Does the process of training a neural network to solve a task tend to use all of the available weights even when the task could be solved with fewer weights? To address this question we study the effects of pruning fully connected, convolutional and residual models while varying their widths. We find that the proportion of weights that can be pruned without degrading performance is largely invariant to model size. Increasing the width of a model has little effect on the density of the pruned model relative to the increase in absolute size of the pruned network. In particular, we find substantial prunability across a large range of model sizes, where our biggest model is 50 times as wide as our smallest model. We explore three hypotheses that could explain these findings. Source code: [29].

Multi-agent consensus problems can often be seen as a sequence of autonomous and independent local choices between a finite set of decision options, with each local choice undertaken simultaneously, and with a shared goal of achieving a global consensus state. Being able to estimate probabilities for the different outcomes and to predict how long it takes for a consensus to be formed, if ever, are core issues for such protocols. Little attention has been given to protocols in which agents can remember past or outdated states. In this paper, we propose a framework to study what we call `memory consensus protocol'. We show that the employment of memory allows such processes to always converge, as well as, in some scenarios, such as cycles, converge faster. We provide a theoretical analysis of the probability of each option eventually winning such processes based on the initial opinions expressed by agents. Further, we perform experiments to investigate network topologies in which agents benefit from memory on the expected time needed for consensus.

One of the main challenges in integrating different smart applications is security. Among the problems related to security, data access control is one of the most important, given that it involves end-users' privacy. In this paper, we propose an approach to the modelling of data access control interfaces using argumentation-based agents. In particular, we introduce argumentation schemes for data access control, which are based on some of the most relevant models for data access control currently available. Our approach considers not only the usual access control policies, describing which category of agents has access to which category of information, but also emergency policies, describing situations where special emergency access control rules apply. Using argumentation-based agents as access control interfaces allows us to deal with the uncertainty about external information, allowing a correct categorisation of agents that request access to information, as well as allowing agents to expose emergency situations in which emergency access control rules may apply.

Recently, argumentation frameworks have been extended in order to consider trust when defining preferences between arguments, given that arguments (or information that supports the arguments) from more trustworthy sources may be preferred to arguments from less trustworthy sources. Although such literature presents interesting results on argumentation-based reasoning and how agents define preferences between arguments, there is little work taking into account agent strategies for argumentation-based dialogues using such information. In this work, we propose an argumentation framework in which agents consider how much the recipient of an argument trusts others in order to choose the most suitable argument for that particular recipient, i.e., arguments constructed using information from those sources that the recipient trusts. Our approach aims to allow agents to construct more effective arguments, depending on the recipients and on their views on the trustworthiness of potential sources.

The increase in routine clinical data collection coupled with an expectation to exploit this in support of evidence based decision making creates a need for an intelligent model selection system to support clinicians when analysing data because clinicians often lack the statistical expertise to do this independently. In a previous position paper, an argumentation based approach to devise a decision support system for such an application was introduced. This approach ignored the relative strength of arguments for and against alternative models. This paper demonstrates how an extended argumentation framework can be employed to capture and reason with statistical and research domain knowledge that affects the relative strength of arguments. The approach is validated by means of a real-world case study.

Options constitute integral part of modern financial trades, and are priced according to the risk associated with buying or selling certain asset in future. Financial literature mostly concentrates on risk-neutral methods of pricing options such as Black-Scholes model. However, it is an emerging field in option pricing theory to use trading agents with utility functions to determine the option's potential payoff for the agent. In this paper, we use one of such methodologies developed by Othman and Sandholm to design portfolio-holding agents that are endowed with popular option portfolios such as bullish spread, butterfly spread, straddle, etc to price options. Agents use their portfolios to evaluate how buying or selling certain option would change their current payoff structure, and form their orders based on this information. We also simulate these agents in a multi-unit direct double auction. The emerging prices are compared to risk-neutral prices under different market conditions. Through an appropriate endowment of option portfolios to agents, we can also mimic market conditions where the population of agents are bearish, bullish, neutral or non-neutral in their beliefs.

We analyse the value of information in a stock market where information can be noisy and costly, using techniques from empirical game theory. Previous work has shown that the value of information follows a J-curve, where averagely informed traders perform below market average, and only insiders prevail. Here we show that both noise and cost can change this picture, in several cases leading to opposite results where insiders perform below market average, and averagely informed traders prevail. Moreover, we investigate the effect of random explorative actions on the market dynamics, showing how these lead to a mix of traders being sustained in equilibrium. These results provide insight into the complexity of real marketplaces, and show under which conditions a broad mix of different trading strategies might be sustainable.

