Real-time communication is a requirement in various application fields. However, developing such a determinis- tic communication requires low level, deep and advanced knowledge about hardware, real-time operating systems, real-time communication protocols and real-time program- ming. There are numerous real-time communication technolo- gies , , , , ,  which bring their specific properties such as specific hardware, software and infrastruc- tural requirements. The required expert knowledge decelerates the development in cross-domainapplications with real-time communication requirements. For example, surgical robotics in Networked Medical Systems (NMS)  requires deterministic system behavior. Common used communication systems in NMS are e.g. High Definition Multimedia Interface (HDMI) for sending and receiving high quality video data, Transmission Control Protocol/Internet Protocol (TCP/IP)  for exchanging non-real-time data between medical systems and IT systems such as Digital Imaging and Communications in Medicine (DI-
Retrieving a set of documents of the exact same domain of the target documents can often be difficult. As discussed previously, a solution would be to manually label some target documents, but creating a suitable train- ing set may require to label a great amount of documents, thus implying a significant amount of human effort. However, in some cases, a set of labeled documents of a slightly different domain may be available. The difference between the domains may consist in the use of some different terms or in the organization within categories representing slightly different concepts. Considering the general setting of the problem, the traditional techniques seen so far for text categorization might be applied to infer a classifier from available training documents and apply it to target documents. How- ever, due to the differences between the two domains, this would likely not result in an accurate classification, as many of the features known by the classifier would not be found within the target documents. Cases like these would require specific methods to somehow transfer the knowledge extracted from the available training data to the target domain. Through- out the last decade, techniques for transfer learning have been devised to address these cases (Pan and Yang, 2010). Transfer learning generally in- volves solving a problem in a target domain by leveraging knowledge from a source domain whose data is fully available. Cross-domain text catego- rization refers to the specific task of classifying a set of target documents in predefined categories using as support a set of pre-labeled documents of a slightly different domain.
Abstract. The simplicity of using Web 2.0 platforms and services has resulted in an abundance of user-generated content. A signifi- cant part of this content contains user opinions with clear economic relevance - customer and travel reviews, for example, or the articles of well-known and respected bloggers who influence purchase deci- sions. Analyzing and acting upon user-generated content is becoming imperative for marketers and social scientists who aim to gather feed- back from very large user communities. Sentiment detection, as part of opinion mining, supports these efforts by identifying and aggregat- ing polar opinions - i.e., positive or negative statements about facts. For achieving accurate results, sentiment detection requires a correct interpretation of language, which remains a challenging task due to the inherent ambiguities of human languages. Particular attention has to be directed to the context of opinionated terms when trying to re- solve these ambiguities. Contextualized sentiment lexicons address this need by considering the sentiment term’s context in their evalu- ation but are usually limited to one domain, as many contextualiza- tions are not stable across domains. This paper introduces a method which identifies unstable contextualizations and refines the contex- tualized sentiment dictionaries accordingly, eliminating the need for specific training data for each individual domain. An extensive eval- uation compares the accuracy of this approach with results obtained from domain-specific corpora.
Cookies represent an important element of HTTP providing state management to an otherwise stateless protocol. HTTP cookies currently in use are governed by the same origin policy that directs Web browsers to allow cookie sharing only between Web sites in the same DNS domain. As Web applications get richer, data sharing across domain boundaries becomes more important. While practical solutions to cross-domain data sharing exist, in many cases they increase complexity and cost. In this paper we propose a simple mechanism to share cookies using authorizations based on X.509 attribute and public key certiﬁcates. In addition to supporting secure cookie sharing between unrelated domains, it can be beneﬁcial for hosts in the same domain when the currently used same origin policy is deemed too permissive, exposing cookies to leakage and spooﬁng.
we propose a domain extraction tool, Dexter 1 , that extracts a domain-specific corpus from Wikipedia. The algorithm, elaborated in Section 2, retrieves a set of documents from Wikipedia that are closest in discipline to a user-supplied small seed corpus. The size of this extracted set is a user-defined hy- perparameter, and thus controls the trade-off be- tween the specificity of the output corpus and its size. We empirically determine the favorable con- figuration of Dexter, demonstrate its benefits to- wards estimating word embeddings, and conse- quently distractor generation, as well as language models. We also show how, on the aformentioned tasks, Dexter outperforms BootCaT, a popular toolkit to automatically create an Internet-derived corpus (Baroni and Bernardini, 2004). Datasets used in this research are released for public use 2 .
