Muthukkaruppan Annamalai

A Dynamic Measurement of Agent Autonomy in the Layered Adjustable Autonomy Model

In a dynamically interactive systems that contain a mix of humans’ and software agents’ intelligence, managing autonomy is a challenging task. Giving an agent a complete control over its autonomy is a risky practice while manually setting the agent’s autonomy level is an inefficient approach. In this paper, we propose an autonomy measurement mechanism and its related formulae for the Layered Adjustable Autonomy (LAA) model. Our model provides a mechanism that optimizes autonomy distribution, consequently, enabling global control of the autonomous agents that guides or even withholds them whenever necessary. This is achieved by formulating intervention rules on the agents’ decision-making capabilities through autonomy measurement criteria. Our aim is to create an autonomy model that is flexible and reliable.

A conceptual model of layered adjustable autonomy

Autonomy and autonomous agents are currently the most researched topics in autonomous systems. Issues like autonomy adjustment, autonomy level, and the required degree of autonomy to be performed are investigated. Abstracting an autonomy model poses the problem of identifying specific aspects that merit an autonomous system. In this paper, we propose another model of autonomy that conceptualizes autonomy as a spectrum, which is constructed in a layered structure of a multi-agent environment called Layered Adjustable Autonomy (LAA). The autonomy spectrum of the LAA is divided into adjustable-leveled layers. Each of which has distinct attributes and properties that assist an agent in managing the influences of the environment during its decision-making process. The LAA structure is designed to endorse an agent’s qualification to make a decision by setting the degree of autonomy to the agent’s choice of decision-making. An Autonomy Analysis Module (AAM) is also proposed to control and delegate the agent’s actions at specific autonomy levels. Hence, the AAM determines the threshold of the agent autonomy level to act in its qualified layer. Ultimately, the proposed LAA model will be implemented on an air drone for the purpose of testing and refinement.

A Dynamically Adjustable Autonomic Agent Framework

The design and development of autonomous software agents is still a challenging task and needs further investigation. Giving an agent the maximum autonomous capabilities may not necessarily produce satisfactory agent behavior. Consequently, adjustable autonomy has become the hallmark of autonomous systems development that influences an agent to exhibit satisfactory behavior. To perform such influences, however, a dynamic adjustment mechanism is needed to be configured. The influences are costly in time and implementation especially for systems with time-critical domain. They might negatively influence agent decisions and cause system disturbance. In this paper, we propose a framework to govern an agent autonomy adjustment and minimize system disturbance. The main components of the proposed framework are the Planner, Scheduler and Controller (PSC) that conform to the current trends in automated systems. Two modules are also suggested which are Autonomy Analysis Module (AAM) and Situation Awareness Module (SAM). They are accordingly used to distribute the autonomy and provide balance to the system so that it’s local and global desires do not conflict.

Agent-based modelling and simulation of emergency department re-triage

Triage is a process of accessing patients on their severity based on a triage acuity scale in hospital emergency department (ED). Re-triage is a process where the severity of a patients condition is reassessed when there is a clinical need for it. Re-triage does not feature in the conventional triage, where the patients with non-urgent consideration will have to wait to be treated on a first come first serve basis. In this study, we investigate the effect of re-triage on patients waiting time and on the ED service by means of agent-based modelling and simulation. The simulation is based on historical records of patients presenting to the ED of Hospital USM in the year 2011. The result of the simulation shows that the implementation of re-triage in the conventional three-level triage system can significantly Reduce the waiting time of patients with deteriorating clinical conditions, with slight increase in the demand for ED service due to the re-triage activity.

Formulating Dynamic Agents’ Operational State via Situation Awareness Assessment

Managing autonomy in a dynamic interactive system that contains a mix of human and software agent intelligence is a challenging task. In such systems, giving an agent a complete control over its autonomy is a risky practice while manually setting the agent’s autonomy level is an inefficient approach. This paper addresses this issue via formulating a Situation Awareness Assessment (SAA) technique to assist in determining an appropriate agents’ operational state. We propose four operational states of agents’ execution cycles; proceed, halt, block and terminate, each of which is determined based on the agents’ performance. We apply the SAA technique in a proposed Layered Adjustable Autonomy (LAA) model. The LAA conceptualizes autonomy as a spectrum and is constructed in a layered structure. The SAA and the LAA notions are applicable to humans’ and agents’ collaborative environment. We provide an experimental scenario to test and validate the proposed notions in a real-time application.

