LSCITS home University of Bristol Department of Computer Science John Cartlidge - home

  [view publications] [view bibtex] [view Google Scholar]

S. Stotter, J. Cartlidge, & D. Cliff, (In press), “Behavioural investigations of financial trading agents using Exchange Portal (ExPo),” To appear in Springer's LNCS Transactions on Computational Collective Intelligence (TCCI), special edition selected papers from ICAART-2013. [Bibtex] [Journal Link] (24 pages, approx. 7,000 words).

Abstract—Some major financial markets are currently reporting that 50% or more of all transactions are now executed by automated trading systems (ATS). To understand the impact of ATS proliferation on the global financial markets, academic studies often use standard reference strategies, such as AA and ZIP, to model the behaviour of real trading systems. Disturbingly, we show that the reference algorithms presented in the literature are ambiguous, thus reducing the validity of strict comparative studies. As a remedy, we suggest disambiguated standard implementations of AA and ZIP. Using Exchange Portal (ExPo), an open-source financial exchange simulation platform designed for real-time behavioural economic experiments involving human traders and/or trader-agents, we study the effects of disambiguating AA and ZIP, before introducing a novel method of assignment-adaptation (ASAD). Experiments show that introducing ASAD agents into a market with shocks can produce counter-intuitive market dynamics.

J. Cartlidge & P. Clamp, (2014), “Correcting a financial brokerage model for cloud computing: closing the window of opportunity for commercialisation,” Journal of Cloud Computing: Advances, Systems and Applications, vol. 3, no. 2, Apr 2014. [Open Access] [Bibtex] [Journal Link] doi:10.1186/2192-113X-3-2 (38 pages, approx. 8,000 words).

Abstract— In April 2012, Rogers and Cliff (R&C) demonstrated a theoretical financial brokerage model for cloud computing that is profitable for the broker, offers reduced costs for cloud users, and generates more predictable demand flow for cloud providers. Relatively cheap, long-term reserved instances (RIs) are bulk-purchased by the broker, and then re-packaged and re-sold as monthly options contracts at a price lower than a user can purchase “on-demand” from the provider. Thus, the broker risks exposure on purchase for margin on sales. R&C's result has generated significant interest in the cloud computing community and is currently the fifth most accessed research paper of all time in the Journal of Cloud Computing: Advances, Systems and Applications.

     Here, we perform an independent replication of R&C's brokerage model using CReST, a discrete event simulation platform for cloud computing developed at the University of Bristol. We identify two implementation problems in R&C's original work: firstly, the broker buys fewer RIs than the model suggests; and secondly, the broker is undercharged for RIs used. We correct R&C's results accordingly: while broker's profits are not as high as R&C suggest, the model still supports the theoretical possibility of a profitable brokerage.

     However, aggressive competition between cloud providers has reduced the cost of cloud services to users and led to the introduction of new secondary markets where users can buy and sell RIs between themselves. This has squeezed the opportunity for an intermediary brokerage. By recalibrating R&C's model to fit current market conditions, we conclude that the commercial viability of R&C's brokerage model has been eradicated. The window of opportunity has now closed.

J. Cartlidge, (2014), “Trading experiments using financial agents in a simulated cloud computing commodity market,” in Proc. 6th Int. Conf. Agents and Artif. Intelligence, Vol. 2 - Agents (ICAART-2014). B. Duval, J. van den Herik, S. Loiseau & J. Filipe, Eds. Angers, France: SciTePress, Mar. 2014, pp. 311-317. [Camera-Ready Copy Available Online] [Bibtex] [Conference Link] doi:10.5220/0004925303110317 (7 pages, approx. 5000 words).

Abstract—In September 2012, Amazon, the leading Infrastructure as a Service (IaaS) provider, launched a secondary marketplace venue for users to buy and sell cloud resources between themselves—the Amazon EC2 Reserved Instance Marketplace (ARIM). ARIM is designed to encourage users to purchase more long-term reserved instances, thus generating more stable demand for the provider and additional revenue through commission on sales. In this paper, we model ARIM using a multi-agent simulation model populated with zero-intelligence plus (ZIP) financial trading agents. We demonstrate that ARIM offers a new opportunity for market makers (MMs) to profit from buying and selling resources, but suggest that this opportunity may be fleeting. We also demonstrate that altering the market mechanism from a retail market (where only sellers post offers; similar to ARIM) to a continuous double auction (where both buyers and sellers post offers) can result in higher sale prices and therefore higher commissions. Since IaaS is a multi-billion dollar industry and currently the fastest growing segment of the cloud computing market, we therefore suggest that Amazon may profit from altering the mechanism of ARIM to enable buyers to post bids.