The increase in routine clinical data collection coupled with an expectation to exploit this in support of evidence based decision making creates the requirement for a system to support clinicians in this analysis. This paper looks at applying argumentation to this problem, by collating all the relevant statistical approaches and their assumptions into a statistical knowledge base and then representing the model selection process through argumentation. This will form the foundation for the development of a prototype that will enable clinicians to answer their research questions with no statistics, informatics or administrative support.

Infrastructure-less mobile ad-hoc networks enable the development of collaborative pervasive applications. Within such dynamic networks, collaboration between devices can be realised through service-orientation by abstracting device resources as services. Recently, a framework for QoS-aware service composition has been introduced which takes into account a spectrum of orchestration patterns, and enables compositions of a better QoS than traditional centralised orchestration approaches. In this paper, we focus on the automated exploration of trade-off compositions within the search space defined by this flexible composition model. For the studied problem, the evaluation of the fitness functions guiding the search process is computationally expensive because it either involves a high-fidelity simulation or actually requires calling the composite service. To overcome this limitation, we have developed efficient surrogate models for estimating the QoS metrics of a candidate solution during the search. Our experimental results show that the use of surrogates can produce solutions with good convergence and diversity properties at a much lower computational effort.

Service-oriented pervasive systems, composed of a large number of devices with heterogeneous capabilities where devices' resources are abstracted as software services, challenge the creation of high-quality composite applications. Resource heterogeneity, dynamic network connectivity, and a large number of highly distributed service providers complicate the process of creating applications with specific QoS requirements. Existing approaches to service composition control the QoS of an application solely by changing the set of participating concrete services which is not suitable for ad-hoc service-based systems characterised by high intermittent connectivity and resource heterogeneity. In this paper, we propose a flexible way of formulating composition configurations suitable for such service-based systems. Our formulation proposes the combined consideration of the following factors that affect the QoS of a composed service: (a) service selection, (b) orchestration partitioning, and (c) orchestrator node selection. We show that the proposed formulation enables the definition of service composition configurations with 49% lower response time, 28% lower network latency, 36% lower energy consumption, and 13% higher success ratio compared to those defined with the traditional approach.

Open multiagent systems (MAS) typically require the participating agents to comply with system-level rules, or norms, and may punish non-compliance. An interesting challenge for designers of such systems is how to provide for dynamic, run-time allocation of norms. In this article, we present a novel, framework-independent approach enabling a norm architecture to be overlaid on any type of multiagent system. We achieve this by defining rights, responsibilities and sanctions in a manner which allows for their run-time assignment and re-assignment to agents in a normative multiagent system. The overlay architecture we propose requires the system to include a normative knowledge (KB) base containing explicit representation of norms as conditional rules, where the preconditions for instantiation of a norm are various run-time occurrences, such as the execution of some agent action, or the occurrence of some event. Our method has both wide generality and flexibility, allowing at one extreme few or no agent roles, each role having many conditional norms, or allowing, at the other extreme, many roles, each having only a small number of conditional norms, or allowing, indeed, any position between these two extremes.

Trust is a natural mechanism by which an autonomous party can deal with the inherent uncertainty regarding the behaviors of other parties and the uncertainty in the information it shares with those parties. Trust is thus crucial in any decentralized system. We build on recent efforts to use argumentation to reason about trust. Specifically, we provide a set of schemes, abstract patterns of reasoning that apply in multiple situations, geared toward trust. We describe, in the form of a set of critical questions, the situations in which the schemes may default.

Within both economics and computer science, many authors have claimed that decentralized or market-based approaches to decision making are superior in general to centralized approaches. The contrary claim has also been made. Unfortunately, these claims are often supported only by informal or anecdotal evidence. In order to assess these competing claims, we present a review of the literature in economics and computer science with a bearing on these issues. Our main findings from this literature survey are as follows: (1) for efficiency assessments, there is wider variance in performance of organizations using market-based control (MBC) than in organizations using centralized control (CC); (2) that MBC and CC have the same efficiency on average, which may explain the observation that (3) human and computer organizations tend to cycle between CC and decentralized control (DC) structures.

Legacy description not available

We propose a novel method for allocating multi-attribute computational resources via competing marketplaces. Trading agents, working on behalf of resource consumers and providers, choose to trade in resource markets where the resources being traded best align with their preferences and constraints. Market-exchange agents, in competition with each other, attempt to provide resource markets that attract traders, with the goal of maximising their profit. Because exchanges can only partially observe global supply and demand schedules, novel strategies are required to automate their search for market niches. Novel attribute-level selection (ALS) strategies are empirically analysed in simulated competitive market environments, and results suggest that using these strategies, market-exchanges can seek out market niches under a variety of environmental conditions.

