Abstract: Robot swarm is a solution for difficult and large scale tasks. However controlling and coordinating a swarm of robots is a challenge because of the complexity and uncertainty of the environment where manual programming of robot behaviours is often impractical. In this study we proposed a hyper-heuristic methodology for swarm robots. It allows robots to create suitable actions based on a set of low-level heuristics. Each heuristic is a behavioural element. With online learning, the robot behaviours can be improved during the executions by autonomous heuristic adjustment. The proposed hyper-heuristic framework is applied to building surface cleaning tasks where multiple separate surfaces exist and the complete surface information is difficult to obtain. Under this scenario, the robot swarm need not only to clean the surfaces efficiently by distributing the robots, but also to move across surfaces by self-assembling into a bridge structure. Experiment results showed the effectiveness of the hyper-heuristic framework as a group of robots were able to autonomously clean multiple surfaces of different layouts without prior programming. Their behaviours can improve over time because of the online learning mechanism.

Abstract: The global growth in urbanisation increases the demand for services including road transport infrastructure, presenting challenges in terms of mobility. In this scenario, optimising the exploitation of urban road network is a pivotal challenge, particularly in the case of unexpected situations. In order to tackle this challenge, approaches based on mixed discrete-continuous planning have been recently proposed and although their feasibility has been demonstrated, there are a lack of informative heuristics for this class of applications. Therefore, existing approaches tend to provide low-quality solutions, leading to a limited impact of generated plans on the actual urban infrastructure. In this work, we introduce the Time-Based heuristic: a highly informative heuristic for PDDL+ planning-based urban traffic control. The heuristic, which has an admissible and an inadmissible variant, has been evaluated considering scenarios that use real-world data.

Abstract: In addition to the news content, most news web pages also contain various noises, such as advertisements, recommendations, and navigation panels. These noises may hamper the studies and applications which require pre-processing to extract the news content accurately. Existing methods of news content extraction mostly rely on non-content features, such as tag path, text layout, and DOM structure. However, without considering topics of the news content, these methods are difficult to recognize noises whose external characteristics are similar to those of the news content. In this paper, we propose a method that combines non-content features and a topic feature based on Dempster-Shafer (DS) theory to increase the recognition accuracy. We use maximal compatibility blocks to generate topics from text nodes and then obtain feature values of topics. Each feature is converted into evidence for the DS theory which can be utilized in the uncertain information fusion. Experimental results on English and Chinese web pages show that combining the topic feature by DS theory can improve the extraction performance obviously.

Abstract: People use the Internet to shop, access information and enjoy entertainment by browsing web sites. At the same time, cyber-criminals operate malicious domains to spread illegal information and acquire money, which poses a great risk to the security of cyberspace. Therefore, it is of great importance to detect malicious domains in the field of cyberspace security. Typically, there are broad research focusing on detecting malicious domains either by blacklist or exploiting the features via machine learning techniques. However, the former is infeasible due to the limited crowd, and the later requires complex feature engineering. Different from most of previous methods, in this paper, we propose a novel lightweight solution named DomainObserver to detect malicious domains. Our technique of DomainObserver is based on dynamic time warping that is used to better align the time series. To the best of our knowledge, it is a new trial to apply passive traffic measurements and time series data mining to malicious domain detection. Extensive experiments on real datasets are performed to demonstrate the effectiveness of our proposed method.

Abstract: Domain squatting is a speculative behavior involving the registration of domain names that are trademarks belonging to popular companies, important organizations or other individuals, before the latters have a chance to register. This paper presents a specific and unconcerned type of domain squatting called “AbbrevSquatting”, the phenomena that mainly happens on institutional websites. As institutional domain names are usually named with abbreviations(i.e., short forms) of the full names or official titles of institutes, attackers can mine abbreviation patterns from existed pairs of abbreviations and full names, and register forged domain names with unofficial but meaningful abbreviations for a given institute. To measure the abuse of AbbrevSquatting, we first mine the common abbreviation patterns used in institutional domain names, and generate potential AbbrevSquatting domain names with a data set of authoritative domains. Then, we check the maliciousness of generated domains with a public API and seven different blacklists, and group the domains into several categories with crawled data. Through a series of manual and automated experiments, we discover that attackers have already been aware of the principles of AbbrevSquatting and are monetizing them in various unethical and illegal ways. Our results suggest that AbbrevSquatting is a real problem that requires more attentions from security communities and institutions' registrars.