School of Computing & Informatics
https://repository.maseno.ac.ke/handle/123456789/1312
Sat, 03 Jun 2023 17:26:52 GMT2023-06-03T17:26:52ZA reinforcement learning multi-agent systems architecture for guaranteeing advancing conversations in task-oriented dialog systems
https://repository.maseno.ac.ke/handle/123456789/5533
A reinforcement learning multi-agent systems architecture for guaranteeing advancing conversations in task-oriented dialog systems
MUGOYE, Kevin. Sindu
The conversational capabilities of a dialog system have a direct impact on the tasks it can accomplish. Solving all conversational issues in dialog systems have the potential to make them serve in complex domains. While this is not achievable, addressing fundamental aspects in conversation is desired to make a task-oriented dialog system (TODS) serve in new domains where they are needed, besides increasing their usefulness. One such aspect is the ability to advance a conversation logically. The primary aim of this study was to develop a novel architecture that will guarantee advancing conversations in TODS. To realize this aim, theories and literature were interrogated that informed the formulation of an agent-based architecture for dialog management. Then implementation of the architecture previously realized in a dialog system prototype. Followed by training the dialog system on initial domain-specific data. And evaluating its performances in a specific domain. The study used exploratory methodology to provide the theories that justified the construction of the multi-agent system (MAS_DM) architecture, while the experimental design was explored to synthesize and train the prototype.The design involved the fusion of agent-based architecture with reinforcement learning technique to enable tracking of context, structure and policy without depending on handcrafted rules. MAS_DM architecture explores learning agents in an unknown environment, where each agent is endowed with the ability to learn and select a policy. Learning and policy selection is sustained through reinforcement learning, eliminating the need for handcrafted rules. The architectural model was evaluated and validated in a prototype Chabot system. The Chatbot system was trained and tested in the maternal healthcare domain and was evaluated by human users. In this context, each user filled out an online questionnaire after successful interaction with the Chatbot. The evaluation parameters were coherence, task success, general performance, user satisfaction and goal achievement. This evaluation adheres to the specifications of Goal Question Metrics and PARAdigm for DIalog System Evaluation frameworks. The key findings were that Chabot’s ability to advance the conversation scored 0.8903, and achieved an overall performance score of 0.553. It achieved a task success rate of 0.936. with a user satisfaction score of 0.775. Based on global acceptable measures, interpreted this task success as substantial, coherence score as substantial, user satisfaction as excellent and the overall performance as good. Where machine learning is involved kappa statistic values above 0.40 are considered exceptional. The results suggest that it is reasonable to conclude that the MAS_DM architecture can be trusted to guarantee conversations that advance logically. The study contributes to the body of knowledge of conversational artificial intelligence by; - developing a novel agent-based architectural model for TODS, demonstrating the practicability of combining multi-agent systems and machine learning toward solving conversational issues and enhancing the capability of TODS.
PhD Thesis
Sat, 01 Jan 2022 00:00:00 GMThttps://repository.maseno.ac.ke/handle/123456789/55332022-01-01T00:00:00ZScript Acquisition: A Crowdsourcing and Text mining approach
https://repository.maseno.ac.ke/handle/123456789/5395
Script Acquisition: A Crowdsourcing and Text mining approach
WANZARE, Lilian Diana Awuor
According to Grice’s (1975) theory of pragmatics, people tend to omit basic information
when participating in a conversation (or writing a narrative) under the assumption that left
out details are already known or can be inferred from commonsense knowledge by the
hearer (or reader). Writing and understanding of texts makes particular use of a specific
kind of common-sense knowledge, referred to as script knowledge. Schank and Abelson
(1977) proposed Scripts as a model of human knowledge represented in memory that stores
the frequent habitual activities, called scenarios, (e.g. eating in a fast food restaurant, etc.),
and the different courses of action in those routines.
This thesis addresses measures to provide a sound empirical basis for high-quality script
models. We work on three key areas related to script modeling: script knowledge acquisition, script induction and script identification in text. We extend the existing repository
of script knowledge bases in two different ways. First, we crowdsource a corpus of 40
scenarios with 100 event sequence descriptions (ESDs) each, thus going beyond the size of
previous script collections. Second, the corpus is enriched with partial alignments of ESDs,
done by human annotators. The crowdsourced partial alignments are used as prior knowledge to guide the semi-supervised script-induction algorithm proposed in this dissertation.
