Impact of Asymmetric Weight Update on Neural Network Training With Tiki-Taka Algorithm

Recent progress in novel non-volatile memory-based synaptic device technologies and their feasibility for matrix-vector multiplication (MVM) has ignited active research on implementing analog neural network training accelerators with resistive crosspoint arrays. While significant performance boost as well as area- and power-efficiency is theoretically predicted, the realization of such analog accelerators is largely limited by non-ideal switching characteristics of crosspoint elements. One of the most performance-limiting non-idealities is the conductance update asymmetry which is known to distort the actual weight change values away from the calculation by error back-propagation and, therefore, significantly deteriorates the neural network training performance. To address this issue by an algorithmic remedy, Tiki-Taka algorithm was proposed and shown to be effective for neural network training with asymmetric devices. However, a systematic analysis to reveal the required asymmetry specification to guarantee the neural network performance has been unexplored. Here, we quantitatively analyze the impact of update asymmetry on the neural network training performance when trained with Tiki-Taka algorithm by exploring the space of asymmetry and hyper-parameters and measuring the classification accuracy. We discover that the update asymmetry level of the auxiliary array affects the way the optimizer takes the importance of previous gradients, whereas that of main array affects the frequency of accepting those gradients. We propose a novel calibration method to find the optimal operating point in terms of device and network parameters. By searching over the hyper-parameter space of Tiki-Taka algorithm using interpolation and Gaussian filtering, we find the optimal hyper-parameters efficiently and reveal the optimal range of asymmetry, namely the asymmetry specification. Finally, we show that the analysis and calibration method be applicable to spiking neural networks.

Parental Perception of Body Weight Status of Their 8-year-old Children: Findings from the European CHOP Study

Background Maternal perception of child weight status in children with overweight or obesity has received a lot of attention but data on paternal perception of children from presumably healthy cohorts are lacking. Objective We aimed to investigate paternal and maternal perception of child weight status at the age of 8 years in a cohort of 591 children from 5 European countries. Material and Methods Included were 8-year-old children and their parents participating in the European Childhood Obesity Project (EU CHOP). Weight and height of children and parents were measured and Body Mass Index (BMI, kg/m2) was calculated. Both parents were asked to assess their perception of child weight status using Eckstein scales and their concern about child overweight. The agreement between mother and father perceptions was assessed by Cohen kappa coefficient and their relationship was analyzed by linear mixed effects models based on ordinal logistic regression, accounting for country, child gender and BMI, parental BMI, level of education, concern and type of feeding during first year of life. Results Data from children and both parents were available for 432 girls and boys. Mean BMI was comparable in boys and girls (16.7 ± 2.31 vs. 16.9 ± 2.87 kg/m2, P = 0.55). In total, 172 children (29.3%) were overweight or obese. There was a high degree of agreement between mother and father perceptions of their child’s weight status (Cohen kappa 0.77). Multivariate modelling showed that perception levels significantly increased with child BMI but were globally lower than assessed. They differed between countries, gender and types of feeding during first year of life, were influenced by education level of the father but were not related to parental BMI and concern about childhood overweight. Conclusions The study showed no overall differences between mothers and fathers in rating their child’s weight status but both parents had a propensity to underestimate their child’s actual weight, particularly in boys. The EU CHOP trial registered at clinicaltrials.gov as NCT00338689.

Produksi, komposisi botani dan kapasitas tampung padang penggembalaan alam Kecamatan Haharu Kabupaten Sumba Timur

