Honeybees are the most specialized insect pollinators and are critical not only for honey production but, also, for keeping the environmental balance by pollinating the flowers of a wide variety of crops. Recording and analyzing bee sounds became a fundamental part of recent initiatives in the development of so-called smart hives. The majority of researches on beehive sound analytics are focusing on swarming detection, a relatively simple binary classification task (due to the obvious difference in the sound of a swarming and a non-swarming bee colony) where machine learning models achieve good performance even when trained on small data. However, in the case of more complex tasks of beehive sound analytics, even modern machine learning approaches perform poorly. First, training such models would need a large dataset but, according to our knowledge, there is no publicly available large-scale beehive audio data. Second, due to the specifics of beehive sounds, efficient noise filtering methods would be required, however, we could not find a noise filtering method that would increase the performance of machine learning models substantially. In this paper, we propose a dynamic noise filtering method applicable on spectrograms (image representations of audio data) which is superior to the most popular image noise filtering baselines. Further, we introduce a multi-class classification task of bee sounds and a large-scale dataset consisting of 10.000 beehive audio recordings. Finally, we provide the results of a large-scale experiment involving various combinations of audio feature extraction and noise filtering methods together with various deep learning models. We believe that the contributions of this paper will facilitate further research in the area of (beehive) sound analytics.

Living Lab, one of the recent emerging smart city concepts, faces long-term sustainability challenges associated with its complexity and breadth of use. To be efficient, it must rely on comprehensive set of information distributed appropriately among all stakeholders to unleash its full innovation potential. This is especially true in the case of positive energy districts, where timely data dissemination is essential for prosumager decisions and their greedy behaviour. This paper interconnects intelligent information exchange, supported by ultra-low latency hybrid access network infrastructure, with the clever use of available fog computing resources to properly disseminate complex energy details to all participating entities. As the optimal distribution of information using proper task offloading is the convergence problem, we recalled higher-order neural units that helped maintain computational and energy efficiency in conjunction with the preservation of the overall system stability. We have achieved a reliable hourly energy consumption prediction with a computationally very lightweight alternative to commonly used deep neural network approaches that can be deployed on available smart appliances with ease. The application and simulation were performed on the dataset provided by one of Europe’s smart city pioneers, where the prosumager positive energy district transition has already started.

Robots working in unstructured environments must be capable of sensing and interpreting their surroundings. One of the main obstacles of deep-learning-based models in the field of robotics is the lack of domain-specific labeled data for different industrial applications. In this article, we propose a sim2real transfer learning method based on domain randomization for object detection with which labeled synthetic datasets of arbitrary size and object types can be automatically generated. Subsequently, a state-of-the-art convolutional neural network, YOLOv4, is trained to detect the different types of industrial objects. With the proposed domain randomization method, we could shrink the reality gap to a satisfactory level, achieving 86.32% and 97.38%$\mathrm{{mAP}}_{50}$scores, respectively, in the case of zero-shot and one-shot transfers, on our manually annotated dataset containing 190 real images. Our solution fits for industrial use as the data generation process takes less than 0.5 s per image and the training lasts only around 12 h, on a GeForce RTX 2080 Ti GPU. Furthermore, it can reliably differentiate similar classes of objects by having access to only one real image for training. To our best knowledge, this is the only work thus far satisfying these constraints.

Nowadays, the Passive Optical Networks (PONs) technology is widely deployed in broadband access networks. This paper deals with the security issues of Gigabit PON (GPON) standardized by the International Telecommunications Union (ITU), namely, standard ITU-T G.984 that is widely implemented in Europe these days. We describe and analyze the security of this standard and show its security risks. In spite of that transmitted data are encrypted to provide their confidentiality on a multipoint fibre connection, session secret keys during their establishment can be observed by adversaries. To address this security flaw, we propose a key establishment protocol that securely sets the session secret keys between two communication parties in GPON. Furthermore, we provide the security analysis of the proposed protocol.

This work proposes a novel approach - personalized forecasting - to take into account the sequential effect in predicting student performance (PSP). Instead of using all historical data as other methods in PSP, the proposed methods only use the information of the individual students for forecasting his/her own performance. Moreover, these methods also encode the "student effect" (e.g. how good/clever a student is, in performing the tasks) and "task effect" (e.g. how difficult/easy the task is) into the models. Experimental results show that the proposed methods perform nicely and much faster than the other state-of-the-art methods in PSP.

Web search heuristics based on Fagin's threshold algorithm assume we have the user profile in the form of particular attribute ordering and a fuzzy aggregation function representing the user combining function. Having these, there are sufficient algorithms for searching top-k answers. Finding particular attribute ordering and aggregation for a user still remains a problem. In this short paper our main contribution is a proof of concept of a new iterative process of acquisition of user preferences and attribute ordering .

We present a middleware system UPRE enabling personalized Web search for users with different preferences. The input for UPRE is user evaluation of some objects in scale from the worst to the best. Our model is inspired by existing models of distributed middleware search. We use both inductive and deductive tasks to find user preferences and consequently best objects