For agents to fulfill their potential of being intelligent and adaptive, it is useful to model their interaction protocols as executable entities that can be referenced, inspected, composed, shared, and invoked between agents, all at runtime. We use the term first‐class protocol to refer to such protocols. Rather than having hard‐coded decision‐making mechanisms for choosing their next move, agents can inspect the protocol specification at runtime to do so, increasing their flexibility. In this article, we show that propositional dynamic logic (PDL) can be used to represent and reason about the outcomes of first‐class protocols. We define a proof system for PDL that permits reasoning about recursively defined protocols. The proof system is divided into two parts: one for reasoning about terminating protocols, and one for reasoning about nonterminating protocols. We prove that proofs about terminating protocols can be automated, while proofs about nonterminating protocols are unable to be automated in some cases. We prove that, for a restricted class of nonterminating protocols, proofs about them can be transformed to proofs about terminating protocols, making them automatable.

Many websites nowadays allow social networking between their users in an explicit or implicit way. In this work, we show how the theory of argumentation schemes can provide a valuable help to formalize and structure on-line discussions and user opinions in decision support and business oriented websites that hold social networks among their users. A real study case is considered and analysed. Then, guidelines for website and system design are provided to enhance social decision support and recommendations with argumentation.

We introduce a method for strategy acquisition in nonzero-sum n -player games and empirically validate it by applying it to a well-known benchmark problem in this domain, namely, the double-auction market. Many existing approaches to strategy acquisition focus on attempting to find strategies that are robust in the sense that they are good all-round performers against all-comers. We argue that, in many economic and multiagent scenarios, the robustness criterion is inappropriate; in contrast, our method focuses on searching for strategies that are likely to be adopted by participating agents, which is formalized as the size of a strategy's basins of attraction under the replicator dynamics.

We consider the issue of representing coalitional games in multi-agent systems with externalities (i.e., in systems where the performance of one coalition may be affected by other co-existing coalitions). In addition to the conventional partition function game representation (PFG), we propose a number of new representations based on a new notion of externalities. In contrast to conventional game theory, our new concept is not related to the process by which the coalitions are formed, but rather to the effect that each coalition may have on the entire system and vice versa. We show that the new representations are fully expressive and, for many classes of games, more concise than the conventional PFG. Building upon these new representations, we propose a number of approaches to solve the coalition structure generation problem in systems with externalities. We show that, if externalities are characterised by various degrees of regularity, the new representations allow us to adapt coalition structure generation algorithms that were originally designed for domains with no externalities, so that they can be used when externalities are present. Finally, building upon Rahwan et al. [16] and Michalak et al. [9], we present a unified method to solve the coalition structure generation problem in any system, with or without externalities, provided sufficient information is available.

The authors are grateful for financial support received from the UK EP-SRC through the project Market-Based Control of Complex Computational Systems (GR/T10657/01). The authors are also thankful to Jennifer McManus, School of English, University of Liverpool for excellent editorial assistance.

Deliberation dialogues occur when two or more participants seek to jointly agree on an action or a course of action in some situation. We present the first formal framework for such dialogues, grounding it in a theory of deliberative reasoning from the philosophy of argumentation. We further fully articulate the locutions and rules of a formal dialogue game for this model, so as to specify a protocol for deliberation dialogues. The resulting protocol is suitable for dialogues between computational entities, such as autonomous software agents. To assess our protocol, we consider it against various records of human deliberations, against normative principles for the conduct of human dialogues, and with respect to the outcomes produced by dialogues undertaken according to the protocol. © 2007 Wiley Periodicals, Inc. Int J Int Syst 22: 95–132, 2007.

This article introduces the special issue of ACM Transactions on Autonomous and Adaptive Systems devoted to research papers arising from the three Technical Forum Group meetings held in 2004 and 2005 that were organized and sponsored by the European FP6 Coordination Action AgentLink III.

In this paper we apply a general account of practical reasoning to arguing about legal cases. In particular, we describe how the reasoning in one very well known property law case can be reconstructed in terms of our account. We begin by summarising our general approach which uses instantiations of an argumentation scheme to provide presumptive justifications for actions, and critical questions to identify arguments which attack these justifications. These arguments and attacks are organised into argumentation frameworks to identify the status of individual arguments. Different beliefs about, and perspectives on, the issue are represented by different agents based on the Belief-Desire-Intention model, and conditions under which these agents may advance justifications and attack them are described. We model the different views of our case in these terms, describe the resulting argumentation frameworks, and relate them to the original majority and dissenting opinions. We contend that this approach both shows the worth of the general approach and its applicability to the legal domain.

In this paper we attempt to give an account of reasoning with legal cases contextualised within a general theory of persuasion in practical reasoning. We begin by presenting our general theory, concentrating on the variety of ways in which a particular position can be attacked. We then apply our theory to the legal domain, illustrating our approach by a case study based on the well known CATO system. From this we conclude that it is possible to see reasoning with legal cases as a particular instantiation of our general theory. We identify some points of interest for discussion, and conclude by stating our intended directions for future work.