Jambalaya: The closest visualisation fit for the protégé ontology conceptual-relationship tracer

Retrieving the relevant background information about the conceptual-relationships defined in an ontology is a prerequisite for evaluating the competency of the ontology. For this purpose, a Conceptual-Relationship Tracer (CRT) for Protégé ontologies has been developed in a previous work. Protégé ontologies are web-ontologies developed using the Protégé ontology editor. The CRTs textual output is sometimes hard to trace, making the intended information difficult to understand. We regard, a distinct visualisation of the output can help to better convey the intended information, thus improving its understandability. The candidates are TGViz, OntoViz and Jambalaya, the three independently developed Protégé visualiser plugins. In this paper, we provide a set of visualisation factors to qualitatively compare the utility and the usability potentials of the candidates, and to decide which one is best at providing the requisite visualisation support for the CRT. The results of the analyses show that Jambalaya is the closest visualisation fit.

Task-Based Modelling of the Triage Domain Knowledge

Triage is a decision-making process that classifies incoming patients for presentational urgency in Emergency Departments (EDs). There are issues with triage reliability in EDs, which we can be resolved through uniform application of a robust triage scale. However, the complex robust triaging knowledge is not easy to understand or recalled for timely decision-making. Therefore, we suggest the development of a knowledge-based triage decision support system to help triage officers to make correct and consistent triage decisions. Consequently, we pursued knowledge engineering to construct the models of the knowledge in order to make explicit the conceptualisation of the assumptions and constraints in triage decision-making. We regard task as a rationale basis for modelling the purposive domain knowledge. Consequently, the paper discusses the modelling of the domain knowledge to support the triage decision-making task. The triage decision-making task model is presented in a complementary paper. Together, the knowledge models can be viewed as meta models that provide the conceptual guiding principles for the consequent design of the triage decision support system.

Content Quality of Clustered Latent Dirichlet Allocation Short Summaries

Latent Dirichlet Allocation (LDA) is a commonly used topic model based summarisation method. However, the generated summaries contain words that are somewhat general and unrelated to the topic. Since the summary depends on word distribution in the input documents and, because the topic signature feature values are averaged across all documents, we think clustering can help to overcome this problem. Therefore, this work sets out to investigate whether clustering the input documents beforehand (clusLDA) can help to improve the content quality of the generated summaries. The words in a LDA summary are weighted and a short summary of 0.67% of the input text size is constituted using significant words proportionately drawn from the clustered summaries. The divergence probabilities of the resulting summaries are compared against the summary produced by LDA without clustering (UnclusLDA). The results are validated using input of various text sizes and different clustering techniques. And, our findings indicate that clustering does not necessarily help to improve the content quality of short summaries.

A knowledge-based approach in video frame processing using Iterative Qualitative Data Analysis

The ability to acquire, identify and represent the knowledge that a human expert has about a particular domain is a powerful key method in the development of a knowledge-based computer system. This paper demonstrates a methodology for acquiring and analyzing data based on semi-structured interview responses conducted upon human experts. Human experts are asked to determine the acceptability of an image containing person(s) in a sequence of images. Different experts may have different judgments and collectively the image values or attributes from their subjective judgment may contribute to the main factors of consideration in determining the overall image acceptability. The aim of this paper is to identify the most appropriate image attributes used by human experts during an image selection task which is in line with our research objectives. We discuss the knowledge acquisition task by adopting the Iterative Qualitative Data Analysis (IQDA) approach and represent the knowledge into a set of filtering attributes.

Policy gaps in the virtual world

The convergence of computer and communication technologies has brought the people around the world to be more connected electronically in a virtual world. The success of Internet has opened up possibilities to create many applications in the space of the virtual world, such as the virtual libraries, the virtual shopping malls and the virtual communities, which have availed opportunities to learn, do business, socialise and entertain in new ways. While in the real world we have policies to govern the physical systems and processes, these policies lose their effect the moment they enter the virtual world, mainly because of their vulnerability to abuse and misuse in a world where much of what is happening is not visible. The purpose of this paper is to inform about the faltering policy issues of the real world that give rise to policy gaps in the virtual world. We discuss this concern in the context of five practical virtual world application areas.