P. Clamp & J. Cartlidge, (2013), “Pricing the cloud: An adaptive brokerage for cloud computing,” in Proc. 5th Int. Conf. Advances in System Simulation (SIMUL-2013). M. Bauer & P. Lorenz, Eds. Venice, Italy: IARIA XPS Press, Oct 2013, pp. 113-121. [Open Access] [Camera-Ready Copy Available Online] [Bibtex] [Conference Link] (9 pages, approx. 6,500 words).

Abstract—Using a multi-agent social simulation model to predict the behavior of cloud computing markets, Rogers & Cliff (R&C) demonstrated the existence of a profitable cloud brokerage capable of benefitting cloud providers and cloud consumers alike. Functionally similar to financial market brokers, the cloud broker matches provider supply with consumer demand. This is achieved through options, a type of derivatives contract that enables consumers to purchase the option, but not the obligation, of later purchasing the underlying asset—a cloud computing virtual machine instance—for an agreed fixed price. This model benefits all parties: experiencing more predictable demand, cloud providers can better optimize their workflow to minimize costs; cloud users access cheaper rates offered by brokers; and cloud brokers generate profit from charging fees. Here, we replicate and extend the simulation model of R&C using CReST—an opensource, discrete event, cloud data center simulation modeling platform developed at the University of Bristol. Sensitivity analysis reveals fragility in R&C's model. We address this by introducing a novel method of autonomous adaptive thresholding (AAT) that enables brokers to adapt to market conditions without requiring a priori domain knowledge. Simulation results demonstrate AAT's robustness, outperforming the fixed brokerage model of R&C under a variety of market conditions. We believe this could have practical significance in the real-world market for cloud computing.

G. Baxter & J. Cartlidge, (2013), “Flying by the seat of their pants: What can High Frequency Trading learn from aviation?,” in Proc. 3rd Int. Conf. Application and Theory of Automation in Command and Control Systems (ATACCS-2013). Naples, Italy: ACM, May 2013, pp. 64-73. [Camera-Ready Copy Available Online] [Bibtex] [Conference Link] (10 pages, approx. 8,500 words).

Abstract—As we build increasingly large scale systems (and systems of systems), the level of complexity is also rising. We still expect people to intervene when things go wrong, however, and to diagnose and fix the problems. Aviation has a history of developing systems with a very good safety record. Domains such as high frequency trading (HFT), however, have a much more chequered history. We note that there are several parallels that can be drawn between aviation and HFT. We highlight the ironies of automation that apply to HFT, before going on to identify several lessons that have been used to improve safety in aviation and show how they can be applied to increase the resilience of HFT in particular.

J. Cartlidge & D. Cliff, (2013), “Comparison of Cloud Middleware Protocols and Subscription Network Topologies using CReST, the Cloud Research Simulation Toolkit,” in Proc. 3rd Int. Conf. Cloud Computing and Services Science (CLOSER-2013). F. Desprez, D. Ferguson, E. Hadar, F. Leymann, M. Jarke & M. Helfert, Eds. Aachen, Germany: SciTePress, May 2013, pp. 58-68. [Camera-Ready Copy Available Online] [Bibtex] [Conference Link] [Presentation Slides] (11 pages, approx. 6,000 words).
Shortlisted for Best Paper Award.

Abstract—We introduce the Cloud Research Simulation Toolkit (CReST), a new cloud computing simulation tool designed to enable cloud providers to research and test their systems before release. We compare CReST with other known cloud simulation tools and demonstrate the utility of CReST by evaluating different distributed middleware protocols and associated subscription network topologies for robustness and reliability. Our results extend previous work and demonstrate that the published literature contains inaccuracies. CReST has been released as open-source under a Creative Commons license on SourceForge, with the intention that it can be used and extended by the cloud computing research community.

J. Cartlidge & D. Cliff, (2013), “Evidencing the “robot phase transition” in experimental human-algorithmic markets,” in Proc. 5th Int. Conf. Agents and Artif. Intelligence, Vol. 1 - Agents (ICAART-2013). J. Filipe & A. Fred, Eds. Barcelona, Spain: SciTePress, Feb. 2013, pp. 345-352. [Camera-Ready Copy Available Online] [Bibtex] [Conference Link] [Poster] (8 pages, approx. 5,500 words).