Given a feature space and a set of annotated objects defined in that space, a new ontology is automatically generated to describe domain knowledge. This is realized by considering the spatial arrangement of the objects in the feature space, using the concepts defined in the previously described semantic model. The generated ontology will have two parts; a fixed part holding fundamental concepts regarding the application domain and the objects, and a dynamically generated part holding the generalized spatial relations between the objects. The fixed part of the concept hierarchy in the automatically generated ontology consists of the class Core Elements, which is superset of all classes in the automatically generated ontology, and four main subclasses of Core Elements (Fig. 1c):
(Goode, 1960; Piotrowski, 1979) postulate that non-work conflict with work increases experienced relationship stress, thereby producing negative effects on mental health (and presumably life satisfaction). Non-work conflict with work is assumed to occur when roles played in one life domain (e.g., work) negatively impact roles played in another (e.g., family) either because of role overload (i.e., not enough time to meet all role demands) or because of conflict in the context required of two different roles. For example, a Professor may feel the need to be task-oriented with subordinates at work but relationship oriented with a Child who is to celebrate the tenth birthday and wants a bash (Adekola, 2006). While the presence of role overload can be readily measured by the amount of time conflict experienced between work and non-work responsibilities, more effects that are indirect will most likely indicate conflict between role context requirements. One such indicator of conflict in role context requirements may be the amount of stress present in personal relationships. That is, personal relationships may suffer when the behaviors required in one role are in conflict with, and subsequently spillover to, behaviors required in another role.
In most applications of machine learning, domain experts provide domain specific knowledge. From pre- vious experience it is known that domain experts are unable to provide all relevant knowledge in advance, but need to see some results of machine learning first. Interactive machine learning, where experts and ma- chine learning algorithm improve the model in turns, seems to solve this problem. In this position paper, we propose to use arguments in interaction between machine learning and experts. Since using and under- standing arguments is a practical skill that humans learn in everyday life, we believe that arguments will help experts to better understand the models, facilitate easier elicitation of new knowledge from experts, and can be intuitively integrated in machine learning. We describe an argument-based dialogue, which is based on a series of steps such as questions and arguments, that can help obtain from a domain expert exactly that knowledge which is missing in the current model.
To meet these requirements our domain model uses Ge- ographic Information Systems (GIS) support to allow state determination to reflect sensor and actuator position and spread. The domain model provides a uniform and scalable method of generating observation models that measure the quality of the sensor infrastructure and allows us to deter- mine where noise tolerant planning and optimisation strate- gies are appropriate.
BOARD corpus can be used to classify utterances in a new domain, that of the AMITI ES GE ´ corpus. We achieve almost 80% of the upper baseline performance over the AMITI ES ´ GE corpus, when judged using our lenient scoring mecha- nism - scoring 55.7% using the cross-domain cues, com- pared to the 70.8% when using in-domain cues. When us- ing the strict measure we still achieve around 60% of the upper bound performance, both results being a substan- tial improvement over the baseline measure of 20%, cor- responding to the most frequent tag. This is a significant result, which confirms the idea that cues can be sufficiently general across domains to be used in classification. However, whilst the experiment using SWITCHBOARD cor- pus derived cues to classify AMITI ES GE ´ data works well, the same is not true in reverse. There are two possible ex- planations for this result. It could be related to the size of data available for training, although our experiments in this area seem to suggest otherwise. We believe that the com- position of the training data is a more crucial element. The fact that SWITCHBOARD corpus data is not domain specific, and, although the DA distribution in this corpus is skewed, it contains enough data for the major classes to be effective on new data. Although the AMITI ES GE ´ contains a lot of ques- tions and statements, there is very little of the other signif- icant categories, such as <backchannels>, a key DA in the
As a step forward, in this paper we take into consideration the possible effects of the backhaul segment in the mobile infrastructure. In this case, the decision making process is not only driven by the service configuration and the experienced performance in the radio link, but it also takes into account the most suitable adaptation actions in order to overcome possible congestion situations in the mobile backhaul. This new functionality is stated in this paper as a cross– domain feature, assuming that the radio and wired segments of a mobile network are considered as two different domains.