Padang penggembalaan perlu didata dan dikelola karena sebagai sumber pakan yang murah dan mudah diperoleh untuk dikonsumsi ternak. Tujuan penelitian untuk mengetahui produksi, komposisi botani dan kapasitas tampung di Kecamatan Haharu Kabupaten Sumba Timur. Penelitian ini telah dilaksanakan di padang penggembalaan alam Kecamatan Haharu, Kabupaten Sumba Timur. Penelitian dilaksanakan pada bulan Juni sampai Agustus 2021. Metode yang digunakan dalam penelitian adalah metode survei dan pengukuran serta pengamatan langsung di lapangan. Pengukuran produksi hijauan menggunakan metode “Actual Weight Estimate” dengan menggunakan kuadran ukur 1 m x 1 m. Data yang diperoleh ditabulasi dan dihitung untuk mendapatkan total produksi hijauan pakan, komposisi botani, dan kapasitas tampung. Hasil penelitian menunjukkan bahwa padang penggembalaan Kecamatan Haharu memiliki produksi bahan segar sebesar 2.109,512 kg/ha, produksi bahan kering sebesar 1.304,754 kg/ha, nilai Summed Dominance Ratio (SDR) rumput 98,74%, legum 1,08 %, dan gulma 0,17 %. Padang penggembalaan ini juga memiliki kapasisitas tampung sebesar 0,03 ST/ha/tahun. Kesimpulan, padang padang penggembalaan alam di Kecamatan Haharu didominasi oleh rumput alam dengan kapasitas tampung yang rendah pada musim kemarau.

Regression Equations for Weight Estimation in Paediatric Resuscitation

Background: Weight estimation is critical in paediatric resuscitation, as stopping to weigh a child could influence their survival. Weight estimation methods used in New Zealand (NZ) are not accurate for the population, increasing the complexity of prescribing medication and selecting equipment. Aim: Develop regression equations (RE) to predict the weight of NZ children based on height, sex, age and ethnicity to be deployed in a mobile application (Weight Estimation Without Waiting). Methods: The RE was derived from retrospective regression modelling of a large existing dataset. Data were presented using descriptive statistics and calculation of means, limits of agreement and the proportion of weight estimates within a percentage of actual weight. Conclusion: The RE developed in this study outperformed existing age-based weight estimation methods while providing a method to ensure that weight estimation techniques evolve with NZ children.

Paediatric diagnostic reference levels for common radiological examinations using the European guidelines

Objective: The purpose of this study was to explore the feasibility to determine regional diagnostic reference levels (RDRLs) for paediatric conventional and CT examinations using the European guidelines and to compare RDRLs derived from weight and age groups, respectively. Methods: Data were collected from 31 hospitals in 4 countries, for 7 examination types for a total of 2978 patients. RDRLs were derived for each weight and age group, respectively, when the total number of patients exceeded 15. Results: It was possible to derive RDRLs for most, but not all, weight-based and age-based groups for the seven examinations. The result using weight-based and age-based groups differed substantially. The RDRLs were lower than or equal to the European and recently published national DRLs. Conclusion: It is feasible to derive RDRLs. However, a thorough review of the clinical indications and methodologies has to be performed previous to data collection. This study does not support the notion that DRLs derived using age and weight groups are exchangeable. Advances in knowledge: Paediatric DRLs should be derived using weight-based groups with access to the actual weight of the patients. DRLs developed using weight differ markedly from those developed with the use of age. There is still a need to harmonize the method to derive solid DRLs for paediatric radiological examinations.

The Correlation between the CT Liver Volume and the Actual Weight of the Resected Cirrhotic Liver in Transplantation Patients

Objective: Volumetric assessment with computed tomography (CT), known as CT volumetry, is the preferred method for estimating future liver remnant. However, the data regarding the usage of CT volumetry to estimate future liver remnant of the diseased liver is still lacking. This study was designed to evaluate the correlation between the liver volume, calculated by CT, and the actual weight of the resected liver in patients who underwent orthotopic liver transplantation.Material and Methods: A total of 32 patients having underwent liver transplantation; from March 2009 to June 2015, were included. A radiologist retrospectively reviewed the pre-operative CT and performed the volume measurement. Statistical analysis was performed to determine the relationship between the estimated liver volume and the actual liver weight.Results: The estimated liver volume was significantly different among the cirrhosis of different etiology (p-value=0.001 for the total liver volume and p-value=0.003 for the functional liver volume). Compared with the total liver volume, the functional liver volume had a stronger correlation with the actual weight of the resected liver (r=0.955 vs. r=0.786). The following formula can be used to accurately estimate the expected weight of the resected liver (expected liver weight: ELW), based on the estimated functional liver volume (FLV) derived by CT volumetry: ELW=489.531+(0.618*FLV). The R-squared for this regression model was 0.914.Conclusion: CT volumetry is reliable and accurate in predicting the actual amount of the resected liver parenchyma in cirrhotic patients.

