Abstract Locating multiple objects has become an important task in multimedia research and applications due to the common nature of real-world images. Object localization requires a large number of visual annotations, such as bounding boxes or segmentation, but the annotation process is labour-intensive and sometimes inextricable for human experts in complex domains such as manufacturing and medical fields. Moving beyond single object localization, this paper presents a weakly semi-supervised learning framework based on Graph Transformer Networks using Class Activation Maps to LOcate Multiple Objects (GraphLOMO) in images without visual annotations. Our method overcomes the computational challenges of gradient-based CAM while integrating topological information and prior knowledge into object localization. Moreover, we investigate the higher order of object inter-dependencies with the use of 3D adjacency matrix for better performance. Extensive empirical experiments are conducted on MS-COCO and Pascal VOC to establish a suitable performance measure and baselines, as well as a state-of-the-art for weakly semi-supervised multi-object localization.

With the emergence of next generation networks, Network Traffic Classification (NTC) has seen greater importance in network management and security. Recently, Channel State Information (CSI) based WiFi sensing techniques have shown their potential for NTC applications [1], [2] as a privacy-preserving yet effective tool. As CSI could be prone to interference, this paper examines the performance of CSI-based NTC models under interference-induced adversarial attacks. Specifically, the impact of spectral allocation of the interference, underlying interfering network traffic type, and physical location of the interference are studied and quantified. We conducted experiments using off-the-shelf devices to test the NTC performance, with and without the adversarial interference attack of ping, buffered video streaming, and live video streaming network traffics. Subsequently, we found that spectral allocation of the attacking interference and the underlying traffic types of interference could be used to deceive the established NTC model, and different network traffic types show different robustness across the interference cases. Namely, ping suffers the most in the spectrally manipulated attack, with the classification accuracy down to as low as 23.2%, whereas Twitch might be completely misidentified as other traffic in underlying traffic type controlled attack.

Harnessing human computation through crowdsourcing offers an alternative approach to solving complex problems, especially those that are relatively easy for humans but difficult for computers. Micro-tasking platforms such as Amazon Mechanical Turk have attracted large, on-demand work force of millions of workers as well as hundreds of thousands of job requesters. Achieving high quality results by putting humans in the loop is one of the main goals of these crowdsourcing systems. We study the effects of different user interface designs on the performance of crowdsourcing systems. Our results indicate that user interface design choices have a significant effect on crowdsourced worker performance.

As Cloud Computing adoption by enterprise customers grows, so too the need for optimal utilisation of their virtual resources. Likewise, cost pressures on cloud providers with a utility business model e.g. Amazon Web Services, would also need to optimise the utilisation of their physical infrastructure. Clearly, the ability to predict demand would be valuable. We introduce Boost red, an automatic demand predictor for the cloud. BoostPred's design goals are to require no human expert intervention in making accurate predictions from noisy real world demand signals. We evaluate the accuracy of Boost red using noisy real-world signals which reveal its potential and current shortcomings.

It is well-known that software failure data often contain noise, making the reliability estimation problematic. In particular, the kind of data noise inherent in software failure data is biased. There is no upper bound for the value of a noisy data point, but there is a lower bound of zero. This may lead to over-optimistic estimation of the reliability when using maximum likelihood or least square methods based on standard software reliability models. We attempt to address this problem by modeling software failure data using machine learning techniques such as support vector machine regression and generalized additive models, which have mechanisms that are capable of dealing with data noise. We then analyse the results from machine learning modeling, and compare them to that of some generalized linear modeling techniques that are equivalent to standard software reliability models. The validity of the machine learning modeling of noisy software failure data is evaluated through this comparison.

Weblog has quickly evolved into a new information and knowledge dissemination channel. Yet it is not easy to discover weblog communities through keyword search. The main contribution of this paper is the study of weblog communities from the perspective of social network analysis. We proposed a new way of collecting and preparing data for weblog community discovery. The data collection stage focuses on gaining knowledge of the strength of social ties between weblogs. The strength of social ties and the clustering feature of social network guided the discovery of weblog communities.

Here we propose an adaptive local pruning method for association rules. Our method exploits the exact mapping between a certain class of association rules, namely those whose consequents are singletons and backward directed hypergraphs (B-graphs). The hypergraph which represents the association rules is called an association rules network(ARN). Here we present a simple example of an ARN. We further prove several properties of the ARN and apply the results of our approach to two popular data sets.

Web services and service oriented architecture (SOA) represent a radical departure from traditional monolithic, tightly bound, customized applications with proprietary interfaces. Their range however is limited as yet by the very structure and standardization which support their interoperability. While they are able to support heterogeneity of technical components, their ability to support complex business interactions has yet to mature. This paper summarizes research in progress which asks how might the Web services model be mapped to other kinds of IT services and extended to account for services of greater human and organizational complexity through their lifecycle. The research proposes the construction of a generic IT service model and a taxonomy IT services to provide a frame of reference for analysis and evaluation for the extension of the Web services model and for a transition to a service oriented view of IT.