Abstract—Johnson, Zhao, Hunsader, Meng, Ravindar, Carran, and Tivnan (2012) recently suggested the existence of a phase transition in the dynamics of financial markets in which there is free interaction between human traders and algorithmic trading systems ("robots"). Above a particular time-threshold, humans and robots trade with one another; below the threshold all transactions are robot-to-robot. We refer to this abrupt system transition as the "robot phase transition". Here, we conduct controlled experiments where human traders interact with 'robot' trading agents in minimal models of electronic financial markets to see if correlates of the two regimes suggested by Johnson et al. (2012) occur in such laboratory conditions. Our results indicate that when trading robots act on a super-human timescale, the market starts to fragment, with statistically lower human-robot interactions than we would expect from a fully mixed market. We tentatively conclude that this is the first empirical evidence for the robot phase transition occurring under controlled experimental conditions.

S. Stotter, J. Cartlidge, & D. Cliff, (2013), “Exploring assignment-adaptive (ASAD) trading agents in financial market experiments,” in Proc. 5th Int. Conf. Agents and Artif. Intelligence, Vol. 1 - Agents (ICAART-2013). J. Filipe & A. Fred, Eds. Barcelona, Spain: SciTePress, Feb. 2013, pp. 77-88. [Camera-Ready Copy Available Online] [Bibtex] [Conference Link]
Winner of Best Student Paper Award.

Abstract—Automated trading systems in the global financial markets are increasingly being deployed to do jobs previously done by skilled human traders: very often a human trader in the markets simply cannot tell whether the counter-party to a trade is another human, or a machine. Clearly, automated trading systems can easily be considered as “intelligent” software agents. In this paper we report on experiments with software trader agents running the well-known “AA” and “ZIP” strategies, often used as reference benchmarks in previously published studies; here we suggest disambiguated standard implementations of these algorithms. Then, using Exchange Portal (ExPo), an open-source financial exchange simulation platform designed for real-time behavioural economic experiments involving human traders and/or trader-agents, we explore the impact of introducing a new method for assignment adaptation in ZIP. Results show that markets containing only assignment adaptive (ASAD) agents equilibrate more quickly after market shocks than markets containing only “standard” ZIP agents. However, perhaps counterintuitively, in mixed heterogeneous populations of ASAD agents and ZIP agents, ZIP agents outperform ASAD agents. Evidence suggests that the behaviour of ASAD agents act as a new signal in the market that ZIP agents then use to beneficially alter their own behaviour, to the detriment of the ASAD agents themselves.

J. Cartlidge, C. Szostek, M. De Luca, & D. Cliff, (2012), “Too fast too furious: faster financial-market trading agents can give less efficient markets,” in Proc. 4th Int. Conf. Agents and Artif. Intelligence, Vol. 2 - Agents (ICAART-2012), J. Filipe & A. Fred, Eds. Vilamoura, Portugal: SciTePress, Feb. 2012, pp. 126-135. [Available Online] [Bibtex] [Conference Link] [Presentation Slides] doi:10.5220/0003720301260135 (10 pages, approx. 7,500 words).

Abstract—For many of the world's major financial markets, the proportion of market activity that is due to the actions of “automated trading” software agents is rising: in Europe and the USA, major exchanges are reporting that 30%—75% of all transactions currently involve automated traders. This is a major application area for artificial intelligence and autonomous agents, yet there have been very few controlled laboratory experiments studying the interactions between human and software-agent traders. In this paper we report on results from new human-agent experiments using the OpEx experimental economics system first introduced at ICAART-2011. Experiments explore the extent to which the performance of the traders, and of the market overall, is dependent on the speed at which the agents operate. Surprisingly, we found that slowing down the agents increased the market's overall ability to settle to a competitive equilibrium, and that slow-agent markets were more efficient.

J. Cartlidge & D. Cliff, (2012), “Exploring the “robot phase transition” in experimental human-algorithmic markets.” Foresight, The Future of Computer Trading in Financial Markets, Driver Review DR25, Crown Copyright, Apr. 2012. [Available Online] [Available Online 2] [Bibtex] [Abstract] (50 pages, approx. 13,000 words). URN: 12/1058

Abstract—Johnson et al. (2012) recently argued that analysis of millisecond-by-millisecond stock-price movements between 2006 and 2011 suggests the existence of a step-change or phase transition in the dynamics and behaviour of financial markets, in which human traders and automated algorithmic 'robot' trading systems freely interact. Above a particular time-threshold, humans and algorithmic systems trade with one another; below the threshold, there is a sudden change to a market in which humans cannot participate and where all transactions are robot-to-robot. We refer to this abrupt system transition from a mixed human-robot phase to an all-robot phase as the 'robot phase transition'. At sub-second timescales, below the robot transition, Johnson et al. argue that the robot-only market exhibits 'fractures' - ultra-fast swings in price akin to mini flash-crashes - that are undesirable, little understood, and intriguingly appear to be linked to longer-term instability of the market as a whole.

