GRAMOFON: General model-selection framework based on networks

Ensembles constitute one of the most prominent class of hybrid prediction models. One basically assumes that different models compensate each other's errors if one combines them in an appropriate way. Often, a large number of various prediction models are available. However, many of them may share similar error characteristics, which highly depress the error compensation effect. Thus the selection of an appropriate subset of models is crucial. In this paper, we address this issue. As major contribution, for the case if large number of models is present, we propose a network-based framework for model selection while paying special attention to the interaction effect of models. In this framework, we introduce four ensemble techniques and compare them to the state-of-the-art in experiments on publicly available real-world data.     

Model-based segmentation and tracking of head-and-shoulder video objects for real time multimedia services

A statistical model-based video segmentation algorithm is presented for head-and-shoulder type video. This algorithm uses domain knowledge by abstracting the head-and-shoulder object with a blob-based statistical region model and a shape model. The object segmentation problem is then converted into a model detection and tracking problem. At the system level, a hierarchical structure is designed and spatial and temporal filters are used to improve segmentation quality. This algorithm runs in real time over a QCIF size video, and segments it into background, head and shoulder three video objects on average Pentium PC platforms. Due to its real time feature, this algorithm is appropriate for real time multimedia services such as videophone and web chat. Simulation results are offered to compare MPEG-4 performance with H.263 on segmented video objects with respects to compression efficiency, bit rate adaptation and functionality.     

Development of a Green Process for the Preparation of Antioxidant and Pigment-Enriched Extracts from Winery Solid Wastes Using Response Surface Methodology and Kinetics

Red grape pomace (RGP), an abundant wine industry solid waste, was used for the recovery of polyphenols and anthocyanin pigments, using ultrasound-assisted extraction and water/glycerol mixtures as the solvent. Glycerol concentration (C_gl) and liquid-to-solid ratio (R_L/S) were first optimized by implementing Box?Behnken experimental design and the process was further studied through kinetics. The optimal conditions were found to be C_gl = 90% (w/v) and R_L/S = 90 mL g^?1, and under these conditions the extraction of total polyphenols (TP) and total pigments (TPm) obeyed first-order kinetics. Maximum diffusivity (D_e) values were 4.22 × 10^?12 and 12.59 × 10^?12 m² s^?1, for TP and TPm, respectively, and the corresponding activation energies were (E_a) 13.94 and 8.22 kJ mol^?1.     

Complexity distortion theory

Complexity distortion theory (CDT) is a mathematical framework providing a unifying perspective on media representation. The key component of this theory is the substitution of the decoder in Shannon's classical communication model with a universal Turing machine. Using this model, the mathematical framework for examining the efficiency of coding schemes is the algorithmic or Kolmogorov (1965) complexity. CDT extends this framework to include distortion by defining the complexity distortion function. We show that despite their different natures, CDT and rate distortion theory (RDT) predict asymptotically the same results, under stationary and ergodic assumptions. This closes the circle of representation models, from probabilistic models of information proposed by Shannon in information and rate distortion theories, to deterministic algorithmic models, proposed by Kolmogorov in Kolmogorov complexity theory and its extension to lossy source coding, CDT.     

Improving maximum margin matrix factorization

Collaborative filtering is a popular method for personalizing product recommendations. Maximum Margin Matrix Factorization (MMMF) has been proposed as one successful learning approach to this task and has been recently extended to structured ranking losses. In this paper we discuss a number of extensions to MMMF by introducing offset terms, item dependent regularization and a graph kernel on the recommender graph. We show equivalence between graph kernels and the recent MMMF extensions by Mnih and Salakhutdinov (Advances in Neural Information Processing Systems 20, 2008). Experimental evaluation of the introduced extensions show improved performance over the original MMMF formulation.     