We further present a semi-supervised clustering approach to induce script structure from
crowdsourced descriptions of event sequences by grouping event descriptions into paraphrase sets and inducing their temporal order. The proposed semi-supervised clustering
model better handles order variation in scripts and extends script representation formalism,
Temporal Script graphs, by incorporating "arbitrary order" equivalence classes in order to
allow for the flexible event order inherent in scripts.
In the third part of this dissertation, we introduce the task of scenario detection, in which
we identify references to scripts in narrative texts. We curate a benchmark dataset of annotated narrative texts, with segments labeled according to the scripts they instantiate. The
dataset is the first of its kind. The analysis of the annotation shows that one can identify scenario references in text with reasonable reliability. Subsequently, we proposes a benchmark
model that automatically segments and identifies text fragments referring to given scenarios. The proposed model achieved promising results, and therefore opens up research on
script parsing and wide coverage script acquisition
Phd Thesis(Donation)
Tue, 01 Jan 2019 00:00:00 GMThttps://repository.maseno.ac.ke/handle/123456789/53952019-01-01T00:00:00ZCompact operators on sequence and function spaces: Characterisations and duality results
https://repository.maseno.ac.ke/handle/123456789/4342
Compact operators on sequence and function spaces: Characterisations and duality results
AYWA, Shem
Since the early seventies numerous papers appeared in which the authors considered the
question "For which Banach spaces X and Y is the space K(X, Y) of compact linear
operators uncomplemented in the space L(X, Y) of bounded linear operators?". The
most general answer to this question has close connection with the question whether the
scalar sequence space Co of null sequences embeds isomorphically into K(X, Y). In one of
the early papers ([28J (1979)) J. Johnson followed a path through dual spaces of spaces of
operators when he proved that there is a projection on L(X, Y)* with range isomorphic to
K(X, Y)* and kernel the annihilator of K(X, Y) when Y has the bounded approximation
property. Johnson then applied his result to consider L(X, Y) as an isomorphic subspace
of K(X, Y)** and could then derive necessary and sufficient conditions for K(X, Y) to
be reflexive, provided that either X or Y has the bounded approximation property. It
turns out that the spaces X and Y have to be reflexive and that the property K(X, Y) =
L(X, Y) necessarily has to hold for K(X, Y) to be reflexive in this case.
The questions described in the previous paragraph led to research into different directions.
Much work went into the study of K(X, Y) as a subspace of L(X, Y) and the question
when (i.e. for which Banach spaces) is the equality K(X, Y) = L(X, Y) true? For
instance an extensive investigation on when K(X, Y) is an M-ideal in L(X, Y) was done.
And some researchers (for example in the papers [2J and [6]) considered the question on
the equality of K(X, Y) and L(X, Y) in the context of scalar sequence spaces and Banach
function spaces - i.e. where either X or both X and Yare such spaces. Also, especially
in recent papers (for example in [10], [22], [25], [26J and [27]), the same questions and the
question about projections from L(X, Y)* onto K(X, Y)* were considered in the setting
of Banach spaces which fail the approximation property. These studies also extended
to similar research activities in the setting of locally convex spaces (for example in the
papers [8], [19J and [20]).
The objective in the present thesis is to contribute to the above mentioned study, in the
following ways:
(a) In line with recent developments we want to show the existence of a suitable
projection onto the space K(X. Y)*, which will allow us to find necessary and sufficient
conditions for the reflexivity of K(X, Y) without relying on the presence of the bounded
approximation property on X or Y. The idea is to put recent work of others in connection
with continuous dual spaces of spaces of bounded linear operators in a suitable framework
and to improve the present known results and techniques.
(b) Use techniques from the theory of vector sequence spaces to simplify proofs of
existing results in connection with the equality K(X, Y) = L(X, Y) when X is a Banach
scalar sequence space and then to extend the existing results to include more general
sequence spaces X.