     Here, we report on using a complementary approach to the historical data analysis employed by Johnson et al.: in March 2012 we conducted laboratory-style experiments where human traders interacted with algorithmic trading agents (i.e., robots) in minimal experimental models of electronic financial markets using De Luca's (2011) OpEx artificial financial exchange. Our aim was to see if correlates of the two regimes suggested by Johnson et al. occur in such laboratory conditions. Our results thus far do indeed indicate that when trading robots act on a super-human timescale, the market starts to fragment, with statistically lower human-robot interactions than we would expect from a fully mixed market. In contrast, when robotic trader agents are slowed to a thinking-and-reaction speed similar to that of humans, less fragmentation is observed. We tentatively conclude that this is evidence for the robot transition occurring in controlled experimental financial market systems.

     The work reported here also explore the effects of increasing the degree of realism in the laboratory experiments: we find that some statistically significant effects may be consequences of constraints introduced to make the analysis of the experiment results easier, and we report on our discovery of a problem with earlier OpEx experiments that casts doubt on their results.

M. De Luca, C. Szostek, J. Cartlidge, & D. Cliff, (2011), “Studies of interactions between human traders and algorithmic trading systems.” Foresight, The Future of Computer Trading in Financial Markets, Driver Review DR13, Crown Copyright, Sep. 2011. [Available Online] [Available Online 2] [Bibtex] [Abstract] (60 pages, approx. 23,000 words). URN: 11/1232

Abstract—In recent years there has been a very significant increase in the percentage of trades in the global financial markets that are initiated and executed by automated “robot” algorithmic trading software systems, autonomously performing trading roles that a decade or more ago would have been performed by human traders. The anonymity of many current electronic trading systems, operated by major exchanges and multilateral trading facilities, mean that an individual trader, whether human or robot, never knows if the counterparty to a particular trade is a human or not. There are commonly-quoted estimates that the proportion of robot-executed trades is approaching 30%-70% on major European and US equity exchanges. In foreign-exchange markets, where there are no central exchanges, the proportion of spot (immediate-execution) transactions that are executed by robots is widely believed to be even higher. From this, it is clear that the current global financial markets involve a very significant degree of interaction between human and robot traders.

     The interactions between human traders in electronic markets has long been studied in the field known as Experimental Economics, and more recently the interactions between software-agent traders in electronic markets has been the topic of various abstract research studies in so-called Agent-based Computational Economics (ACE). These two research fields are largely distinct: the first studies markets populated entirely by human traders; the second studies markets populated entirely by algorithmic software-agent traders. There is a surprising lack of studies of the interactions between human traders and robot traders. That is, there is very little scientific literature that explores heterogeneous markets, populated by both humans and robots.

     In this document we review the very small amount of published literature that does describe scientific studies of interactions between human and robot traders under experimental conditions. We contend that the relative lack of such studies is a serious omission from the literature. We propose that De Luca’s (2010) Open Exchange (OpEx) open-source design for studying human-robot interactions in electronic marketplaces should be used as a free de facto standard for future work in this area. We illustrate the use of OpEx by summarising recently published peer-reviewed accounts of early experiments with OpEx, and then present a detailed description and analysis of results from some new experiments, conducted specifically for this review document, where we relax some of the artificial experimental constraints that have been used in earlier work.

     Experiments with the OpEx system indicate that the previously reported outperformance of the algorithmic trading systems over humans may well be related to the artificial nature of the experiment design that was employed in the earlier research: a design essentially unchanged since the first experimental economics results were published in the early 1960’s. When the flow of orders in the market was trickled in gradually (rather than all orders being released simultaneously, which was an artificial constraint in the designs of the earlier experiments) the performance of the software agents was no longer so clearly superior. Furthermore, when the software agents were slowed to operate on the same sort of timescales that human traders act on, the data we have thus far indicates that the market dynamics alter, but further experiments would be required to establish the significance of this with appropriate levels of certainty.

