DIVISION OF ENGINEERING AND APPLIED SCIENCES
HARVARD UNIVERSITY
CS 266: Biologically-inspired Distributed and Multi-agent SystemsFall 2004 |
| Class Research Projects |
Here is a list of projects that were explored by students in this class. Also, here is a list of suggested projects that was handed out early in the term.
Theoretical Approaches
Title: How Does the Behavior of Small Groups Constrain That of Larger Groups?
Authors: Dan Iancu and Daniel Yamins
Abstract: For any locally-defined one-dimensional algorithm f, it is possible to define the "group behavior" that f generates among groups of any given size. Since for any given fixed local neighborhood size r, there are finitely many local algorithms which operate on neighborhoods of that size, the number of possible sets of group behaviors for such algorithms is itself a finite number. Hence there is a relationship between the behavior patterns of small-sized groups and those of large-sized groups, which we will explore through theory and experimental simulations.
Title: Centralized versus Decentralized Team Formation: A comparison
Authors: Adam Juda and Ben Lubin
Abstract: When trying to form teams, centralized solutions are not aware of private information but can find optimal solutions for what they do know. Decentralized solutions can more readily take advantage of private information, but optimal solutions may not be found. We will investigate under what circumstances each methodology performs better.
Title: Ant-Based Computing
Authors: Loizos Michael
Abstract: In this project we investigate the theoretical and practical feasibility of building an ant-based computer. The main challenge is to use the simple, local, and inherently probabilistic protocols followed by ants, in order to solve a problem (namely, circuit evaluation), which is by its nature complex and exhibits a robust deterministic global behavior. This project's goal is to propose a practically plausible way of implementing an ant-based computer, and to subsequently establish some theoretical guarantees on the expected behavior of such an implementation.
Biologically-inspired Systems Design
Title: Swarm-inspired, Agent-based Dynamic Load-Balancing of Communicating Tasks in a Distributed Computing System
Authors: Prashanth Bungale and Alexander Allain
Abstract: We consider the problem domain of dynamic load balancing of computation tasks in a distributed system, where the tasks can include dynamically varying sets of related, communicating (interacting) tasks, among which the communication patterns cannot be determined a priori. Our goal is to perform load balancing such that the communication overhead is minimized. We propose to investigate the application of agent-based swarm-intelligence techniques towards building a solution for this problem. Considering the success of these techniques in the biological and social worlds, we would expect this research direction to strive towards including self-organization, adaptation and resilience as fundamental properties of the system.
Title: Self Aggregation in Swarm Robots
Authors: Jimming Cheng, Winston Cheng
Abstract: Our project addresses programmable self-aggregation of mobile swarm robots with limited communication and proximity sensing abilities. Major challenges are 1) negotiating a stable coordinate system with rough triangulation 2) recruitment 3) blocking. We hope to achieve success by putting the robots into a cohesive swarm with gas expansion model behaviors.
Title: Smart Intersections
Authors: Nick Elprin
Abstract: The project explores the use of local rules as a means of dynamically changing traffic signal behavior. The idea is that traffic signals can tune their behavior based on the directional throughput at neighboring intersections to better accommodate changing traffic patterns.
Title: Target Detection in Wireless Sensor Networks using an Artificial Immune System
Authors: Wilfred Yeung, DJ Lee, [Ya'ir Aizenman]
Abstract: We will attempt a novel approach at solving the problem of target detection using wireless networks. This problem boils down to distinguishing signal from noise, and to this end we plan to apply principles gleaned from the study of Artificial Immune Systems. The natural immune system solves a problem of discriminating self from non-self using a vastly distributed network of sensors; in this sense, target detection should be a natural and effective application of immune system ideas.
Title: Adaptive Congestion Avoidance in Sensor Networks Inspired by Quorum Sensing
Authors: Bor-rong Chen
Abstract: Quorum sensing is a mechanism for communicating "density" of entities in certain area. Bacteria use quorum sensing to trigger actions which can succeed only when certain density is reached. Congestion in sensor networks can be viewed as high density of packets gathered in some area of the network. This project aims to use quorum sensing model to signal congestion avoidance mechanism in sensor networks.
