Scalable Distributed Model Checking: Experiences, Lessons, and Expectations: (invited talk)

In recent years a lot of effort has been put into using parallel and distributed environments to solve the computational and space complexity bottlenecks in model checking systems.In this talk we will survey early efforts, trying to pinpoint the reasons for their limited success. We will then outline the evolution of a method that did succeed in making its way into industrial tools, its current status, and its roadmap given existing technological trends. Finally, we will describe the high-performance platforms on the horizon and their impact on the design of scalable distributed model checking software.     

Auditing black-box models for indirect influence

Data-trained predictive models see widespread use, but for the most part they are used as black boxes which output a prediction or score. It is therefore hard to acquire a deeper understanding of model behavior and in particular how different features influence the model prediction. This is important when interpreting the behavior of complex models or asserting that certain problematic attributes (such as race or gender) are not unduly influencing decisions. In this paper, we present a technique for auditing black-box models, which lets us study the extent to which existing models take advantage of particular features in the data set, without knowing how the models work. Our work focuses on the problem of indirect influence: how some features might indirectly influence outcomes via other, related features. As a result, we can find attribute influences even in cases where, upon further direct examination of the model, the attribute is not referred to by the model at all. Our approach does not require the black-box model to be retrained. This is important if, for example, the model is only accessible via an API, and contrasts our work with other methods that investigate feature influence such as feature selection. We present experimental evidence for the effectiveness of our procedure using a variety of publicly available data sets and models. We also validate our procedure using techniques from interpretable learning and feature selection, as well as against other black-box auditing procedures. To further demonstrate the effectiveness of this technique, we use it to audit a black-box recidivism prediction algorithm.     

Enabling teachers to explore grade patterns to identify individual needs and promote fairer student assessment

Exploring student test, homework, and other assessment scores is a challenge for most teachers, especially when attempting to identify cross-assessment weaknesses and produce final course grades. During the course, teachers need to identify subject weaknesses in order to help students who are struggling with a particular topic. This identification often needs to happen across multiple assessment data points and should be considered in comparison to the class’s progress as a whole. When determining grades, fairness to all is essential, but there are special needs for students who did poorly on one exam or had a steadily increasing grasp of the subject. We present eduViz, a visualization tool designed to help teachers explore and assign grades. Teachers can see the trajectory of student scores, the relationship of a particular student to the class, and use categories they have defined in order to filter their assessment information. Query response is immediate and all logical comparisons are possible. Teachers can easily compare their query to the class or per student average as well as view scores by raw point total or percentage. Additionally, eduViz provides a grade assignment interface which allows teachers to view sorted student scores in a scatterplot. This scatterplot is coupled with a unique partition slider which allows users to move color coordinated bands on the scatterplot to indicate grade ranges. As these grade ranges are set, a histogram is updated to show the number of students assigned to each grade range. These features give teachers new and powerful ways to explore and assign grades so that they can better understand student strengths and weaknesses and make the most of the time they have available. Interviews with 16 expert teachers indicate that eduViz is a success across fields, provides teachers with a useful tool to understand and help their classes, and encourages reflective practice.     

A work-efficient distributed algorithm for reachability analysis

This work presents a novel distributed symbolic algorithm for reachability analysis that can effectively exploit, as needed, a large number of machines working in parallel. The novelty of the algorithm is in its dynamic allocation and reallocation of processes to tasks and in its mechanism for recovery from local state explosion. As a result, the algorithm is work-efficient: it utilizes only those resources that are actually needed. In addition, its high adaptability makes it suitable for exploiting the resources of very large and heterogeneous distributed, nondedicated environments. Thus, it suitable for verifying very large systems. We implemented our algorithm in a tool called Division. Our experimental results show that the algorithm is indeed work-efficient. Although the goal of this research is to check larger models, the results also indicate that the algorithm can obtain high speedups, because communication overhead is very small.     

A Platform-Independent Distributed Runtime for Standard Multithreaded Java

JavaSplit is a portable runtime environment for distributed execution of standard multithreaded Java programs. It gains augmented computational power and increased memory capacity by distributing the threads and objects of an application among the available machines. Unlike previous works, which either forfeit Java portability by using a nonstandard Java Virtual Machine (JVM) or compromise transparency by requiring user intervention in making the application suitable for distributed execution, JavaSplit automatically executes standard multithreaded Java on any heterogenous collection of Java-enabled machines. Each machine carries out its part of the computation using nothing but its local standard (unmodified) JVM. Neither the programmer nor the person submitting the program for execution needs to be aware of JavaSplit or its distributed nature. We evaluate the efficiency of JavaSplit on several combinations of operating systems, JVM implementations, and communication hardware.     