J. Cartlidge & I. Sriram, (2011), “Modelling resilience in cloud-scale data centres,” in Proc. 23rd European Modeling & Simulation Symposium (EMSS-2011), A. G. Bruzzone et al., Eds. Rome, Italy: University of Genoa Press, Sep. 2011, pp. 299-307. [Available Online] [Bibtex] [Abstract] [Conference Link] [Presentation Slides] (9 pages, approx. 6,000 words).

Abstract—The trend for cloud computing has initiated a race towards data centres (DC) of an ever-increasing size. The largest DCs now contain many hundreds of thousands of virtual machine (VM) services. Given the finite lifespan of hardware, such large DCs are subject to frequent hardware failure events that can lead to disruption of service. To counter this, multiple redundant copies of task threads may be distributed around a DC to ensure that individual hardware failures do not cause entire jobs to fail. Here, we present results demonstrating the resilience of different job scheduling algorithms in a simulated DC with hardware failure. We use a simple model of jobs distributed across a hardware network to demonstrate the relationship between resilience and additional communication costs of different scheduling methods.

J. Cartlidge & D. Ait-Boudaoud, (2011), “Autonomous virulence adaptation improves coevolutionary optimization,” IEEE Trans. Evol. Comput., vol. 15, no. 2, pp. 215–229, Apr. 2011. [Available Online] [Bibtex] [Abstract] [Journal Link] doi:10.1109/TEVC.2010.2073471 (15 pages, approx. 13,000 words).

Abstract—A novel approach for the autonomous virulence adaptation (AVA) of competing populations in a coevolutionary optimization framework is presented. Previous work has demonstrated that setting an appropriate virulence, v, of populations accelerates coevolutionary optimization by avoiding detrimental periods of disengagement. However, since the likelihood of disengagement varies both between systems and over time, choosing the ideal value of v is problematic. The AVA technique presented here uses a machine learning approach to continuously tune v as system engagement varies. In a simple, abstract domain, AVA is shown to successfully adapt to the most productive values of v. Further experiments, in more complex domains of sorting networks and maze navigation, demonstrate AVA’s efficiency over reduced virulence and the layered Pareto coevolutionary archive.

J. Cartlidge & S. Phelps, (2011), “Estimating demand for dynamic pricing in electronic markets,” GSTF Journal on Computing (JoC), vol. 1, no. 2, pp. 128–133, Feb. 2011. [Available Online] [Bibtex] [Abstract] [Journal Link] doi:10.5176/2010-2283_1.2.50 (6 pages, approx. 4,500 words).

Abstract—Economic theory suggests sellers can increase revenue through dynamic pricing; selling identical goods or services at different prices. However, such discrimination requires knowledge of the maximum price that each consumer is willing to pay; information that is often unavailable. Fortunately, electronic markets offer a solution; generating vast quantities of transaction data that, if used intelligently, enable consumer behaviour to be modelled and predicted.

     Using eBay as an exemplar market, we introduce a model for dynamic pricing that uses a statistical method for deriving the structure of demand from temporal bidding data. This work is a tentative first step of a wider research program to discover a practical methodology for automatically generating dynamic pricing models for the provision of cloud computing services; a pertinent problem with widespread commercial and theoretical interest.

J. Cartlidge & S. Phelps, (2010), “Estimating consumer demand from high-frequency data,” in Proc. Ann. Int. Academic Conf. Business Intelligence and Data Warehousing (BIDW-2010), K. Kumar, Ed. Singapore: Global Science and Technology Forum (GSTF), Jul. 2010, pp. 132–138. [Available Online] [Bibtex] [Abstract] [Conference Link] doi:10.5176/978-981-08-6308-1_14 (7 pages, approx. 5,000 words).

Abstract—Price discrimination offers sellers the possibility of increasing revenue by capturing a market’s consumer surplus: arising from the low price elasticity segment of customers that would have been prepared to pay more than the current market price. First degree price discrimination requires a seller to know the maximum (reserve) price that each consumer is willing to pay. Discouragingly, this information is often unavailable; making the theoretical ideal a practical impossibility.

     Electronic commerce offers a solution; with loyalty cards, transaction statements and online accounts all providing channels of customer monitoring. The vast amount of data generated—eBay alone produces terabytes daily—creates an invaluable repository of information that, if used intelligently, enables consumer behaviour to be modelled and predicted. Thus, once behavioural models are calibrated, price discrimination can be tailored to the level of the individual.

