Information geometric measurements of generalisation

Neural networks can be regarded as statistical models, and can be analysed in a Bayesian framework. Generalisation is measured by the performance on independent test data drawn from the same distribution as the training data. Such performance can be quantified by the posterior average of the information divergence between the true and the model distributions. Averaging over the Bayesian posterior guarantees internal coherence; Using information divergence guarantees invariance with respect to representation. The theory generalises the least mean squares theory for linear Gaussian models to general problems of statistical estimation. The main results are: (1)~the ideal optimal estimate is always given by average over the posterior; (2)~the optimal estimate within a computational model is given by the projection of the ideal estimate to the model. This incidentally shows some currently popular methods dealing with hyperpriors are in general unnecessary and misleading. The extension of information divergence to positive normalisable measures reveals a remarkable relation between the dlt dual affine geometry of statistical manifolds and the geometry of the dual pair of Banach spaces Ld and Ldd. It therefore offers conceptual simplification to information geometry. The general conclusion on the issue of evaluating neural network learning rules and other statistical inference methods is that such evaluations are only meaningful under three assumptions: The prior P(p), describing the environment of all the problems; the divergence Dd, specifying the requirement of the task; and the model Q, specifying available computing resources.

An exploratory study to determine applicability of nano-hardness and micro-compression measurements for yield stress estimation

The superior properties of ferritic/martensitic steels in a radiation environment (low swelling, low activation under irradiation and good corrosion resistance) make them good candidates for structural parts in future reactors and spallation sources. While it cannot substitute for true reactor experiments, irradiation by charged particles from accelerators can reduce the number of reactor experiments and support fundamental research for a better understanding of radiation effects in materials. Based on the nature of low energy accelerator experiments, only a small volume of material can be uniformly irradiated. Micro and nanoscale post irradiation tests thus have to be performed. We show here that nanoindentation and micro-compression testing on T91 and HT-9 stainless steel before and after ion irradiation are useful methods to evaluate the radiation induced hardening.

The Proinflammatory Environment in Potential Heart and Lung Donors: Prevalence and Impact of Donor Management and Hormonal Therapy

BACKGROUND: Brain stem death can elicit a potentially manipulable cardiotoxic proinflammatory cytokine response. We investigated the prevalence of this response, the impact of donor management with tri-iodothyronine (T3) and methylprednisolone (MP) administration, and the relationship of biomarkers to organ function and transplant suitability. METHODS: In a prospective randomized double-blinded factorially designed study of T3 and MP therapy, we measured serum levels of interleukin-1 and -6 (IL-1 and IL-6), tumor necrosis factor-alpha (TNF-alpha), C-reactive protein, and procalcitonin (PCT) levels in 79 potential heart or lung donors. Measurements were performed before and after 4 hr of algorithm-based donor management to optimize cardiorespiratory function and +/-hormone treatment. Donors were assigned to receive T3, MP, both drugs, or placebo. RESULTS: Initial IL-1 was elevated in 16% donors, IL-6 in 100%, TNF-alpha in 28%, CRP in 98%, and PCT in 87%. Overall biomarker concentrations did not change between initial and later measurements and neither T3 nor MP effected any change. Both PCT (P =0.02) and TNF-alpha (P =0.044) levels were higher in donor hearts with marginal hemodynamics at initial assessment. Higher PCT levels were related to worse cardiac index and right and left ventricular ejection fractions and a PCT level more than 2 ng x mL(-1) may attenuate any improvement in cardiac index gained by donor management. No differences were observed between initially marginal and nonmarginal donor lungs. A PCT level less than or equal to 2 ng x mL(-1) but not other biomarkers predicted transplant suitability following management. CONCLUSIONS: There is high prevalence of a proinflammatory environment in the organ donor that is not affected by tri-iodothyronine or MP therapy. High PCT and TNF-alpha levels are associated with donor heart dysfunction. (C) 2009 Lippincott Williams & Wilkins, Inc.