Mechanical consequences of cellular force generation

The discovery that adherent tissue cells actively sustain internal tension and exert mechanical forces on their surroundings has opened new vistas in the field of cell and tissue mechanics. Cellular forces, generated by acto-myosin contractility, play a central role in numerous aspects of cell behavior and function. Apart from the various specific functions that cells perform by applying forces (e.g., wound healing, remodeling of the extracellular matrix and muscle contraction), cells also apply stresses as a generic means for sensing and responding to the mechanical nature of their environment. In addition, the internal tension plays a role in actively controlling the elastic moduli and shape stability of the cell. In this review, we survey recent theoretical and experimental studies of the physical consequences of cell mechanical activity including its role in cell morphology, adhesion strength, stress-fiber polarization, and the elastic properties of cells. We also discuss the role of cell mechanics in orienting cellular assemblies and in the response of cells to external loads.     

The quest for runware: on compositional, executable and intuitive models

We believe that future models of complex software and systems will combine the crucial traits of intuitiveness, compositionality, and executability. The importance of each of these to modeling is already well recognized, but our vision suggests a far more powerful synergy between them. First, models will be aligned with cognitive processes used by humans to think about system behavior and will be understood, and perhaps creatable, by almost anyone. Second, one will be able to build models incrementally, adding to, refining or sculpting away already-specified behaviors without changing most existing parts of the model. Third, there will be powerful ways to execute such intuitive and compositional models, in whole or in part, at any stage of the development. The presence of these three traits in a single artifact will blur the boundaries between natural-language requirements, formal models, and actual software, bringing in its wake a major advance in the way systems are built, and in their cost and quality. We propose the term runware to refer to this kind of higher level artifact.     

Ni–Fe Catalysts Based on Perovskite-type Oxides for Dry Reforming of Methane to Syngas

LaNi_(1−x)Fe_xO₃ (x=0, 0.2, 0.4 and 0.7) perovskite-type catalysts were modified by the partial substitution of nickel by iron, aiming to increase the stability and resistance to carbon deposition during the methane dry reforming reaction. The results showed that a suitable combination of precipitation and calcination steps could result in oxides with the desired structure and with improved properties from the point of view of heterogeneous catalysis. The partial substitution of Ni by Fe in the perovskite structure resulted in decreasing rates of conversion of both reactants. However, the stability of the catalyst during the reaction was highly increased. These substituted catalysts were shown to be stable and the LaNi_0.8Fe_0.2O₃ catalyst, calcined at 800 °C for 5 h, was the most active in the reaction conditions.     

Structural Characterization of the Drug Translocation Path of MRP1/ABCC1

The rapid clearance of drugs from human cells is carried out by MRP1 and other proteins of the ABC transporter superfamily. Selective mutations carried out by DeGorter indicated that replacement of Y324 by phenylalanine (but not by tryptophan or alanine) enhances the capacity of the protein to extrude various drugs. In this study we investigate the effect of mutation on the structure of the isolated transmembrane domain of MRP1 through molecular dynamics simulations of the protein embedded in a POPC membrane. The simulations reveal a persistent tendency of the translocation path to experience a partial constriction, losing ∼50 % of the water inside the conducting path. While the Wt, Y324W and Y324A transporters all experienced the same constriction, the Y324F transporter, the one having a higher clearance rate than the Wt, retains a fully open configuration. The structure of the Y324F mutant reveals an alternate set of stabilizing interactions that force a kink in transmembrane helix 6, which keeps the protein in a fully open outward-facing configuration thus providing a molecular-level account for the higher activity of the mutant. The ability of the simulations to corroborate the experimental observations implies that the homology model of MRP1 is a proper representation of the internal interactions between the residues in the protein, and can be used as a reliable model for studying the human multidrug resistance function of the MRP1 protein.     

High‐order global–regional model interaction: Extension of Carpenter's scheme

A two‐dimensional global–regional model interaction problem for linear time‐dependent waves is considered. The setup, which is sometimes called ‘one‐way nesting,’ arises in numerical weather prediction as well as in other fields concerning waves in very large domains. It involves the interaction of a coarse global model and a fine limited‐area (regional) model through an ‘open boundary.’ The multiscale nature of this general problem is described. The Carpenter scheme, originally proposed in a note by K. M. Carpenter in 1982 for this type of problem, is then revisited, in the context of the linear scalar wave equation. The original Carpenter scheme is based on the Sommerfeld radiation operator and thus is associated with low‐order accuracy. By replacing the Sommerfeld operator with the high‐order Hagstrom–Warburton absorbing operator, a modified Carpenter open‐boundary condition emerges, which possesses high‐order accuracy. This boundary condition is incorporated in a computational scheme, which uses finite element discretization in space and Newmark time‐stepping. Error analysis and numerical tests for wave guides demonstrate the performance of the modified scheme for combinations of incoming and outgoing waves. Copyright © 2008 John Wiley & Sons, Ltd.