(c) Exposing that recent studies in connection with "absolutely summing multipliers"
and " sequences in the range of a vector measure" are intertwined, we intend to extend
the concept of "absolutely summing multiplier" to more general types of "summing multipliers"
and to apply our work to consider properties of Banach space valued operators
on scalar sequence spaces.
What are our contributions in this thesis?
* Firstly, we introduce an operator ideal approach which seems to provide a natural
setting in which to consider the existence of projections from L(X, Y)* onto K(X, Y)* and
derive necessary and sufficient conditions for the reflexivity of K(X, Y) in the absence
of the approximation property. Thus we simplify the proofs of existing results in the
literature and also generalise these results to such an extend that at least the well known
spaces without the approximation property are included.
* Secondly, in line with modern trend to provide proofs for theorems about operators
on Banach spaces which do not rely on the approximation properties, we expand the
concept of conjugate ideal to introduce the operator dual space of spaces of bounded
linear operators. It turns out that the operator dual space is a handy tool to study
inclusion theorems for spaces of operators. Also, using operator dual techniques, we are
able to prove existing characterisations of continuous dual spaces of important classes of
operators without relying on the continuity of the trace functional with respect to the
nuclear norm - thus the proofs do not depend on the approximation property.
* Thirdly, we provide a direct and easy proof of a known result which provides necessary
and sufficient condition for all weak p-summable sequences in a Banach space to be norm
null. Our proof uses sequence space arguments only, thereby allowing us to extend the
proof to more general sequence spaces, including certain Orlicz sequence spaces. Applying
these results, together with some known characterisations of operators on sequence spaces
in terms of vector sequence spaces, we succeed on the one hand to provide easier proofs
for existing results about necessary and sufficient conditions for the equalities K(1!.P, X) =
L(1!.P,X) and K(co, X) = L(co, X) and on the other hand to obtain further improvements.
* An absolutely summing multiplier of a Banach space X is a scalar sequence (ai) such
that (aixi) is absolutely summable in X for all weakly absolutely summable sequences (Xi)
in X. Recently there were several papers by Spanish mathematicians about sequences in
the range of a vector measure. We expose the fact that these concepts are intertwined and
thereby show that various results in one of the papers about sequences in the range of a
vector measure can easily be obtained, using the absolutely summing multiplier concept.
In the last chapter of the thesis we generalise the idea of absolutely summing multiplier
to that of p-summing multiplier, A- summing multiplier and even more general, (A, E)-
summing multiplier and use these concepts to obtain results about Banach space valued
bounded linear operators on A.
Donation
Fri, 01 Jan 1999 00:00:00 GMThttps://repository.maseno.ac.ke/handle/123456789/43421999-01-01T00:00:00ZUtility-Based Agent for Vehicle Driver Behaviour Modelling
https://repository.maseno.ac.ke/handle/123456789/4007
Utility-Based Agent for Vehicle Driver Behaviour Modelling
IMENDE, James Obuhuma
Knowledge on driver behaviour is a major factor that can possibly aid in future strategies for minimising if not fully controlling road fatalities. The behaviour of human vehicle drivers is the main cause of road accidents and is also the factor which has so far proved to be the most difficult to establish and model. Studies conducted on driver behaviour modelling have been limited by five factors: study methodology; vehicle model compatibility; cost; overestimation of critical driving events; and scope for driver behaviour monitoring. Probabilistic reasoning and intelligence, which are critical in modelling under stochastic environments are lacking in the applied methodologies. Fortunately, a combination of computing and communication technologies now makes it possible to model the behaviour of drivers operating in complex environments. The main objective of the research was to model human vehicle driver behaviour using a utility-based agent. To realise this objective, the research identified parameters that describe the behaviour of a human vehicle driver operating under diverse environments, formulated a vehicle driver behaviour dataset and developed and evaluated a vehicle driver agent that can operate in a complex environment. A sample of 30 drivers was used, with tonnes of data collected and analysed. Vehicle position coordinates, speed, direction, altitude, time and a reflected signal signifying the presence of an obstacle were collected using the Global Positioning System (GPS) comprising of satellites, GPS receivers and a server. Data analysis generated a driver behaviour dataset that was used in the preparation of the driver agent through three main phases: training, validation and testing. The driver agent was founded on Mixture Models with Bayesian inferencing techniques that performed driver behavioural pattern recognition and predictive analyses. The agent’s actions under dynamic conditions were evaluated against sets of performance standards, yielding mean success rates of over 68% accuracies and over 70% F-scores, +/- 5. This was an indicator of the appropriateness of the data collection tools and techniques, data analysis algorithms and the driver behaviour dataset. The significance of the study is three-fold. First, the function of the system could be extended to providing advisory services to drivers in real-time. Second, data gathered from the system could be used by road safety stakeholders to vet drivers and to diagnose causes of road accidents. Finally, the resulting knowledge-base could establish standards of rationality in driving and/or formulate rules for use in driverless vehicle control systems.
