Curvature-driven smoothing: a learning algorithm for feed-forward networks

The performance of feed-forward neural networks in real applications can be often be improved significantly if use is made of a-priori information. For interpolation problems this prior knowledge frequently includes smoothness requirements on the network mapping, and can be imposed by the addition to the error function of suitable regularization terms. The new error function, however, now depends on the derivatives of the network mapping, and so the standard back-propagation algorithm cannot be applied. In this paper, we derive a computationally efficient learning algorithm, for a feed-forward network of arbitrary topology, which can be used to minimize the new error function. Networks having a single hidden layer, for which the learning algorithm simplifies, are treated as a special case.

Incorporating curvature information into on-line learning

We analyse the dynamics of a number of second order on-line learning algorithms training multi-layer neural networks, using the methods of statistical mechanics. We first consider on-line Newton's method, which is known to provide optimal asymptotic performance. We determine the asymptotic generalization error decay for a soft committee machine, which is shown to compare favourably with the result for standard gradient descent. Matrix momentum provides a practical approximation to this method by allowing an efficient inversion of the Hessian. We consider an idealized matrix momentum algorithm which requires access to the Hessian and find close correspondence with the dynamics of on-line Newton's method. In practice, the Hessian will not be known on-line and we therefore consider matrix momentum using a single example approximation to the Hessian. In this case good asymptotic performance may still be achieved, but the algorithm is now sensitive to parameter choice because of noise in the Hessian estimate. On-line Newton's method is not appropriate during the transient learning phase, since a suboptimal unstable fixed point of the gradient descent dynamics becomes stable for this algorithm. A principled alternative is to use Amari's natural gradient learning algorithm and we show how this method provides a significant reduction in learning time when compared to gradient descent, while retaining the asymptotic performance of on-line Newton's method.

Phakometric measurement of ocular surface radii of curvature, axial separations and alignment in relaxed and accommodated human eyes

Measurements (autokeratometry, A-scan ultrasonography and video ophthalmophakometry) of ocular surface radii, axial separations and alignment were made in the horizontal meridian of nine emmetropes (aged 20-38 years) with relaxed (cycloplegia) and active accommodation (mean ± 95% confidence interval: 3.7 ± 1.1 D). The anterior chamber depth (-1.5 ± 0.3 D) and both crystalline lens surfaces (front 3.1 ± 0.8 D; rear 2.1 ± 0.6 D) contributed to dioptric vergence changes that accompany accommodation. Accommodation did not alter ocular surface alignment. Ocular misalignment in relaxed eyes is mainly because of eye rotation (5.7 ± 1.6° temporally) with small amounts of lens tilt (0.2 ± 0.8° temporally) and decentration (0.1 ± 0.1 mm nasally) but these results must be viewed with caution as we did not account for corneal asymmetry. Comparison of calculated and empirically derived coefficients (upon which ocular surface alignment calculations depend) revealed that negligible inherent errors arose from neglect of ocular surface asphericity, lens gradient refractive index properties, surface astigmatism, effects of pupil size and centration, assumed eye rotation axis position and use of linear equations for analysing Purkinje image shifts. © 2004 The College of Optometrists.

Phakometric measurement of ocular surface radius of curvature and alignment: evaluation of method with physical model eyes

Ophthalmophakometric measurements of ocular surface radius of curvature and alignment were evaluated on physical model eyes encompassing a wide range of human ocular dimensions. The results indicated that defocus errors arising from imperfections in the ophthalmophakometer camera telecentricity and light source collimation were smaller than experimental errors. Reasonable estimates emerged for anterior lens surface radius of curvature (accuracy: 0.02–0.10 mm; precision 0.05–0.09 mm), posterior lens surface radius of curvature (accuracy: 0.10–0.55 mm; precision 0.06–0.20 mm), eye rotation (accuracy: 0.00–0.32°; precision 0.06–0.25°), lens tilt (accuracy: 0.00–0.33°; precision 0.05–0.98°) and lens decentration (accuracy: 0.00–0.07 mm; precision 0.00–0.07 mm).

