One-on-one human tutoring is a very effective method of instruction and researchers are attempting to achieve similar levels of success with intelligent tutoring systems (ITSs), which replace the human tutor with a computer tutor. While ITSs succeed at tutoring quantitative problem solving about as well as human tutors, they do not show the same success with conceptual topics. In the work discussed here, we focus on conceptual kinematics physics, where ITSs show learning gains over simple methods (e.g. reading a text), but the students' learning gains are not as good as with a human tutor.

ITS researchers have explored methods of improving ITS effectiveness. One dimension considered is the format used to represent the concepts. Some use a natural language representation (e.g. dialogues or text). Others use visual representations, either graphics or illustrations, to convey concepts. Sometimes the visuals are presented with a natural language representation while other times the visuals are provided with almost no text. Still others have found switching between visual representations during tutoring to be better than using just one visual representation.

Another dimension considered is how adaptive to make the ITS. ITS adaptability is the customization of content and/or delivery to the learner. Generally, more adaptive systems are better than less adaptive systems, but the difficulty is figuring out what should be adaptive and how to adapt. Within the context of the representation discussion above, adapting visual representations to individuals is still unexplored. My research looks at determining whether illustrations or graphs should be shown to students during conceptual physics dialogue-based tutoring. Prior findings suggest that different students learn better from different visual representations and that these differences may change during tutoring. The work presented here will focus on how the tutor should pick which kind of visual to show the student to better-encourage learning.

Access control schemes come in all shapes and sizes, which makes choosing the right one for a particular application a challenge. Yet today's techniques for comparing access control schemes completely ignore the setting in which the scheme is to be deployed. In this work, we present a formal framework for comparing access control schemes with respect to a particular application. The analyst's main task is to evaluate an access control scheme in terms of how well it implements a given access control workload (a formalism that we introduce to represent an application's access control needs). One implementation is better than another if it has stronger security guarantees, and in this paper we introduce several such guarantees: correctness, homomorphism, AC-preservation, safety, administration-preservation, and compatibility. The scheme that admits the implementation with the strongest guarantees is deemed the best fit for the application. We demonstrate the use of our framework by evaluating two workloads on ten different access control schemes.

The U.S. Department of Energy has identified resilience and energy consumption as key challenges for future extreme-scale systems. All checkpoint/restart methods require I/O to local or remote storage. Efforts are under way to minimize the amount of data movement and increase scalability. Nevertheless, the energy consumed by fault resilience methods will increase with system size. It is therefore important to understand the performance overhead in conjunction with the energy consumption of each fault resilience method. In this paper we explore throttling CPU power consumption during I/O intensive checkpoint operations of real applications. We find that 10% total energy savings are possible with little impact on application time to solution.

In this work we consider the problem of learning a positive semidefinite kernel matrix from relative comparisons of the form: "object A is more similar to object B than it is to C", where comparisons are given by humans. Existing solutions to this problem assume many comparisons are provided to learn a high quality kernel. However, this can be considered unrealistic for many real-world tasks since relative assessments require human input, which is often costly or difficult to obtain. Because of this, only a limited number of these comparisons may be provided. In this work, we explore methods for aiding the process of learning a kernel with the help of auxiliary kernels built from more easily extractable information regarding the relationships among objects. We propose a new kernel learning approach in which the target kernel is defined as a conic combination of auxiliary kernels and a kernel whose elements are learned directly. We formulate a convex optimization to solve for this target kernel that adds only minor overhead to methods that use no auxiliary information. Empirical results show that in the presence of few training relative comparisons, our method can learn kernels that generalize to more out-of-sample comparisons than methods that do not utilize auxiliary information, as well as similar methods that learn metrics over objects.