Many recent works have demonstrated that unsupervised sentence representations of neural networks encode syntactic information by observing that neural language models are able to predict the agreement between a verb and its subject. We take a critic
al look at this line of research by showing that it is possible to achieve high accuracy on this agreement task with simple surface heuristics, indicating a possible flaw in our assessment of neural networks' syntactic ability. Our fine-grained analyses of results on the long-range French object-verb agreement show that contrary to LSTMs, Transformers are able to capture a non-trivial amount of grammatical structure.
Multimodal Machine Translation (MMT) enriches the source text with visual information for translation. It has gained popularity in recent years, and several pipelines have been proposed in the same direction. Yet, the task lacks quality datasets to i
llustrate the contribution of visual modality in the translation systems. In this paper, we propose our system under the team name Volta for the Multimodal Translation Task of WAT 2021 from English to Hindi. We also participate in the textual-only subtask of the same language pair for which we use mBART, a pretrained multilingual sequence-to-sequence model. For multimodal translation, we propose to enhance the textual input by bringing the visual information to a textual domain by extracting object tags from the image. We also explore the robustness of our system by systematically degrading the source text. Finally, we achieve a BLEU score of 44.6 and 51.6 on the test set and challenge set of the multimodal task.
The insurer's obligation to declare insured risk data is the most important
obligation in the insurance contract. Based on this declaration, the
insured is able to determine all the terms of the contract.
In this research, we define the concept of visual saliency in biology
and how it is described in computer science using the concept of
saliency maps, and how to use these maps to detect salient
objects in digital images. We also conduct experiment
s using
several algorithms to detect salient objects, and describe how to
quantify the quality of the results using clear and well-defined
standards.
Jgroup integrates the object group paradigm with the distributed object model of Java
RMI, providing a platform which is suitable for developing partitionable distributed
applications. Jgroup depends on RMI in all its interactions; whether internal
for
coordination between object group replicas, or external for communicating clients with
object group. Because of the dynamic of network which is caused by joining new servers
and leaving another ones to object group, or caused by partitioning, Partitionable Group
Membership Service tracks this changes to provide each member with a report called view.
The view contains a list of members which can communicate and coordinate activities.
The advantage of group membership in Jgroup is the ability to continue in providing
service in each partition, instead of limiting it in one partition.
When partitions merge, State Merging Service of Jgroup constructs a new global
consistent state, to reconcile any divergence caused by conflict updates in the different
partitions.
Group Membership Service is required that a view is installed only after agreement
is reached on its composition among the servers included in the view (Agreement On View
property). To achieve this property; many of Estimation messages are exchanged between
the servers, which causes overhead on the network.
This article improves the performance of group membership algorithm which is
responsible for achieving the agreement, through allowing for the first server detects the
new change in membership to send its estimation to other servers, instead of doing that by
each server.
Results show that the enhanced algorithm reduces the number of exchanged estimate
messages, and takes approximately the same period of time to reach to agreement on view
as in the default algorithm.
In our research we offer detailed study of one of the data
mining functions within the text data using the object properties in
databases. It studies the possibility of applying this function on the
Arabic texts. We use procedural query language P
L / SQL that
deals with the object of Oracle databases.
Data mining model Has been built. It works on classification
of Arabic texts documents using SVM algorithm for indexing of
texts and texts preparation, Naïve Bayes algorithm to classify data
after transformation it into nested tables. So we made an evaluation
of the obtained results and conclusions.
This paper presents a method integrating database with Jgroup
based on Hibernate, which is one of Object Relational Mapping
tools. We compare between the performance of Jgroup integrated
with Hibernate and the performance of RMI integrated with
Hibernate. The results show that Jgroup/Hibernate outperforms
RMI/Hibernate when the number of clients increases.
The increasing reliance on network systems in day-to-day activities requires that they
provide available and reliable services. Jgroup provides available service through creating
multiple replicas of the same service on multiple devices. Jgroup ach
ieves reliable service
by maintaining the shared state between the replicas and coordinating their activities
through Remote Method Invocation. Unlike Jgroup, JavaGroups uses message passing to
implement coordination between the replicas.
In this paper, we compare Jgroup and JavaGroups for different Group Method
Invocation modes. These modes are Anycast and Multicast in Jgroup, GET_FIRST and
GET_ALL in JavaGroups.
This paper also improves the performance of ARM (Autonomous Replication
Management) which is embedded with Jgroup (Jgroup/ARM) for supporting fault
tolerance, through finding a new solution to handle group failure where all remaining
replicas fail in rapid succession. In this new solution, only one replica (the group leader)
issues renew events (IamAlive) periodically, instead of sending it by every replica in the
group, with taking the same period to discover group failure by Replication Manager.
Results of Comparison show that JavaGroups is faster than Jgroup when a single
replica is used, whereas Jgroup outperforms JavaGroups with increasing number of
replicas. The invocation delay in JavaGroups increases noticeably with increasing the size
of array passed into the invoked method which make JavaGroups unsuitable for
applications which require exchanging big sizes of data and use large number of servers,
whereas Jgroup is suitable for that.
Results show that the new proposal reduces the number of renew events to 37.5% at
most, and Jgroup/ARM takes approximately the same period of time to discover group
failure as in Meling solution.