Title:A Low-Power Swarm-Inspired Routing Protocol
Authors: Geoffrey Mainland
Abstract: We are implementing a low-power routing protocol inspired by insect behavior, in particular firefly synchronization. Decentralized synchronization and clusterhead-formation algorithms are used to perform very low power packer routing, and the trade-off between latency and energy consumption can be tuned by the programmer.
Title: Biologically Inspired Time Synchronization for Wireless Sensor Networks
Authors: Uri Brawn, Geetika Tewari, Geoffrey Werner-Allen
Abstract: We propose to design and evaluate Firefly Inspired Time Synchronization (FITS), a protocol for distributed time synchronization based on the pulse-coupled integrate and fire model embedded in biological swarms. The goals behind our design include low communication bandwidth utilization, scalability in multi-hop networks, and robustness against topology changes and node failures. We intend to evaluate the performance of FITS on the MicaZ platform, using both TOSSIM, the TinyOS mote simulator, amd Motelab, a 38 node sensor cluster deployed in Maxwell Dworkin.
Title: Dynamic Robustness in Unreliable Wireless Sensor Networks
Authors: Jonathan McPhie [Ivin Baker]
Abstract: The goal of this project is to examine methods by which an unreliable network can compensate for a failed node beyond simply assuring that the other nodes are still operational. Rather, we will experiment with increasing sampling rates in nearby nodes as a means of functional compensation. Title: Handbook of Gradients
Authors: Philip Hendrix
Abstract: Many papers have been written about ways to coordinate multiple agents, biological or mechanical, with limited computing and communication resources in order to perform complex tasks. I hope to gain an understanding of many of these methods, point out the merits and drawbacks of each, and relate them with one another. In particular, I plan to focus on models that use various types of gradients in order to implement structure of individual agents.
Synthetic Biology and Models of Biology
Title: Synthetic Biology via directed evolution
Authors: Chris Clearfield, Illya Bomash
Abstract: We are investigating the construction of biological computation through a process that combines engineering design ala BioBricks with evolution, by allowing random promoter recombination to assemble random networks within a cell that monitors or reports its own fitness. We plan to do theoretical work on the likelihood of evolving into a working state. We want to try to apply this technique to the construction a novel network, like a cellular integrator.
Title: Cell Topology in Epithelial Tissue
Authors: Ankit Patel
Abstract: In the fruitfly wing, cells divide over many generations to rapidly grow the size of the wing tissue. We will investigate models of cleavage plane choice in these cells to understand what implications different models have for high level properties of the tissue. The goal is to connect hypotheses of local cell behavior with global tissue properties.
Title: From Sequence and Expression to Network in the development of C. Elegans
Authors: Joseph Barillari
Abstract: The worm C. Elegans is unique among research organisms in that scientists have determined the exact course by which each of the nematode's 959 body cells grows from the its egg, making it an ideal model for studying genetic regulation of development. Tavazoie and Beer (2004) suggest that genetic regulatory networks of unicellular organisms can be mapped by combining sequence data with microarray expression data, analysis that they extend as a proof-of-concept to C. elegans. The aim of this project is to combine Tavazoie's approach with the well-characterized spatial information on C. elegans development, along with some algorithmic enhancements, to produce a more accurate C. elegans network---which will in turn give rise to readily generalizable software for multicellular organism network discovery.
Title: Systematic cell programming for pattern formation
Authors: Chris Bockman, Lan Zhang
Abstract: The goal of the project is to construct a system to achieve a given global pattern with a colony of identically programmed cells. Given a global pattern, the system aims to express the serial process derived from the Origami Shape Language (OSL) with an intermediate rule-based asynchronous language similar to the Microbial Colony Language (MCL), and eventually map it to a cellular circuit composed of genetic logic gates.
| CS266, fall 2004 |