Wed, 01 Jan 2020 00:00:00 GMThttps://repository.maseno.ac.ke/handle/123456789/40072020-01-01T00:00:00ZA Model of the Relationships Between Availability Mechanisms and Outage Sources in Cloud Computing
https://repository.maseno.ac.ke/handle/123456789/3635
A Model of the Relationships Between Availability Mechanisms and Outage Sources in Cloud Computing
MBOGHOLI, JOHN SAUL MSAGHA
ABSTRACT
The use of cloud computing has been growing exponentially since its inception. Availability of the cloud, however, has been a problem for users and Cloud Service Providers (CSPs) alike; outages have been on the rise. This problem could be attributed to the fact that engineers building Availability Mechanisms (AMs) and those studying outage causes do not work collectively. The general objective of the study was to develop and evaluate an availability mechanism model for service outages in cloud computing environments. The study specifically sought to: identify the causes of outages in cloud computing environments, identify availability mechanisms in use in cloud computing environments, formulate a model that establishes correspondences between AMs and outages, and evaluate the performance of the model by measuring its service availability levels in cloud computing environments in relation to the settings of the cloud computing system parameters. A model was developed called the Ferris Wheel of Availability (FWA) model. The model was developed by relating AMs to outage causes, with AMs being conjugate in nature in relation to the respective outage causes. There were seven categories of AMs and seven categories of outage causes; AMs were categorized as cluster management, component redundancy, limit detection policy, checkpointing, node management, Active-X variant and fault tolerance. Outage causes were categorized as configuration issues, hardware issues, resource exhaustion, security issues, node failures, network issues and natural disasters. Testing of the model was done using CloudSim, a discrete, deterministic simulator that allows users to set up their customized configurations and run them in it. The simulator was configured to run each outage cause individually and the applicable AMs were then injected simultaneously and output recorded. Each outage cause had two AMs, and the findings confirmed the effectiveness of the proposed model structure in increasing service availability at infrastructure level. Key findings were that checkpointing is not effective as an AM against resource exhaustion, and that effective management of a cluster results in effective management of the nodes in it. It was not conclusive as to whether limit detection policy was effective as an AM against security issues. The study also suggested that limit detection policy be renamed limit prevention policy. The study introduced a new availability parameter called execution availability and recommended its use together with service availability in predicting overall availability at infrastructure level. The key contributions of the study were: development of the FWA model that establishes correspondences between AMs and outage causes since a model that establishes these correspondences had not been developed before; discovery of the relationships between AMs and outage causes based on simulation tests and consequent analysis; and introduction of an availability parameter called execution availability that measures the ratio of tasks allocated versus tasks executed. It is recommended to study the feasibility of merging two or more simulators to achieve results which were inconclusive using one simulator; an extension to the simulator in use may also be investigated. The use of the FWA model at CSP level is also recommended as it assists analysts and developers to build for availability from the very foundation as opposed to adapting a wait-and-see attitude in countering outages as they occur. The outcome of the study points to suggest that the application of the FWA model in a cloud computing infrastructure has the potential to increase availability in the cloud.
Wed, 01 Jan 2020 00:00:00 GMThttps://repository.maseno.ac.ke/handle/123456789/36352020-01-01T00:00:00Z