Cross-lingual For cross-lingual experiments, we tune our system’s parameters by training a sys- tem on the data for three languages with sufficient amounts of data (namely, German, Spanish and Brazilian Portuguese), and using English data as a development set. We then train a new model also using multi-task learning (with these tuned param- eters) using only source training data, and report performance on the target test set. This allows us to estimate performance when no data is available for the language of interest (unsupervised adapta- tion).
linguistic uncertainty are strongly domain-specific. In other words, the linguistic means to express them may change from domain to domain. For instance, the abbreviated form of szerintem “I think”, sztem is very often used in social media texts as a doxastic cue but it is never used in its short form in standard texts. Thus, adding in-domain data to the training set might provide examples of such cases typical of social media language use. Also, it should be noted that precision values are relatively high for doxastic and epistemic cues even in the cross-domain settings. This might be related to the fact that these types of uncertainty cues occur rarely in social media texts and even if they occur, they are mostly different from the linguistic means used in standard texts. So, the system is unable to identify many of such cues based on the training data but when it marks one cue as doxastic/epistemic, it is most probably a true positive.
– We conduct comprehensive experiments on three large-scale real-world datasets from Foursquare, Yelp and Brightkite to evaluate the performance of CBT- based non-overlapping CDCF. Our experimental results demonstrate that CrossFire is not able to transfer useful knowledge from the source domain to the target domain for the venue recommendation task, being consistent with the previous study of . In particular, we find that the CBT-based technique does not clearly contribute to the improvements observed from CrossFire, compared to the traditional single-domain MF-based models. In- deed, through experiments conducted when equating the source and target domains, we show that such improvements may not be explained by trans- fer of knowledge between source and target domains. We postulate that, in fact, that the improvements are gained from the additional parameters introduced by CrossFire, which makes it more flexible than the traditional single-domain MF-based approaches. In addition, our experimental results on a state-of-the-art sequential-based deep learning venue recommendation framework of  further validate this result.
considers all vehicles in close proximity (see Sec. 4) by a parametrized position range, mainly dependent on current vehicle state. All vehicles within that range which fulfill the current acceptance policy becomes a node, i.e. a master in NFN. All messages are considered for evaluation, since they potentially contain information for safety-critical situa- tions and applications, e.g. CACC. In contrast, FFN uses a cluster-based approach, where vehicles beyond the NFN area are clustered and are represented by a cluster master which in turn has its own NFN. Since information from FFN is potentially used for long-term optimizations or warnings (e.g. premature reaction to an accident) only messages sent by FFN masters which fulfill the current acceptance policy are considered for evaluation. Messages from slaves of the individual clusters can be discarded inherently. Masters are determined by simple First Come First Serve principle. This does not limit the reception of event-driven messages, e.g. in case of a remote accident, since every master has its own NFN and could forward those event-driven messages by the WSM Scheduler as described in Sec. 4. Additionally, our hybrid cluster approach is completely decentralized as it only relies on GPS, speed and direction data which is part of a BSM encapsulated in a WSM. Furthermore the cluster approach is simple to implement and reduces computational complexity, which is a prerequisite in order to meet real-time constraints.
the distance of two domains. The proxy was defined by 2(1 − 2), where is the generalization error of a linear SVM classifier trained on the binary classification problem to distinguish inputs between the source and target domains. Figure 4 shows the results of each pair of domains. We observe several trends: Firstly, the proxy A-distance of within-group domain pairs (i.e., BD and EK) is consistently smaller than that of the cross-group domain pairs (i.e., BK and DE) on all of the hidden spaces. Secondly, the proxy A-distance on the domain specific space is consistently larger than its corresponding value on the hidden space of SO model, as expected. While the proxy A-distance value on domain invariant space is generally smaller than its corresponding value on the hidden space of SO model, except for BK domain pair. A possible explanation is that the balance of classification loss and domain discrepancy loss makes there is still some target domain specific information in the domain invariant space, introduced by the target unlabeled data.
Corpus-Based similarity includes Hyperspace Analogue to Language (HAL) , Latent Semantic Analysis (LSA). In general text mining involves extractions of terms, concept using Bag of Words (BOW) or N-Gram so that each document can be represented by vector space. Thus a matrix is created and machine learning algorithms are applied over here. If matrix columns are large then we apply technique of mathematics called Singular Value Decomposition is reduce number of features maintaining similarity structure among rows. Other techniques are Generalized Latent Semantic Analysis (GLSA) , Explicit Semantic Analysis (ESA), The cross-language explicit semantic analysis (CLESA), Pointwise Mutual Information - Information Retrieval(PMI-IR), Second-order co-occurrence Pointwise mutual information (SCO-PMI) ,Normalized Google Distance (NGD).