David Ackley
Associate Professor
Dept. of Computer Science
University of New Mexico

Abstract:
The growth of the serial digital computer---making a CPU faster with a memory vaster---has now stalled, even as our ability to manufacture more, denser, and cheaper chips continues to expand. Although the efficiency of parallel architectures for general-purpose computation is often questioned, parallel hardware also offers the possibility of improving computational robustness. The future, somehow, will be massively parallel and distributed, components will come and go while computations continue, and---though our ability to pose ever larger computations will remain prodigious---it will often be as important to spend effort on robustness as on efficiency.

In this talk I will suggest that the traditional roles assigned to computer hardware and computer software---so revolutionary in the middle of the last century---are increasingly counterproductive and need to be renegotiated. Although hints of that process can already be seen, I will argue that for computer science as well as society at large, we would be better off recognizing the sea change that is now upon us.

A tabletop computational grid involving dozens of processors will be assembled and demonstrated.

Bio:
David Ackley received his Ph.D. in Computer Science from Carnegie Mellon, and was a member of the technical staff at Bellcore before joining the faculty at UNM. Research interests include artificial life and the connections between computation and biology, distributed and adaptive systems, and making things that do things by themselves.

Lance R. Williams
Associate Professor
Dept. of Computer Science
University of New Mexico

Abstract:
The thing which distinguishes animate from inanimate matter is that animate matter uses information processing to work against entropy and produce a state of increased order in the physical world. Even the simplest single cell organisms are able to translate self-descriptions stored on DNA molecules into copies of themselves. We believe that this remarkable feat, which more than any other defines life itself, is accomplished by means of a process which is intimately related to a topic at the heart of computer science, namely, compilation of programming languages.

Self-replicating systems based on von Neumann's universal constructor lack transparency and for this reason have had virtually no impact in biology. We believe that fundamental computational principles underlying their operation, e.g., self-reflection, are obscured by the complexity of low-level implementations, e.g., cellular automata. We wish to bridge the gap between principles and implementation by means of transparent automatic processes for translating abstract descriptions of self-reproducing machines into physical implementations. The abstract descriptions are expressions in high-level functional programming languages which are compiled into bytecode quines, i.e., virtual-machine fixed-points. Implementation of the bytecodes as lightweight processes, or actors, which accomplish evaluation by means of continuation passing, yields a self-replicating distributed virtual machine--or artificial cell.

Bio:
Lance R. Williams received the BS degree in computer science from the Pennsylvania State University and the MS and PhD degrees in computer science from the University of Massachusetts. Prior to joining UNM, he was a post-doctoral scientist at NEC Research Institute. His researches include computer vision and graphics, neural computation and digital image processing.

Related links: National Science Foundation; National Cancer Insititute; Santa Fe Institute; Qforma; Professor Stephanie Forrest; Professor Shuang Luan

Related links: Computer Science Graduate Student Association;

Sean M. Brennan

Committee Chair:
Barney Maccabe, Director of the Computer Science and Mathematics Division at Oak Ridge National Laboratory
Committee Members:
Wenbo He, Assistant Professor of Computer Science at UNM
Sudharman Jayaweera, Assistant Professor of Electrical and Computer Engineering at UNM
Michael Cai, Los Alamos National Laboratory

Abstract:
The SENSIX software framework uniquely integrates constraint-dominated wireless sensor networks with the flexibility of object-oriented programming models, without violating the principles of either. Though these two computing paradigms are opposite in many ways, SENSIX bridges them to yield a dynamic middleware abstraction unifying low-level resource-aware task reconfiguration and high-level object recomposition. Through the layered approach of SENSIX, the software developer creates a domain-specific sensing architecture by merely defining a customized task specification and utilizing object inheritance. In addition, SENSIX performs better at large scales (on the order of 1000 nodes or more) than other sensor network middleware which do not include such unified facilities for vertical integration.

Guanyu Wang

Committee Chair:
Joe Kniss, Assistant Professor of Computer Science at UNM
Committee Members:
Pradeep Sen, Assistant Professor of Electrical and Computer Engineering at UNM
Lance Williams, Associate Professor of Computer Science at UNM

Abstract:
I will propose a simple and robust method for image and volume data segmentation based on manifold distance metrics. In this approach, pixels in an image are not considered as points with color values arranged in a grid. In this way, a new data set is built by a transform function from one traditional 2D image or 3D volume to a manifold in higher dimension feature space. Multiple possible feature spaces like position, gradient and probabilistic measures are studied and experimented. Graph algorithm and probabilistic classification are involved. Both time and space complexity of this algorithm is O(N). With appropriate choice of feature vector, this method could produce similar qualitative and quantitative results to other algorithms like Level Sets and Random Walks. Analysis of sensitivity to parameters is presented. Comparison between segmentation results and ground-truth images is also provided to validate of the robustness of this method.

Manjunath Gorentla Venkata

Committee Chair:
Patrick Bridges, Associate Professor of Computer Science at UNM
Committee Members:
Dorian Arnold, Assistant Professor of Computer Science at UNM
Rolf Riesen, Principal Member of the Technical Staff, Scalable Systems Software, Sandia National Laboratories
Nasir Ghani, Associate Professor of Electrical and Computer Engineering at UNM

Abstract:
Modern HPC applications, for example adaptive mesh refinement and multi-physics codes, have dynamic communication characteristics which result in poor performance on current MPI implementations. The degraded application performance can be attributed to a mismatch between changing application requirements and static communication library functionality. To improve the performance of these applications, MPI libraries should change their protocol functionality in response to changing application requirements, and tailor their functionality to take advantage of hardware capabilities.

This dissertation describes PRO-MPI, a framework for constructing profile-driven reconfigurable MPI libraries; these libraries use past application characteristics (profiles) to dynamically change their functionality to match the changing application requirements. The framework addresses the challenges of designing and implementing the reconfigurable MPI libraries, which include collecting and reasoning about application characteristics to drive the protocol reconfiguration and defining abstractions required for implementing these reconfigurations. Two prototype reconfigurable MPI implementations based on the framework -- Open PRO-MPI and Cactus PRO-MPI -- are also presented to demonstrate the utility of the framework.

To demonstrate the effectiveness of reconfigurable MPI libraries, this dissertation presents experimental results to show the impact of using these libraries on the application performance. The results show that PRO-MPI improves the performance of important HPC applications and benchmarks. They also show that the application performance improves significantly when exact profiles are available, and the performance improves considerably when only approximate profiles are available.

Related links: Colloquium description