The effect of corneal thickness and corneal curvature on pneumatonometer measurements

Purpose. The purpose of this study was to investigate the influence of corneal topography and thickness on intraocular pressure (IOP) and pulse amplitude (PA) as measured using the Ocular Blood Flow Analyzer (OBFA) pneumatonometer (Paradigm Medical Industries, Utah, USA). Methods. 47 university students volunteered for this cross-sectional study: mean age 20.4 yrs, range 18 to 28 yrs; 23 male, 24 female. Only the measurements from the right eye of each participant were used. Central corneal thickness and mean corneal radius were measured using Scheimpflug biometry and corneal topographic imaging respectively. IOP and PA measurements were made with the OBFA pneumatonometer. Axial length was measured using A-scan ultrasound, due to its known correlation with these corneal parameters. Stepwise multiple regression analysis was used to identify those components that contributed significant variance to the independent variables of IOP and PA. Results. The mean IOP and PA measurements were 13.1 (SD 3.3) mmHg and 3.0 (SD 1.2) mmHg respectively. IOP measurements made with the OBFA pneumatonometer correlated significantly with central corneal thickness (r = +0.374, p = 0.010), such that a 10 mm change in CCT was equivalent to a 0.30 mmHg change in measured IOP. PA measurements correlated significantly with axial length (part correlate = -0.651, p < 0.001) and mean corneal radius (part correlate = +0.459, p < 0.001) but not corneal thickness. Conclusions. IOP measurements taken with the OBFA pneumatonometer are correlated with corneal thickness, but not axial length or corneal curvature. Conversely, PA measurements are unaffected by corneal thickness, but correlated with axial length and corneal radius. These parameters should be taken into consideration when interpreting IOP and PA measurements made with the OBFA pneumatonometer.

The interrogation and multiplexing of long period grating curvature sensors using a Bragg grating based, derivative spectroscopy technique

A long period grating is interrogated with a fibre Bragg grating using a derivative spectroscopy technique. A quasi-linear relationship between the output of the sensing scheme and the curvature experienced by the long period grating is demonstrated, with a sensitivity of 5.05 m and with an average curvature resolution of 2.9 × 10-2 m-1. In addition, the feasibility of multiplexing an in-line series of long period gratings with this interrogation scheme is demonstrated with two pairs of fibre Bragg gratings and long period gratings. With this arrangement the cross-talk error between channels was less than ± 2.4 × 10-3 m-1.

Optical bend sensor for vector curvature measurement based on Bragg grating in eccentric core polymer optical fibre

We present an optical bend sensor based on a Bragg grating written in an eccentric core polymer optical fibre. The grating wavelength shifts are studied as a function of bend curvature and fibre orientation and the device exhibits strong fibre orientation dependence, wide bend curvature range of ± 22.7 m-1 and high bend sensitivity of 63 pm/m-1, which is 80 times higher than the reported sensor based on an offset-FBG in standard single mode silica fibre.

The interrogation and multiplexing of long period grating curvature sensors using a Bragg grating based, derivative spectroscopy technique

A long period grating is interrogated with a fibre Bragg grating using a derivative spectroscopy technique. A quasi-linear relationship between the output of the sensing scheme and the curvature experienced by the long period grating is demonstrated, with a sensitivity of 5.05 m and with an average curvature resolution of 2.9 × 10-2 m-1. In addition, the feasibility of multiplexing an in-line series of long period gratings with this interrogation scheme is demonstrated with two pairs of fibre Bragg gratings and long period gratings. With this arrangement the cross-talk error between channels was less than ± 2.4 × 10-3 m-1.

Dynamic two-axis curvature measurement using multicore fiber Bragg gratings interrogated by arrayed waveguide gratings

We describe the use of arrayed waveguide gratings (AWGs) in the interrogation of fiber Bragg gratings (FBGs) for dynamic strain measurement. The ratiometric AWG output was calibrated in a static deflection experiment over a ±200 με range. Dynamic strain measurement was demonstrated with a FBG in a conventional single-mode fiber mounted on the surface of a vibrating cantilever and on a piezoelectric actuator, giving a resolution of 0.5 με at 2.4 kHz. We present results of this technique extended to measure the dynamic differential strain between two FBG pairs within a multicore fiber. An arbitrary cantilever oscillation of the multicore fiber was determined from curvature measurements in two orthogonal axes at 1125 Hz with a resolution of 0.05 m-1. © 2006 Optical Society of America.