     Here, we introduce a statistical method designed to model the behaviour of bidders on eBay to estimate demand functions for individual item classes. Using eBay’s temporal bidding data—arrival times, price, user id—the model generates estimates of individual reserve prices for market participants; including hidden, or censored, demand not directly contained within the underlying data. Market demand is then estimated, enabling eBay power sellers—large professional bulk-sellers—to optimize sales and increase revenue. Proprietary software automates this process: analyzing data; modelling behaviour; estimating demand; and generating sales strategy.

     This work is a tentative first step of a wider, ongoing, research program to discover a practical methodology for automatically calibrating models of consumers from large-scale high-frequency data. Multi-agent systems and artificial intelligence offer principled approaches to the modelling of complex interactions between multiple individuals. The goal is to dynamically model market interactions using realistic models of individual consumers. Such models offer greater flexibility and insight than static, historical, data analysis alone.

J. Cartlidge, (2008), “Dynamically adapting parasite virulence to combat coevolutionary disengagement (abstract),” in Proc. 11th Int. Conf. Simulation and Synthesis Living Systems (Alife-11), S. Bullock et al., Eds. Winchester, UK: MIT Press, Aug. 2008, p. 757. [Available Online] [Bibtex] [Abstract] [Conference Link] [Presentation Slides] (1 page, approx. 500 words).

Abstract— Participating in an evolutionary arms-race, natural coevolutionary predator-prey and host-parasite systems often exhibit accelerated evolution. Competitive Coevolutionary Genetic Algorithms (CCGAs) attempt to harness this evolutionary acceleration by engaging (multiple) evolving populations in competitive self-play.

     By evaluating individuals competitively, CCGAs afford the possibility of tackling problems that are illdefined, open-ended and lacking in formalism. This offers CCGAs a potential advantage over more traditional Genetic Algorithms (GAs) when fitness evaluation is difficult to operationally define. Analogously, one imagines that it is much easier to formally define the rules of the game of tennis than it is to define tennis playing ability. In practice, however, defining an appropriate game is often non-trivial.

     Competitive evaluation leaves CCGAs susceptible to some adverse evolutionary dynamics. One such hindrance is “disengagement”. This occurs when one coevolving population gets the upper hand and begins to easily outperform the other. Since it becomes impossible to discriminate between individuals according to ability, the selection gradient disappears and the coevolving populations begin to stagnate. The result is a stymied system that is left to flounder aimlessly.

     To prevent disengagement, Cartlidge has previously introduced the “Reduced Virulence” technique (Cartlidge & Bullock, 2004, Evol. Comp., 12, p.193). This technique helps avoid disengagement by reigning in a population that inherits an advantageous bias. Rather than reward individuals that maximally damage a competitor, Reduced Virulence favors individuals that give opponents a chance. Perhaps counter-intuitively, Reduced Virulence enables accelerated evolutionary progress by disadvantaging a population’s most successful individuals.

     In this work, Reduced Virulence undergoes a rigorous sensitivity analysis in the Counting Ones domain (introduced by Watson & Pollack, 2001, GECCO, p.702, Morgan Kaufmann); an analytically tractable substrate designed to highlight the dynamics of coevolution. Following intuition, it is shown that for optimal performance, virulence should be increasingly reduced as the asymmetrical bias (and thus likelihood of disengagement) between coevolving populations increases. Interestingly, even when coevolution is unbiased, “Maximum Virulence”—equivalent to the canonical fitness evaluation of “reward all victories”—is shown not to be ideal. Thus, results suggest that (in the Counting Ones domain) when population sizes are small, it is never the case that the canonical coevolutionary setup should be favored. The generality of this result, however, is an open question.

     Utilizing this information, a novel “Dynamic Virulence” algorithm is introduced. This algorithm adapts population virulence over time as populations evolve. It is shown that Dynamic Virulence is able to cope with varying bias better than fixed virulence and allows the discovery of optimal solutions under a much wider range of conditions than any individual fixed virulence setting.

     Finally, it is discussed how analyzing the role of virulence in artificial systems may allow us to better understand virulence in nature. For instance, perhaps there is potential for a “Reduced Virulence” approach to tackling infectious diseases. Rather than killing mosquitoes to eradicate malaria, one could alternatively encourage malaria-resistant strains that are better able to survive.

J. Cartlidge & S. Bullock, (2004), “Combating coevolutionary disengagement by reducing parasite virulence,” Evol. Comput., vol. 12, no. 2, pp. 193–222, Summer 2004. [Available Online] [Bibtex] [Abstract] [Journal Link] doi:10.1162/106365604773955148 (30 pages, approx. 15,000 words).

