.
Paper: ps, pdf.
James L. Carroll, Todd S. Peterson, Nancy E. Owens, "Memory-guided Exploration in Reinforcement Learning," in The 2001 International Joint Conference on Neural Networks, (IJCNN 2001).
Abstract: The life-long learning architecture attempts to create an adaptive agent through the incorporation of prior knowledge over the lifetime of a learning agent. Our paper focuses on task transfer in reinforcement learning and specifically in Q-learning. There are three main model free methods for performing task transfer in Q-learning: direct transfer, soft transfer and memory-guided exploration. In direct transfer Q-values from a previous task are used to initialize the Q-values of the next task. Soft transfer initializes the Q-values of the new task with a weighted average of the standard initialization value and the Q-values of the previous task. In memory-guided exploration the Q-values of previous tasks are used as a guide in the initial exploration of the agent. The weight that the agent gives to its past experience decreases over time. We explore stability issues related to the off-policy nature of memory-guided exploration and compare memory-guided exploration to soft transfer and direct transfer in three different environments.
Paper: ps, pdf.
Todd S. Peterson, Nancy E. Owens, and James L. Carroll, "Towards Automatic Shaping in Robot Navigation," in The 2001 IEEE International Conference on Robotics and Automation (ICRA, 2001).
Abstract: Shaping is a potentially powerful tool in reinforcement learning applications. Shaping often fails to function effectively because of a lack of understanding about its effects when applied in reinforcement learning settings and the use of inadequate algorithms in its implementation. Because of these difficulties current shaping techniques require some form of manual intervention. We examine some of the principles involved in shaping and present a new algorithm for automatic transferral of knowledge which uses Q-values established in a previous task to guide exploration in the learning of a new task. This algorithm is applied to two different but related robot navigation tasks.
Paper: ps, pdf.
Dr. J. Bart Czirr, David B. Merrill, David Buehler, Thomas K. McKnight, James L. Carroll, Thomas Abbott, Eva Wilcox, " Capture-gated Neutron Spectrometry," in Nuclear Instruments and Methods A, June 2000.
Abstract: The applications of a new inorganic scintillator, lithium gadolinium borate, to neutron dosimetry and spectroscopy, are described. A dosimeter using this material registers, in separate energy bins, thermal, epithermal and MeV neutrons. A spectrometer for MeV neutrons has a calculated energy resolution of 10% FWHM.
Keywords: Neutron detector; Spectrometry; Organic scintillator
Abstract: Enos's wrestle "Before God" has many similarities to the Jacob story in the Old Testament. Better understanding the themes and morals of the Jacob story can aid our understanding of the implications and message of the book of Enos. From this comparison we learn that the greatest blessings that God has for us are often only attained after intense spiritual effort on our part.
Draft available upon request.
James L. Carroll, "An Interpreters History of the Israelite Three Room Temple Design," in The Ninth Annual BYU Religious Education Student Symposium, 2007.
Abstract: In 1842, Joseph Smith revealed the endowment which connected the temple tradition with the creation, the Garden of Eden, and the ritual activities of Adam and his descendents immediately after the fall. Ancient Israel under the Law of Moses also built several temples. These temples were not primarily used for endowments as we understand them today, but were part of an Aaronic order designed to be preparatory in nature. These temples each contained three main rooms or divisions. We survey several of the most important LDS and non-LDS theories that attempt to explain the meaning of these rooms and show that even most non-LDS scholars believe that the Israelite temple design was connected with the creation, the Garden of Eden, and the fall of man. Thus, although these temples were not identical to the modern temple pattern revealed through the Endowment, they were designed to teach many of the same fundamental gospel principles. The majority of these scholarly opinions were given many years after Joseph revealed the Endowment, illustrating the inspiration of the prophet.
Paper: pdf, Presentation slides: ppt.
James L. Carroll, "Egyptian Craft Guild Initiations," in Studia Antiqua, The Journal of the Student Society for Ancient Studies, 2007, vol. 5:1 p. 17-44.