Simultaneous measurement of strain and curvature using superstructure fibre Bragg gratings

We report the first demonstration of the simultaneous measurement of strain and curvature, with temperature compensation, using a single superstructure fibre Bragg grating (SFBG). The SFBG exhibits the properties of both the fibre Bragg grating (FBG) and the long period fibre grating (LPG) such that its spectral response facilitates strain measurement from the wavelength shift of the FBG-like characteristic, and independent measurement of curvature from the LPG-like mode-splitting characteristic. The dependence of the LPG mode-splitting on the mode order has also been investigated and utilised for the measurement of very small curvatures.

Biconical-taper-assisted fiber interferometer with modes coupling enhancement for high-sensitive curvature measurement

A modal interferometer based on multimode-singlemode-multimode fiber structure built with a biconical taper for fiber curvature measurement is proposed and experimentally demonstrated. Due to the tapered singlemode fiber acting as a high-efficient mode power converter to enhance the modes coupling, curvature sensor with improved sensitivity is achieved by monitoring the defined fringe visibility of the interference spectrum. The measuring range can be tuned by changing the waist diameter of the fiber taper. Meanwhile, the sensor shows an intrinsic ability to overcome the influence of temperature cross-sensitivity and the power fluctuation of light source. The advantages of easy fabrication, high-quality spectrum with improved sensitivity, and small hysteresis will provide great potential for practical applications of the sensor. © 2013 Springer-Verlag Berlin Heidelberg.

Laparoscopic greater curvature plication in morbidly obese women with type 2 diabetes:effects on glucose homeostasis, postprandial triglyceridemia and selected gut hormones

Background: Laparoscopic greater curvature plication (LGCP) is an emerging bariatric procedure that reduces the gastric volume without implantable devices or gastrectomy. The aim of this study was to explore changes in glucose homeostasis, postprandial triglyceridemia, and meal-stimulated secretion of selected gut hormones [glucose-dependent insulinotropic polypeptide (GIP), glucagon-like peptide-1 (GLP-1), ghrelin, and obestatin] in patients with type 2 diabetes mellitus (T2DM) at 1 and 6 months after the procedure. Methods: Thirteen morbidly obese T2DM women (mean age, 53.2 ± 8.76 years; body mass index, 40.1 ± 4.59 kg/m2) were prospectively investigated before the LGCP and at 1- and 6-month follow-up. At these time points, all study patients underwent a standardized liquid mixed-meal test, and blood was sampled for assessment of plasma levels of glucose, insulin, C-peptide, triglycerides, GIP, GLP-1, ghrelin, and obestatin. Results: All patients had significant weight loss both at 1 and 6 months after the LGCP (p≤0.002), with mean percent excess weight loss (%EWL) reaching 29.7 ;plusmn2.9 % at the 6-month follow-up. Fasting hyperglycemia and hyperinsulinemia improved significantly at 6 months after the LGCP (p<0.05), with parallel improvement in insulin sensitivity and HbA1c levels (p<0.0001). Meal-induced glucose plasma levels were significantly lower at 6 months after the LGCP (p<0.0001), and postprandial triglyceridemia was also ameliorated at the 6-month follow-up (p<0.001). Postprandial GIP plasma levels were significantly increased both at 1 and 6 months after the LGCP (p<0.0001), whereas the overall meal-induced GLP-1 response was not significantly changed after the procedure (p ;gt0.05). Postprandial ghrelin plasma levels decreased at 1 and 6 months after the LGCP (p<0.0001) with no significant changes in circulating obestatin levels. Conclusion: During the initial 6-month postoperative period, LGCP induces significant weight loss and improves the metabolic profile of morbidly obese T2DM patients, while it also decreases circulating postprandial ghrelin levels and increases the meal-induced GIP response. © 2013 Springer Science+Business Media New York.