Stress and Deformation Analysis of Batch Plates using Finite Element Method

The aim of the paper is to check the load-bearing capacity of batch plates. The batch plates are used for coating process of ball joints in a gas nitriding furnace. The plates have circular shape with two types of circular holes, the first type of holes are used for gas flow and the second type of holes are used for ball joints. The plates are laid on the supporting frame. The frame is supported in four suspension eyes, it is loaded by the actual weight of the plate and the plate is loaded by the weight of the stored ball joints. From the results is clear that adjustments need to be made on the supporting frame for the safe operation.

Comparison of ultrasonographic estimated fetal weight and actual birth weight.

Objective: To determine the accuracy of estimated fetal weight by ultrasound compared with actual birth weight. Study Design: Cross Sectional study. Setting: Department of Radiology, Fatima Memorial Hospital, Lahore. Period: September 2017 to January 2018. Material & Methods: A sample of 139 pregnant women who fulfilled the inclusion and exclusion criteria were included in this study Ultrasonography of full term pregnant women was performed to determine the comparison and accuracy with the actual weight of baby at birth. Results: In a sample of 139 pregnant women, the mean age was 27.8±4.2 years (with minimum age of 20 years and maximum age of 40 years). Ultrasonographic estimated fetal weight and actual birth weight was compared by using paired t-test. No significant difference was found between estimated fetal weight and actual birth weight with P-value 0.237. Conclusion: Ultrasound is highly sensitive, good, reliable, safe and accurate modality for estimation of fetal weight. There is no significant difference between fetal weight and actual birth weight.

Comparative study between Johnson’s formula and Dare’s formula of fetal weight estimation at term

Background: Prediction of fetal weight is one of the methods towards effective management of pregnancy and delivery. To assess and compare the accuracy of clinical and sonographic fetal weight estimation in predicting birth weight at term pregnancy, patients who were in latent or in active phase of labour. In the present study, an effort is made to compare two different clinical methods and USG and relate to the actual weight of the baby at birth.Methods: It is a prospective observational study of one hundred pregnant women satisfying the criteria, consenting for the study was recruited. Both USG and clinical methods will be done and compared with estimated the fetal weight. Weight of the baby at birth will be measured.Results: All the three methods had significant relationship with the baby weight. Percentage error was least with USG and the standard deviation of error was lower with Dare’s formula. The standard deviation was minimal for Dare`s formula EFW followed closely by USG.Conclusions: It can be concluded that Dare’s formula of clinical methods can be a potential option to be promoted in predicting the fetal weight in the absence of USG facilities. Training in this method is very important and can be an integral part in managing pregnancy during delivery in primary care setting.

Prediction of Energy Consumption by Ships at the port using Deep Learning

The harbours using green ports have become a common mode of enabling the use of environment friendly energy consumption. In this paper, two major contributions are made: reduction of energy consumption in the ports by using ships; prediction of energy consumption with respect to a green port. The characteristics that will play a crucial role in energy consumption of ships are considered and a detailed analysis has been performed to predict the energy consumed by the ships. Deep learning methodologies such as, K-Nearest Regression (KNR), Linear Regression (LR), BP Network (BP), Random Forest Regression (RF) and Gradient Boosting Regression (GBR) are used to determine the different characteristics of the ships that are used while the external features of the ports are given as input. To determine the efficiency of the proposed work, k-fold cross validation is also incorporated. Based on feature importance, the crucial features of the algorithm are selected. The influence of different changing aspects on the ship's energy usage is identified, and reduction methods are implemented appropriately. According to the observed data, the most essential factors that may be utilised to estimate energy consumption of the ship are efficiency of facilities, actual weight, deadweight tonnage, and net tonnage. As the efficiency increases, there is also a significant reduction and the power consumption of the ship at the rate of 8% and 32% in port and berth respectively.