Cadmium in the environment and high risk population groups

Cadmium has been widely used in various industries for the past fifty years, with current world production standing at around 16,755 tonnes per year. Very little cadmium is ever recycled and the ultimate fate of all cadmium is the environment. In view of reports that cadmium in the environment is increasing, this thesis aims to identify population groups 'at risk' of receiving dietary intakes of cadmium up to or above the current Food and Agricultural Organisation/World Health Organisation maximum tolerable intake of 70 ug/day. The study involves the investigation of one hundred households (260 individuals) who grow a large proportion of their vegetable diet in garden soils in the Borough of Walsall, part of an urban/industrial area in the United Kingdom. Measurements were made of the cadmium levels in atmospheric deposition, soil, house dust, diet and urine from the participants. Atmospheric deposition of cadmium was found to be comparable with other urban/industrial areas in the European Community, with deposition rates as high as 209 g ha-1 yr-1. The garden soils of the study households were found to contain up to 33 mg kg-1 total cadmium, eleven times the highest level usually found in agricultural soils. Dietary intakes of cadmium by the residents from food were calculated to be as high as 68 ug/day. It is suggested that with intakes from other sources, such as air, adventitious ingestion, smoking and occupational exposure, total intakes of cadmium may reach or exceed the FAO/WHO limit. Urinary excretion of cadmium amongst a non-smoking, non-occupationally exposed sub-group of the study population was found to be significantly higher than that of a similar urban population who did not rely on home-produced vegetables. The results from this research indicate that present levels of cadmium in urban/industrial areas can increase dietary intakes and body burdens of cadmium. As cadmium serves no useful biological function and has been found to be highly toxic, it is recommended that policy measures to reduce human exposure on the European scale be considered.

Suitability of dysphonia measurements for telemonitoring of Parkinson's disease

We present an assessment of the practical value of existing traditional and non-standard measures for discriminating healthy people from people with Parkinson's disease (PD) by detecting dysphonia. We introduce a new measure of dysphonia, Pitch Period Entropy (PPE), which is robust to many uncontrollable confounding effects including noisy acoustic environments and normal, healthy variations in voice frequency. We collected sustained phonations from 31 people, 23 with PD. We then selected 10 highly uncorrelated measures, and an exhaustive search of all possible combinations of these measures finds four that in combination lead to overall correct classification performance of 91.4%, using a kernel support vector machine. In conclusion, we find that non-standard methods in combination with traditional harmonics-to-noise ratios are best able to separate healthy from PD subjects. The selected non-standard methods are robust to many uncontrollable variations in acoustic environment and individual subjects, and are thus well-suited to telemonitoring applications.

Correlation between p-GaN growth environment with electrical and optical properties of blue LEDs

Two blue (450 nm) light–emitting diodes (LED), which only differ in top p-GaN layer growth conditions, were comparatively investigated. I-V, C-V, TLM, Electroluminescence (EL) and Photoluminescence (PL) techniques were applied to clarify a correlation between MOCVD carrier gas and internal properties. The A-structure grown in the pure N2 environment demonstrated better parameters than the B-structure grown in the N2/H2 (1:1) gas mixture. The mixed growth atmosphere leaded to an increase of sheet resistances of p-GaN layer. EL and PL measurements confirmed the advantage of the pure N2 utilization, and C(VR) measurement pointed the increase of static charge concentration near the p-GaN interface in the B structure.

The automation of protein crystal presentation for x ray diffraction experiments using standing acoustic waves in a microfluidic chip environment

As the pressure continues to grow on Diamond and the world's synchrotrons for higher throughput of diffraction experiments, new and novel techniques are required for presenting micron dimension crystals to the X ray beam. Currently this task is both labour intensive and primarily a serial process. Diffraction measurements typically take milliseconds but sample preparation and presentation can reduce throughput down to 4 measurements an hour. With beamline waiting times as long as two years it is of key importance for researchers to capitalize on available beam time, generating as much data as possible. Other approaches detailed in the literature [1] [2] [3] are very much skewed towards automating, with robotics, the actions of a human protocols. The work detailed here is the development and discussion of a bottom up approach relying on SSAW self assembly, including material selection, microfluidic integration and tuning of the acoustic cavity to order the protein crystals.

The automation of protein crystal presentation for X ray diffraction experiments using standing acoustic waves in a microfluidic chip environment

As the pressure continues to grow on Diamond and the world's synchrotrons for higher throughput of diffraction experiments, new and novel techniques are required for presenting micron dimension crystals to the X ray beam. Currently this task is both labour intensive and primarily a serial process. Diffraction measurements typically take milliseconds but sample preparation and presentation can reduce throughput down to 4 measurements an hour. With beamline waiting times as long as two years it is of key importance for researchers to capitalize on available beam time, generating as much data as possible. Other approaches detailed in the literature [1] [2] [3] are very much skewed towards automating, with robotics, the actions of a human protocols. The work detailed here is the development and discussion of a bottom up approach relying on SSAW self assembly, including material selection, microfluidic integration and tuning of the acoustic cavity to order the protein crystals.