Publications

The goal of this project was to develop a risk analysis meta tool--a tool that enables security analysts both to combine and analyze data from multiple other risk assessment tools on demand. Our approach was based on the innovative self-assembling software technology under development by the project team. This technology provides a mechanism for the user to specify his intentions at a very high level (e.g., equations or English-like text), and then the code self-assembles itself, taking care of the implementation details. The first version of the meta tool focused specifically in importing and analyzing data from Joint Conflict and Tactical Simulation (JCATS) force-on-force simulation. We discuss the problem, our approach, technical risk, and accomplishments on this project, and outline next steps to be addressed with follow-on funding.

Biochemical reactions taking place in living systems that map different inputs to specific outputs are intuitively recognized as performing information processing. Conventional wisdom distinguishes such proteins, whose primary function is to transfer and process information, from proteins that perform the vast majority of the construction, maintenance, and actuation tasks of the cell (assembling and disassembling macromolecular structures, producing movement, and synthesizing and degrading molecules). In this paper, we examine the computing capabilities of biological processes in the context of the formal model of computing known as the random access machine (RAM) [Dewdney AK (1993) The New Turing Omnibus. Computer Science Press, New York], which is equivalent to a Turing machine [Minsky ML (1967) Computation: Finite and Infinite Machines. Prentice-Hall, Englewood Cliffs, NJ]. When viewed from the RAM perspective, we observe that many of these dynamic self-assembly processes - synthesis, degradation, assembly, movement - do carry out computational operations. We also show that the same computing model is applicable at other hierarchical levels of biological systems (e.g., cellular or organism networks as well as molecular networks). We present stochastic simulations of idealized protein networks designed explicitly to carry out a numeric calculation. We explore the reliability of such computations and discuss error-correction strategies (algorithms) employed by living systems. Finally, we discuss some real examples of dynamic self-assembly processes that occur in living systems, and describe the RAM computer programs they implement. Thus, by viewing the processes of living systems from the RAM perspective, a far greater fraction of these processes can be understood as computing than has been previously recognized. © Springer Science+Business Media, Inc. 2006.

Intracellular molecular machines synthesize molecules, tear apart others, transport materials, transform energy into different forms, and carry out a host of other coordinated processes. Many molecular processes have been shown to work outside of cells, and the idea of harnessing these molecular machines to build nanostructures is attractive. Two examples are microtubules and motor proteins, which aid cell movement, help determine cell shape and internal structure, and transport vesicles and organelles within the cell. These molecular machines work in a stochastic, noisy fashion: microtubules switch randomly between growing and shrinking in a process known as dynamic instability; motor protein movement along microtubules is randomly interrupted by the motor proteins falling off. A common strategy in attempting to gain control over these highly dynamic, stochastic processes is to eliminate some processes (e.g., work with stabilized microtubules) in order to focus on others (interaction of microtubules with motor proteins). In this paper, we illustrate a different strategy for building nanostructures, which, rather than attempting to control or eliminate some dynamic processes, uses them to advantage in building nanostructures. Specifically, using stochastic agent-based simulations, we show how the natural dynamic instability of microtubules can be harnessed in building nanostructures, and discuss strategies for ensuring that 'unreliable' stochastic processes yield a robust outcome.

We present two ways in which dynamic self-assembly can be used to perform computation, via stochastic protein networks and self-assembling software. We describe our protein-emulating agent-based simulation infrastructure, which is used for both types of computations, and the few agent properties sufficient for dynamic self-assembly. Examples of protein-network-based computation and self-assembling software are presented. We describe some novel capabilities that are enabled by the inherently dynamic nature of the self-assembling executable code. © Springer-Verlag 2004.

We describe stochastic agent-based simulations of protein-emulating agents to perform computation via dynamic self-assembly. The binding and actuation properties of the types of agents required to construct a RAM machine (equivalent to a Turing machine) are described. We present an example computation and describe the molecular biology and non-equilibrium statistical mechanics, and information science properties of this system.

We present initial results on achieving synthesis of complex software systems via a biophysics-emulating, dynamic self-assembly scheme. This approach offers novel constructs for constructing large hierarchical software systems and reusing parts of them. Sets of software building blocks actively participate in the construction and subsequent modification of the larger-scale programs of which they are a part. The building blocks interact through a software analog of selective protein-protein bonding. Self-assembly generates hierarchical modules (including both data and executables); creates software execution pathways; and concurrently executes code via the formation and release of activity triggering bonds. Hierarchical structuring is enabled through encapsulants that isolate populations of building block binding sites. The encapsulated populations act as larger-scale building blocks for the next hierarchy level. Encapsulant populations are dynamic, as their contents can move in and out. Such movement changes the populations of interacting sites and also modifies the software execution. ''External overrides'', analogous to protein phosphorylation, temporarily switch off undesired subsets of behaviors (code execution, data access/modification) of other structures. This provides a novel abstraction mechanism for code reuse. We present an implemented example of dynamic self-assembly and present several alternative strategies for specifying goals and guiding the self-assembly process.

Future generation automated human biometric identification and verification will require multiple features/sensors together with internal and external information sources to achieve high performance, accuracy, and reliability in uncontrolled environments. The primary objective of the proposed research is to develop a theoretical and practical basis for identifying and verifying people using standoff biometric features that can be obtained with minimal inconvenience during the verification process. The basic problem involves selecting sensors and discovering features that provide sufficient information to reliably verify a person`s identity under the uncertainties caused by measurement errors and tactics of uncooperative subjects. A system was developed for discovering hand, face, ear, and voice features and fusing them to verify the identity of people. The system obtains its robustness and reliability by fusing many coarse and easily measured features into a near minimal probability of error decision algorithm.

A split Hopkinson bar technique has been developed to evaluate the performance of accelerometers that measure large amplitude pulses. An evaluation of this technique has been conducted in the Mechanical Shock Laboratory at Sandia National Laboratories (SNL) to determine its use in the practical calibration of accelerometers. This evaluation consisted of three tasks. First, the quartz crystal was evaluated in a split Hopkinson bar configuration to evaluate the quartz gage`s sensitivity and frequency response at force levels of 18,000, 35,000 and 53,000 N at ambient temperature, {minus}48 C and +74 C. Secondly, the fly away technique was evaluated at shock amplitudes of 50,000, 100,000, 150,000 and 200,000 G (1 G = 9.81 m/s{sup 2}) at ambient temperature, {minus}48 C and +74 C. Lastly, the technique was performed using a NIST calibrated reference accelerometer. Comparisons of accelerations calculated from the quartz gage data and the measured acceleration data have shown very good agreement. Based on this evaluation, the authors expect this split Hopkinson fly away technique to be certified by the SNL Primary Standards Laboratory.