Weber and Lionel

et al. 5 discusses a technique for developing Dynamic Bayesian Networks (DBN)

to formalize the need of solving complex dynamic models. Also to analyze the

dependability estimations obtained by the proposed DBN model and the classical

Markov Chain model, a small valve system is used. This paper introduces the DBNs

as an equivalent model to Markov Chain. The issues considered in Weber and

Lionel et al. 5 are those involving systems whose flow can be demonstrated as

stochastic procedures and where the decision maker’s activities impact the

system behavior. Here the probability distribution over the next states are

determined by the current system state and the action applied to the system.

Due to the uncertainty, the paper considers the process state to be a random

variable that takes its value from a finite state space comparing to the

possible process state. The Markov Chain permits demonstrating the elements of

successions taken by these states. To compute the probability distribution over

the states at a certain time, it uses a compact network form, which is based on

iterative inference. Here, the notion of time is introduced through inference.

The paper states that the proposed method, based on the DBN theory, allows

building DBN structures for demonstrating of the transient development of

complex systems and this proposal can be the solution to model the reliability

of the complex systems. It also states that the acquired model is significantly

more compact and readable than the processes modelled using the Markov Chain.

The major aspect that Weber and Lionel et al.

5, failed to show/talk about is that the proposed methodology fails to

define how the learning calculations of Bayesian Network can add to show the

progression of the system dependability and how the parameters conduct can be

then demonstrated.

Carmona et al.

6 presents a student model to address the Learning system based on the model

proposed by Felder and Sylverman, and implementing the system using Dynamic

Bayesian Networks. The model proposed here is instated as per the outcomes from

the student in the Index of learning Style Questionnaire, and afterward tweaked

amid the course of the interaction using Bayesian model. Finally the model is

used to group objects in repository as suitable or not for a particular

student. The paper tries to resolve the issue from the Learning Object Metadata

(LOM). LOM is a standard to indicate the language structure and semantics of

learning objects utilizing a set of that completely/sufficiently portray a

learning question. The issue with LOM is the retrieval and searching facilities

of learning objects 6. To resolve this issue, Carmona et al. 6, proposes to

“filter” and “sort” the learning objects according to the current student’s

learning and preferences. This paper uses a hybrid approach i.e. combining the

Index of Learning Style Questionnaire (ILSQ) proposed by Felser and Soloman,

and then adding the dynamic Bayesian network (DNB) to further process the

outcome from the ILSQ. The DNB is initialized as the student’s model and the

scores from ILSQ are used as the initial beliefs. Therefore, at whatever point

a new evidences about the preferences of the student arrives, a new time slice

of the Dynamic Bayesian Network is instantiated 6, consequently setting off

the propagation mechanism and updating the beliefs for the Learning System. By

doing this, Carmona et al. 6 claims that it refines the initial values for

each student in the learning system that was acquired from the ILSQ as the

student interacts with it.

Pavlovic et al.

7 presents a model for figure motion analysis based on a novel Dynamic

Bayesian Network (DBN) using Switching Linear Dynamic System (SLDS). The key

component presented in this paper is an inexact Viterbi inference strategy for

defeating the immovability of correct inference in blended state DBNs. The

approach used here for dynamic modelling is an approach that is taken from the

field of biomechanics, which has also been utilized effectively to create

computer designs liveliness and to track abdominal area movement in a user-interface

(UI) setting. The drawback of this model is that it is difficult to track large

objects which is hard to measure using only visual data and may not reduce the

complexity of the model to exploit this confined core interest. To overcome

this difficulty, Pavlovic et al. 7 proposes an alternative strategy for

learning dynamic models from a preparation corpus of watched state space

directions. Finally, this paper illustrate the use of the SLDS structure to

demonstrating figure elements, where it demonstrates the learning of exchanging

models of fronto-parallel walking and running movement from video information.

Also it derives a mixed state version of the Viterbi estimation calculation for

induction in DBNs.

The context-aware

recommender systems (CARS), Adomavicius and Alexander et al. 9 create more

pertinent suggestions by adjusting them to the particular logical circumstance

of the user. The work presented in Adomavicius and Alexander et al. 9

investigates on how relevant data can be utilized to make keen and helpful

recommender systems. It gives a review of the multifaceted thought of setting,

talks about a few methodologies for consolidating logical data in the proposal

process, and outlines the use of such methodologies in a few application regions

where different types of contexts are exploited. The context in the recommender

systems is based on assumption of certain contextual factors such as time,

location, etc. that distinguish the context in which the recommendations are

given. It also assumes that each of these relevant components can have a

structure; the Time factor, for instance, can be characterized in wording of

seconds, minutes, hours, days, months, and years. “What a recommender system

may know about contextual factors”, and “How contextual factors change over

time” are the two aspects of contextual factors that Adomavicius and Alexander

et al. 9, is based on for the

classification of context in recommender systems.

Park, Ji-Oh and

Sung-Bae el al. 10, uses fuzzy

Bayesian Network to propose a context-aware music recommendation system. In

Context-aware music recommendation system, Park, Ji-Oh and Sung-Bae el al.

10, they are using fuzzy Bayesian networks and utility theory. The

context-aware music recommendation system (CA-MARS) exploits fuzzy system to

incorporate diverse source of information. While providing recommendation to

user, context also plays vital role for any decision system (example:

Temperature, Humidity, Noise, Season etc.). The problem with Bayesian Network

is it cannot deal with the diverse information. There is a possibility of loss

of information and it may not reflect the context appropriately. To overcome

this limitation, Park, Ji-Oh and Sung-Bae el al. 10 are proposing fuzzy

system.