Abstract: Initiation seems to have played an important role in Egyptian religion from the beginning of recorded history. Initiations are rites whereby the initiate is symbolically moved from one state of being into another or from one part of the temple into another, the passage involves various trials or tests of knowledge, the rites often deal with death and resurrection, various oaths are taken either of an ethical or of a sacramental nature, and the ceremony itself is usually secret. The initiation paradigm can be seen in the Egyptian funerary literature, the Daily Temple Liturgy, the initiations of the Egyptian priesthood, and the later Isis mystery cult initiations. All these ritual elements can also be seen in modern craft guild initiations, however it is unclear how early this paradigm became part of craft guild initiations. Although it can be shown that craft guilds existed in ancient Egypt from the earliest times, little direct evidence of their nature has remained. However, several elements from the earlier Egyptian initiations show evidence of having been influenced by guild initiations. This indicates that the guild traditions may have adopted the initiation paradigm at a very early stage. If this is the case, then it would have significant ramifications for the origins of modern guild initiations, and would indicate that they are connected to ancient traditions of initiation into the afterlife, and to ancient temple traditions.
James L. Carroll, "The Second Coming of Christ as Covenant Renewal," in The Eighth Annual BYU Religious Education Student Symposium, 2006.
Abstract: The second coming of Christ, especially his appearance at Jerusalem, can be thought of as a covenant renewal ritual in which the people at Jerusalem rediscover the identity of Jehovah, their God, and recommit their lives to him. This covenant renewal ritual is similar in form to the ancient covenant patterns found throughout the ancient Near East and in the Old Testament accounts of Abraham and Joshua. Viewing the second coming in this manner can help us to understand several passages that would otherwise be difficult to interpret, especially the division of the Mount of Olives. In this context, the division of the Mount of Olives can be seen as a re-creation of the sacred landscape that was present at Shechem when the Israelites made their covenant with Jehovah before possessing the Land of Promise.
Draft available upon request.
James L. Carroll, "A Revised Temple Typology" in Hagion Temenos, 2nd Edition ed. Stephen Ricks, Provo Ut. BYU Press, 2005.
Abstract: John M. Lundquist's "Temple Typology" has been highly influential in the past several years. This "Revised Temple Typology" attempts to build upon what he has created by synthesizing several of Lundquist's publications and by adding several new elements to the typology. Further, the typology is reordered, and organized into three main categories, "the Temple Space," "The Temple Rites," and "The Tempe and Community." Short titles have also been added to the typology elements. It is hoped that these changes will improve the use of the typology in teaching situations.
James L. Carroll, "An Expanded View of the Israelite Scapegoat" in Selections From the Seventh Annual BYU Religious Education Student Symposium, 2005.
Abstract: There have been many divergent interpretations of the scapegoat in LDS and other Christian commentaries. Some see the scapegoat as a symbol of Christ, others as a symbol of Satan. In order to better understand the spiritual significance of the scapegoat, this paper analyzes several scriptural stories that are similar in form to the Day of Atonement, specifically stories where one person or group is put to death, and another person or group is cast out or released. It is then possible to understand the Lord's goat and the scapegoat in terms of physical and spiritual death respectively. Such an interpretation can explain how the scapegoat can represent the Savior, who took spiritual death upon himself for those who repent, while at the same time representing Satan, who was cast out of the Father's presence, thus suffering spiritual death for his unrepented sins.
James L. Carroll, "The Reconciliation of Adam and Israelite Temples," in Studia Antiqua, The Journal of the Student Society for Ancient Studies, Winter 2003.
Abstract: Modern researchers have shown that ancient temples were often associated with the creation, the Garden of Eden, and reconciliation. All three of these elements can be found in Genesis 1-3 if one assumes that Adam and Eve repented of their transgression in the Garden as many apocryphal elements attest. The methods of reconciliation that they record form a unifying principal for understanding the significance of the tripartite divisions found in Israelite temples which seem to have represented the heavenly throne of God, the Garden of Eden, and the fallen world where Adam and Eve worked out their reconciliation with God.
Abstract: A survey paper concerning the burning of Incense in its relation to the temple cult of Ancient Israel. We place the burning of incense in its contexts from the religions surrounding Israel, attempt to identify the ingredients mentioned in Ex. 30. We also survey various rabbinic commentaries on the methods for burning incense, and give an LDS interpretation for the meaning of this part of the ancient Israelite temple ritual.