Abstract—While standard evolutionary algorithms employ a static, absolute fitness metric, coevolutionary algorithms assess individuals by their performance relative to populations of opponents that are themselves evolving. Although this arrangement offers the possibility of avoiding long-standing difficulties such as premature convergence, it suffers from its own unique problems, cycling, over-focusing and disengagement.

     Here, we introduce a novel technique for dealing with the third and least explored of these problems. Inspired by studies of natural host-parasite systems, we show that disengagement can be avoided by selecting for individuals that exhibit reduced levels of “virulence”, rather than maximum ability to defeat coevolutionary adversaries. Experiments in both simple and complex domains are used to explain how this counterintuitive approach may be used to improve the success of coevolutionary algorithms.

J. Cartlidge & S. Bullock, (2004), “Unpicking tartan CIAO plots: Understanding irregular coevolutionary cycling,” Adaptive Behaviour, vol. 12, no. 2, pp. 69–92, Jun. 2004. [Available Online] [Bibtex] [Abstract] [Journal Link] doi:10.1177/105971230401200201 (24 pages, approx. 14,000 words).

Abstract—We report results from a series of studies coevolving players for simple Rock–Paper–Scissors games. These results demonstrate that “Current Individual versus Ancestral Opponent” (CIAO) plots, which have been proposed as a visualization technique for detecting both coevolutionary progress and coevolutionary cycling, suffer from ambiguity with respect to an important but rarely discussed class of cyclic behavior. While regular cycling manifests itself as a characteristic banded plot, irregular cycling produces an irregular tartan pattern which is also consistent with random drift through strategy space. Although this tartan pattern is often reported in the literature on coevolutionary algorithms, it has received little attention or analysis. Here we argue that irregular cycling will tend to be more prevalent than regular cycling, and that it corresponds to a class of coevolutionary scenario that is important both theoretically and in practice. As such, it is desirable that we improve our ability to distinguish its occurrence from that of random drift, and other forms of coevolutionary dynamic.

J. Cartlidge, (2004), “Rules of Engagement: Competitive Coevolutionary Dynamics in Computational Systems,” PhD thesis, Sch. Comput., Univ. Leeds, UK. [Available Online] [Bibtex] [Abstract] [Department Link] (210 pages, approx. 75,000 words).

Abstract—Given that evolutionary biologists have considered coevolutionary interactions since the dawn of Darwinism, it is perhaps surprising that coevolution was largely overlooked during the formative years of evolutionary computing. It was not until the early 1990s that Hillis’ seminal work thrust coevolution into the spotlight.

     Upon attempting to evolve fixed-length sorting networks, a problem with a long and competitive history, Hillis found that his standard evolutionary algorithm was producing sub-standard networks. In response, he decided to reciprocally evolve a population of test lists against the sorting network population; thus producing a coevolutionary system. The result was impressive; coevolution not only outperformed evolution, but the best network it discovered was only one comparison longer than the best-known solution. For the first time, a coevolutionary algorithm had been successfully applied to problem-solving.

     Pre-Hillis, the shortcomings of standard evolutionary algorithms had been understood for some time: whilst defining an adequate fitness function can be as challenging as the problem one is hoping to solve, once achieved, the accumulation of fitness-improving mutations can push a population towards local optima that are difficult to escape. Coevolution offers a solution. By allowing the fitness of each evolving individual to vary (through competition) with other reciprocally evolving individuals, coevolution removes the requirement of a fitness yardstick. In conjunction, the reciprocal adaptations of each individual begin to erode local optima as soon as they appear.

     However, coevolution is no panacea. As a problem-solving tool, coevolutionary algorithms suffer from some debilitating dynamics, each a result of the relative fitness assessment of individuals. In a single-, or multi-, population competitive system, coevolution may stabilize at a suboptimal equilibrium, or mediocre stable state; analogous to the traditional problem of local optima. Populations may become highly specialized in an unanticipated (and undesirable) manner; potentially resulting in brittle solutions that are fragile to perturbation. The system may cycle; producing dynamics similar to the children's game rock-paper-scissors. Disengagement may occur, whereby one population out-performs another to the extent that individuals cannot be discriminated on the basis of fitness alone; thus removing selection pressure and allowing populations to drift. Finally, coevolution's relative fitness assessment renders traditional visualization techniques (such as the graph of fitness over time) obsolete; thus exacerbating each of the above problems.