Aerosol lidar intercomparison in the framework of the EARLINET project. 2. Aerosol backscatter algorithms

An intercomparison of aerosol backscatter lidar algorithms was performed in 2001 within the framework of the European Aerosol Research Lidar Network to Establish an Aerosol Climatology (EARLINET). The objective of this research was to test the correctness of the algorithms and the influence of the lidar ratio used by the various lidar teams involved in the EARLINET for calculation of backscatter-coefficient profiles from the lidar signals. The exercise consisted of processing synthetic lidar signals of various degrees of difficulty. One of these profiles contained height-dependent lidar ratios to test the vertical influence of those profiles on the various retrieval algorithms. Furthermore, a realistic incomplete overlap of laser beam and receiver field of view was introduced to remind the teams to take great care in the nearest range to the lidar. The intercomparison was performed in three stages with increasing knowledge on the input parameters. First, only the lidar signals were distributed; this is the most realistic stage. Afterward the lidar ratio profiles and the reference values at calibration height were provided. The unknown height-dependent lidar ratio had the largest influence on the retrieval, whereas the unknown reference value was of minor importance. These results show the necessity of making additional independent measurements, which can provide us with a suitable approximation of the lidar ratio. The final stage proves in general, that the data evaluation schemes of the different groups of lidar systems work well.     

Economics of Dynamic Pricing in a Reputation Brokered Agent Mediated Marketplace

We present a framework to study the microeconomic effects in a reputation brokered Agent mediated Knowledge Marketplace, when we introduce dynamic pricing algorithms. We study the market with computer simulations of multiagent interactions. In this marketplace, the seller reputations are updated in a collaborative fashion based on the performance of the user in the delegated tasks. To the best of our knowledge this is the first agent mediated marketplace where the agents use dynamic pricing based on dynamically updated reputations. The framework can be used to investigate the different equilibria reached, based on the level of intelligence of the selling agents, the level of price-importance elasticity of the buying agents, and the level of unemployment in the marketplace. Preliminary experiments addressing these issues are presented.     

Visualising deep network time-series representations

Neural Computing and Applications - Despite the popularisation of machine learning models, more often than not, they still operate as black boxes with no insight into what is happening inside the...     

Cone Calorimeter and Thermogravimetric Analysis of Glass Phenolic Composites Used in Aircraft Applications

The increasing use of composite materials in aircraft cabins and structures poses significant challenges in order to maintain and improve the fire safety of aviation. In this work, the flammability characteristics of a commercial glass-fibre reinforced phenolic composite (GFRP) used for aircraft cabin partitions and furnishing are investigated experimentally. Thermogravimetric analysis under inert atmosphere at several heating rates provided information on the thermal decomposition process. The degradation process is modelled with one and two-step mechanisms using the Ozawa–Flynn–Wall iso-conversional method and the GPYRO numerical code which utilizes a genetic algorithm optimization scheme. The estimated activation energy and pre-exponential factor values, especially in the two-step case (77.18 and 104.69 kJ/mol and 2.60 × 106 and 3.19 × 106 min−1 for the first and the second step respectively), recover reasonably well the conversion degree and its derivative. Tests with a cone calorimeter (CC), performed at different incident heat fluxes, provided information on the reaction to fire characteristics of the material and the influence of the heat flux on the combustion process. In general, combustion proceeds in two stages, flaming and smoldering combustion. The CC results assisted by scanning electron microscopy photos provide information on the charring characteristics of the material. The critical heat flux for ignition and the corresponding ignition temperature are estimated, correlating heat fluxes with time to ignition. Thermally thin and thick models are considered, as well as a modified technique bridging the gap between these limit cases and therefore valid for thermally thin and thick but also intermediate conditions (more pertinent in the present case). The results for this latter approach are $$\dot{q}^{\prime\prime}_{ig,cr}$$ ~ 20 kW/m2 and Tig = 469°C, providing also complementing information on thermophysical properties, such as thermal diffusivity, α = 1.23 × 10−7 m2/s, thermal conductivity, k = 0.325 W/(m K) and specific heat capacity, c = 1.330 kJ/(kg K). This work provides information on the reaction to fire characteristics of GFRP, but also on physical and flammability properties in a form suitable to be used in numerical codes, for the prediction of fire and evacuation scenarios. The influence of the reinforcement structure on the fire behaviour of the composite is also illustrated and discussed.