Search

This paper deals with the microtoponymy of the Berkovo village, which is situated on the left bank of Beli drim, in the area called Prekoruplje. Its accurate location is between Ozdrim and Pećka Bistrica confluence with the Drim. Research materials have been recorded by interviewing locals in the year of 1997 (Orthodox Serbs and Catholic Albanians). Some of the entries that are used in this paper, concerning village Berkovo, are taken from the work of Mileta Bukumirić called Onomastic of Prekoruplje (second part). The paper was published in Belgrade in the Onomatology contributions in 1982.

Fault analysis poses a serious threat to embedded security devices, especially smart cards. In particular, modeling faults and finding effective practical approaches that are also generic is considered to be of interest for smart card industry. In this work we propose a novel methodology to deal with a difficult question of choosing multiple parameters required for effective faults. To this aim, we investigate several algorithms and find a new promising direction using evolutionary computation. Our experimental results on some of the smart cards used today show the potential of this new approach. Our best algorithm is a tailored search strategy especially developed for the purpose of finding the best choice of parameters for glitching. With this approach we found some of off-the-shelf devices, although secured against this type of attacks, still vulnerable.

Task-based modeling with recurrent neural networks (RNNs) has emerged as a popular way to infer the computational function of different brain regions. These models are quantitatively assessed by comparing the low-dimensional neural representations of the model with the brain, for example using canonical correlation analysis (CCA). However, the nature of the detailed neurobiological inferences one can draw from such efforts remains elusive. For example, to what extent does training neural networks to solve common tasks uniquely determine the network dynamics, independent of modeling architectural choices? Or alternatively, are the learned dynamics highly sensitive to different model choices? Knowing the answer to these questions has strong implications for whether and how we should use task-based RNN modeling to understand brain dynamics. To address these foundational questions, we study populations of thousands of networks, with commonly used RNN architectures, trained to solve neuroscientifically motivated tasks and characterize their nonlinear dynamics. We find the geometry of the RNN representations can be highly sensitive to different network architectures, yielding a cautionary tale for measures of similarity that rely representational geometry, such as CCA. Moreover, we find that while the geometry of neural dynamics can vary greatly across architectures, the underlying computational scaffold---the topological structure of fixed points, transitions between them, limit cycles, and linearized dynamics---often appears universal across all architectures., Comment: Presented at NeurIPS 2019

Evolutionary computation methods are successfully applied in solving of combinatorial optimization problems. Since the “No Free Lunch” theorem states that there is no single best algorithm to solve all possible problems, throughout the years many algorithms and their modifications have emerged. When a new algorithm is developed, one question that naturally arises is how it compares to other algorithms, whether for some specific problem or in general performance. Because of the stochastic nature of systems involved, usually the only possible way of deriving the answer is to perform extensive experimental analysis. In this paper we provide an overview of possible approaches in the experimental analysis, and describe statistical methods that could be used. Furthermore, we outline similarities and differences between these methods, which lead to a discussion of important issues that need to be resolved when using these methods.

According to European Strategy for sustainable, competitive and secure energy, which guidelines are described in two documents: ‘’Green Paper: a European Strategy for Sustainable, Competitive and Secure Energy’’ and ‘’White Paper: Energy for the Future: Renewable Sources of Energy’’, it is predicted that share of renewable energy resources in total energy balance will raise from present 6% up to 15% until 2015. Croatia, as candidate country for EU admittance, with growing dependency upon imported energy because of continuous depletion of own energy resources, prior oil and gas, needs to follow EU strategic aims to achieve diversification of energy sources and implement and promote renewable energy resources. This paper presents strategy of geothermal resources development in Republic of Croatia for the period of 2007-2030 in cascade and cogeneration principle of energy utilization. These projections of geothermal energy development are part of comprehensive Strategy of Mineral Resources Development which is made by Faculty of Mining, Geology and Petroleum Engineering for Ministry of Economy, Labour and Entrepreneurship.

Optimization of geothermal energy production process means the minimization of all energy losses from the reservoir conditions to the user. As the available energy is being utilized mostly in the wellbore and in the surface equipment, process optimization requires scientific access including the extraction technology parameters. Specific energy on the geothermal wellhead is calculated for two possible cases. The first embraces only geothermal water production, while the other takes into account the saturated steam production as well. Each of these working conditions defines unambiguously designed pressure on the wellhead. The steam and water energy ratio, in function of predicted sink temperature for reinjection of geothermal water, points out the possibilities for commercialization of reservoir Velika Ciglena.

Recurrent neural networks (RNNs) are a widely used tool for modeling sequential data, yet they are often treated as inscrutable black boxes. Given a trained recurrent network, we would like to reverse engineer it--to obtain a quantitative, interpretable description of how it solves a particular task. Even for simple tasks, a detailed understanding of how recurrent networks work, or a prescription for how to develop such an understanding, remains elusive. In this work, we use tools from dynamical systems analysis to reverse engineer recurrent networks trained to perform sentiment classification, a foundational natural language processing task. Given a trained network, we find fixed points of the recurrent dynamics and linearize the nonlinear system around these fixed points. Despite their theoretical capacity to implement complex, high-dimensional computations, we find that trained networks converge to highly interpretable, low-dimensional representations. In particular, the topological structure of the fixed points and corresponding linearized dynamics reveal an approximate line attractor within the RNN, which we can use to quantitatively understand how the RNN solves the sentiment analysis task. Finally, we find this mechanism present across RNN architectures (including LSTMs, GRUs, and vanilla RNNs) trained on multiple datasets, suggesting that our findings are not unique to a particular architecture or dataset. Overall, these results demonstrate that surprisingly universal and human interpretable computations can arise across a range of recurrent networks., Presented at NeurIPS 2019