     This thesis attempts to better understand and address the problems of coevolution through the design and analysis of simple coevolutionary models. “Reduced virulence”— a novel technique specifically designed to tackle disengagement—is developed. Empirical results demonstrate the ability of reduced virulence to combat disengagement both in simple and complex domains, whilst outperforming the only known competitors. Combining reduced virulence with diversity maintenance techniques is also shown to counteract mediocre stability and over-specialization.

     A critique of the CIAO plot—a visualization technique developed to detect coevolutionary cycling—highlights previously undocumented ambiguities; experimental evidence demonstrates the need for complementary visualizations. Extending the scope of visualization, a first exploration into coevolutionary steering is performed; a technique allowing the user to interact with a coevolutionary system during run-time. Using a simple model incorporating reduced virulence, the coevolutionary steering demonstration highlights the future potential of such tools for both research and education.

     The role of neutrality in coevolution is discussed in detail. Whilst much emphasis is placed upon neutral networks in the evolutionary computation literature, the nature of coevolutionary neutrality is generally overlooked. Preliminary ideas for modelling coevolutionary neutrality are presented.

     Finally, whilst this thesis is primarily aimed at a computing audience, strong reference to evolutionary biology is made throughout. Exemplifying potential crossover, the CIAO plot, a tool previously unused in biology, is applied to a simulation of E. Coli, with results confirming empirical observations of real bacteria.

J. Cartlidge & S. Bullock, (2003), “Caring versus sharing: How to maintain engagement and diversity in coevolving populations,” in Proc. 7th Eur. Conf. Artif. Life (ECAL’03), W. Banzhaf et al., Eds. Dortmund, Germany: Springer Verlag, Sep. 2003, pp. 299–308. [Available Online] [Bibtex] [Abstract] doi:10.1007/978-3-540-39432-7_32 (10 pages, approx. 5,000 words).

Abstract—Coevolutionary optimisation suffers from a series of problems that interfere with the progressive escalating arms races that are hoped might solve difficult classes of optimisation problem. Here we explore the extent to which encouraging moderation in one coevolving population (termed parasites) can alleviate the problem of coevolutionary disengagement. Results suggest that, under these conditions, disengagement is avoided through maintaining variation in relative fitness scores. In order to explore whether standard diversity maintenance techniques such as resource sharing could achieve the same effects, we compare moderating virulence with resource sharing in a simple matching game. We demonstrate that moderating parasite virulence differs significantly from resource sharing, and that its tendency to prevent disengagement can also reduce the likelihood of coevolutionary optimisation halting at mediocre stable states.

S. Bullock, J. Cartlidge, & M. Thompson, (2002), “Prospects for computational steering of evolutionary computation,” in Workshop Proc. 8th Int. Conf. Artif. Life, E. Bilotta et al., Eds. Sydney, Australia: MIT Press, Dec. 2002, pp. 131–137. [Available Online] [Bibtex] [Abstract] [Conference Link] (7 pages, approx. 4,500 words).

Abstract—Currently, evolutionary computation (EC) typically takes place in batch mode: algorithms are run autonomously, with the user providing little or no intervention or guidance. Although it is rarely possible to specify in advance, on the basis of EC theory, the optimal evolutionary algorithm for a particular problem, it seems likely that experienced EC practitioners possess considerable tacit knowledge of how evolutionary algorithms work. In situations such as this, computational steering (ongoing, informed user intervention in the execution of an otherwise autonomous computational process) has been profitably exploited to improve performance and generate insights into computational processes. In this short paper, prospects for the computational steering of evolutionary computation are assessed, and a prototype example of computational steering applied to a coevolutionary algorithm is presented.

J. Cartlidge & S. Bullock, (2002), “Learning lessons from the common cold: How reducing parasite virulence improves coevolutionary optimization,” in Proc. Congr. Evol. Comput. (CEC’02), D. Fogel et al., Eds. Honolulu, HI: IEEE Press, Jun. 2002, pp. 1420–1425. [Available Online] [Bibtex] [Abstract] [Conference Link] doi:10.1109/CEC.2002.1004451 (7 pages, approx. 4,500 words).

Abstract—Inspired by the virulence of natural parasites, a novel approach is developed to tackle disengagement, a detrimental phenomenon coevolutionary systems sometimes experience. After demonstrating beneficial results in a simple model, minimum-comparison sorting networks are coevolved, with results suggesting that moderating parasite virulence can help in practical problem domains.


©John Cartlidge.  Page last modified on: