Empirical evidence for ultrametric structure in multi-layer perceptron error surfaces

Combinatorial optimization problems share an interesting property with spin glass systems in that their state spaces can exhibit ultrametric structure. We use sampling methods to analyse the error surfaces of feedforward multi-layer perceptron neural networks learning encoder problems. The third order statistics of these points of attraction are examined and found to be arranged in a highly ultrametric way. This is a unique result for a finite, continuous parameter space. The implications of this result are discussed.

Visualization of learning in multilayer perceptron networks using principal component analysis

This paper is concerned with the use of scientific visualization methods for the analysis of feedforward neural networks (NNs). Inevitably, the kinds of data associated with the design and implementation of neural networks are of very high dimensionality, presenting a major challenge for visualization. A method is described using the well-known statistical technique of principal component analysis (PCA). This is found to be an effective and useful method of visualizing the learning trajectories of many learning algorithms such as back-propagation and can also be used to provide insight into the learning process and the nature of the error surface.

Influence of process parameters on part quality and mechanical properties for DMLS and SLM with iron-based materials

Rapid manufacturing is an advanced manufacturing technology based on layer-by-layer manufacturing to produce a part. This paper presents experimental work carried out to investigate the effects of scan speed, layer thickness, and building direction on the following part features: dimensional error, surface roughness, and mechanical properties for DMLS with DS H20 powder and SLM with CL 20 powder (1.4404/AISI 316L). Findings were evaluated using ANOVA analysis. According to the experimental results, build direction has a significant effect on part quality, in terms of dimensional error and surface roughness. For the SLM process, the build direction has no influence on mechanical properties. Results of this research support industry estimating part quality and mechanical properties before the production of parts with additive manufacturing, using iron-based powders

Topographic mappings and feed-forward neural networks

This thesis is a study of the generation of topographic mappings - dimension reducing transformations of data that preserve some element of geometric structure - with feed-forward neural networks. As an alternative to established methods, a transformational variant of Sammon's method is proposed, where the projection is effected by a radial basis function neural network. This approach is related to the statistical field of multidimensional scaling, and from that the concept of a 'subjective metric' is defined, which permits the exploitation of additional prior knowledge concerning the data in the mapping process. This then enables the generation of more appropriate feature spaces for the purposes of enhanced visualisation or subsequent classification. A comparison with established methods for feature extraction is given for data taken from the 1992 Research Assessment Exercise for higher educational institutions in the United Kingdom. This is a difficult high-dimensional dataset, and illustrates well the benefit of the new topographic technique. A generalisation of the proposed model is considered for implementation of the classical multidimensional scaling (¸mds}) routine. This is related to Oja's principal subspace neural network, whose learning rule is shown to descend the error surface of the proposed ¸mds model. Some of the technical issues concerning the design and training of topographic neural networks are investigated. It is shown that neural network models can be less sensitive to entrapment in the sub-optimal global minima that badly affect the standard Sammon algorithm, and tend to exhibit good generalisation as a result of implicit weight decay in the training process. It is further argued that for ideal structure retention, the network transformation should be perfectly smooth for all inter-data directions in input space. Finally, there is a critique of optimisation techniques for topographic mappings, and a new training algorithm is proposed. A convergence proof is given, and the method is shown to produce lower-error mappings more rapidly than previous algorithms.

Basis set superposition error effects, excited-state potential energy surface and photodynamics of thymine

En aquesta tesi he estudiat l'efecte de l'error de superposició de base (BSSE) en la planaritat d'algunes molècules. He observat que l'ús d'alguns mètodes de càlcul amb determinades funcions de base descriuen mínims d'energia no planars per les bases nitrogenades de l'ADN. He demostrat que aquests problemes es poden arreglar utilitzant el mètode Counterpoise per corregir el BSSE en els càlculs. En aquesta tesi també he estudiat la fotofísica de la timina i els resultats mostren que existeixen dos camins de relaxació des de l'estat excitat que permeten la regeneració de l'estructura inicial de forma ultraràpida.

Full-dimensional analytical ab initio potential energy surface of the ground state of HOI

Extensive ab initio calculations using a complete active space second-order perturbation theory wavefunction, including scalar and spin-orbit relativistic effects with a quadruple-zeta quality basis set were used to construct an analytical potential energy surface (PES) of the ground state of the [H, O, I] system. A total of 5344 points were fit to a three-dimensional function of the internuclear distances, with a global root-mean-square error of 1.26 kcal mol(-1). The resulting PES describes accurately the main features of this system: the HOI and HIO isomers, the transition state between them, and all dissociation asymptotes. After a small adjustment, using a scaling factor on the internal coordinates of HOI, the frequencies calculated in this work agree with the experimental data available within 10 cm(-1). (C) 2011 American Institute of Physics. [doi: 10.1063/1.3615545]

Processing optimization of probiotic yogurt containing glucose oxidase using response surface methodology

Exposure to oxygen may induce a lack of functionality of probiotic dairy foods because the anaerobic metabolism of probiotic bacteria compromises during storage the maintenance of their viability to provide benefits to consumer health. Glucose oxidase can constitute a potential alternative to increase the survival of probiotic bacteria in yogurt because it consumes the oxygen permeating to the inside of the pot during storage, thus making it possible to avoid the use of chemical additives. This research aimed to optimize the processing of probiotic yogurt supplemented with glucose oxidase using response surface methodology and to determine the levels of glucose and glucose oxidase that minimize the concentration of dissolved oxygen and maximize the Bifidobacterium longum count by the desirability function. Response surface methodology mathematical models adequately described the process, with adjusted determination coefficients of 83% for the oxygen and 94% for the B. longum. Linear and quadratic effects of the glucose oxidase were reported for the oxygen model, whereas for the B. longum count model an influence of the glucose oxidase at the linear level was observed followed by the quadratic influence of glucose and quadratic effect of glucose oxidase. The desirability function indicated that 62.32 ppm of glucose oxidase and 4.35 ppm of glucose was the best combination of these components for optimization of probiotic yogurt processing. An additional validation experiment was performed and results showed acceptable error between the predicted and experimental results.

Infiltration from surface and buried point sources: The average wetting water content

The assumption in analytical solutions for flow from surface and buried point sources of an average water content, (θ) over bar, behind the wetting front is examined. Some recent work has shown that this assumption fitted some field data well. Here we calculated (θ) over bar using a steady state solution based on the work by Raats [1971] and an exponential dependence of the diffusivity upon the water content. This is compared with a constant value of (θ) over bar calculated from an assumption of a hydraulic conductivity at the wetting front of 1 mm day(-1) and the water content at saturation. This comparison was made for a wide range of soils. The constant (θ) over bar generally underestimated (θ) over bar at small wetted radii and overestimated (θ) over bar at large radii. The crossover point between under and overestimation changed with both soil properties and flow rate. The largest variance occurred for coarser texture soils at low-flow rates. At high-flow rates in finer-textured soils the use of a constant (θ) over bar results in underestimation of the time for the wetting front to reach a particular radius. The value of (θ) over bar is related to the time at which the wetting front reaches a given radius. In coarse-textured soils the use of a constant value of (θ) over bar can result in an error of the time when the wetting front reaches a particular radius, as large as 80% at low-flow rates and large radii.

Repeatability of maximal voluntary force and of surface EMG variables during voluntary isometric contraction of quadriceps muscles in healthy subjects

The repeatability of initial values and rate of change of EMG signal mean spectral frequency (MNF), average rectified values (ARV), muscle fiber conduction velocity (CV) and maximal voluntary contraction (MVC) was investigated in the vastus medialis obliquus (VMO) and vastus lateralis (VL) muscles of both legs of nine healthy male subjects during voluntary, isometric contractions sustained for 50 s at 50% MVC. The values of MVC were recorded for both legs three times on each day and for three subsequent days, while the EMG signals have been recorded twice a day for three subsequent days. The degree of repeatability was investigated using the Fisher test based upon the ANalysis Of VAriance (ANOVA), the Standard Error of the Mean (SEM) and the Intraclass Correlation Coefficient (ICC). Data collected showed a high level of repeatability of MVC measurement (normalized SEM from 1.1% to 6.4% of the mean). MNF and ARV initial values also showed a high level of repeatability (ICC > 70% for all muscles and legs except right VMO). At 50% MVC level no relevant pattern of fatigue was observed for the VMO and VL muscles, suggesting that other portions of the quadriceps might have contributed to the generated effort. These observations seem to suggest that in the investigation of muscles belonging to a multi-muscular group at submaximal level, the more selective electrically elicited contractions should be preferred to voluntary contractions. (C) 2001 Elsevier Science Ltd. All rights reserved.

Repeatability of surface EMG variables in the sternocleidomastoid and anterior scalene muscles

In this study we examined the repeatability and reliability of the surface electromyographic (sEMG) signal mean frequency (MNF), average rectified value (ARV) and conduction velocity (CV) measured for the sternocleidomastoid (SCM) and the anterior scalene (AS) muscles in nine healthy volunteers during 15-s isometric cervical flexion contractions at 50% of the maximal voluntary contraction level over 3 non-consecutive days. Repeatability and reliability estimates were obtained for the initial values and rates of change of each sEMG variable by using both the Intraclass Correlation Coefficient (ICC) and the normalised standard error of the mean (nSEM). Results from SCM indicated good levels of repeatability for the initial value and slope of ARV (ICC > 65%). For the AS, high levels of repeatability were identified for the initial value of MNF (ICC > 70%) and the slope of ARV (ICC > 75%). Values of nSEM in the range 2.8-7.2% were obtained for the initial values of MNF and CV for both SCM and AS, indicating clinically acceptable measurement precision. The low value obtained for the nSEM of the initial value of MNF for the AS, in combination with the high ICC, indicates that of all of the variables examined, this variable could offer the best normative index to distinguish between subjects with and without neck pain, and represents the sEMG variable of choice for future evaluation purposes.

Surface stickiness of drops of carbohydrate and organic acid solutions during convective drying: Experiments and modeling

Drying kinetics of low molecular weight sugars such as fructose, glucose, sucrose and organic acid such as citric acid and high molecular weight carbohydrate such as maltodextrin (DE 6) were determined experimentally using single drop drying experiments as well as predicted numerically by solving the mass and heat transfer equations. The predicted moisture and temperature histories agreed with the experimental ones within 6% average relative (absolute) error and average difference of +/- 1degreesC, respectively. The stickiness histories of these drops were determined experimentally and predicted numerically based on the glass transition temperature (T-g) of surface layer. The model predicted the experimental observations with good accuracy. A nonsticky regime for these materials during spray drying is proposed by simulating a drop, initially 120 mum in diameter, in a spray drying environment.

Deploying public surface transit to forward messages in DTN

Delay Tolerant Network (DTN) is a communication architecture enabling connectivity in a topology with unregular end-to-end network connection. DTN enables communication in environments with cross-connectivity, large delays and delivery time variations, and a high error rate. DTN can be used in vehicular networks where public transport get involved. This research aims to analyze the role of public transit as a DTN routing infrastructure. The impact of using public transit as a relay router is investigated by referencing the network performance, defined by its delivery ratio, average delay and overhead. The results show that public transit can be used as a backbone for DTN in an urban scenario using existing protocols. This opens insights for future researches on routing algorithm and protocol design.

Basis set superposition error-counterpoise corrected potential energy surfaces : application to hydrogen peroxide ... X (X=F-, Cl-, Br-, Li+, Na+) complexes

Møller-Plesset (MP2) and Becke-3-Lee-Yang-Parr (B3LYP) calculations have been used to compare the geometrical parameters, hydrogen-bonding properties, vibrational frequencies and relative energies for several X- and X+ hydrogen peroxide complexes. The geometries and interaction energies were corrected for the basis set superposition error (BSSE) in all the complexes (1-5), using the full counterpoise method, yielding small BSSE values for the 6-311 + G(3df,2p) basis set used. The interaction energies calculated ranged from medium to strong hydrogen-bonding systems (1-3) and strong electrostatic interactions (4 and 5). The molecular interactions have been characterized using the atoms in molecules theory (AIM), and by the analysis of the vibrational frequencies. The minima on the BSSE-counterpoise corrected potential-energy surface (PES) have been determined as described by S. Simón, M. Duran, and J. J. Dannenberg, and the results were compared with the uncorrected PES

C-H···O H-bonded complexes: how does basis set superposition error change their potential-energy surfaces?

Geometries, vibrational frequencies, and interaction energies of the CNH⋯O3 and HCCH⋯O3 complexes are calculated in a counterpoise-corrected (CP-corrected) potential-energy surface (PES) that corrects for the basis set superposition error (BSSE). Ab initio calculations are performed at the Hartree-Fock (HF) and second-order Møller-Plesset (MP2) levels, using the 6-31G(d,p) and D95++(d,p) basis sets. Interaction energies are presented including corrections for zero-point vibrational energy (ZPVE) and thermal correction to enthalpy at 298 K. The CP-corrected and conventional PES are compared; the unconnected PES obtained using the larger basis set including diffuse functions exhibits a double well shape, whereas use of the 6-31G(d,p) basis set leads to a flat single-well profile. The CP-corrected PES has always a multiple-well shape. In particular, it is shown that the CP-corrected PES using the smaller basis set is qualitatively analogous to that obtained with the larger basis sets, so the CP method becomes useful to correctly describe large systems, where the use of small basis sets may be necessary

How does basis set superposition error change the potential surfaces for hydrogen-bonded dimers?

We describe a simple method to automate the geometric optimization of molecular orbital calculations of supermolecules on potential surfaces that are corrected for basis set superposition error using the counterpoise (CP) method. This method is applied to the H-bonding complexes HF/HCN, HF/H2O, and HCCH/H2O using the 6-31G(d,p) and D95 + + (d,p) basis sets at both the Hartree-Fock and second-order Møller-Plesset levels. We report the interaction energies, geometries, and vibrational frequencies of these complexes on the CP-optimized surfaces; and compare them with similar values calculated using traditional methods, including the (more traditional) single point CP correction. Upon optimization on the CP-corrected surface, the interaction energies become more negative (before vibrational corrections) and the H-bonding stretching vibrations decrease in all cases. The extent of the effects vary from extremely small to quite large depending on the complex and the calculational method. The relative magnitudes of the vibrational corrections cannot be predicted from the H-bond stretching frequencies alone