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.

"Back to Basics" Approach for Improving Maternal Health Care Services Utilization in Lao PDR.

Earlier studies claim inadequate knowledge and understanding of maternal health care service among women and families account for low demand and utilization of these services. This study explores maternal service utilization in Lao PDR by interviewing women, families, health service providers, and community members in Xiengkhuang province. In general, women's attitude and acceptance of maternal health care were positive, with many expressing appreciation and need for maternal health services. Nevertheless, utilization of maternal health services remained poor largely due to dissatisfaction with services: inaccessibility, irregular services, fund shortage, poor facilities, and problems dealing with male health service providers. It appears that utilization of maternal health care services in Lao PDR is as much a health system concern as it is a user preference issue. This article advocates a "back to basics" approach for reviewing maternal health care infrastructure systems in developing countries to improve community health services and utilization.

Provider perspectives on constraints in providing maternal, neonatal and child health services in the Lao People's democratic republic: a qualitative study.

BACKGROUND: To reduce its high maternal and neonatal mortality rate and meet Millennium Development Goals four and five, Lao PDR has adopted a national 'Strategy and Planning Framework of Implementation of Maternal, Neonatal and Child Health Services'. This paper reports on implementation constraints identified in three demonstration sites. METHODS: The objectives of this paper are to analyse health worker perceptions of the implementation of the strategy and constraints faced during implementation. A qualitative design was used with interviews conducted at health facilities in three demonstration provinces. Data were collected through key interviews with provincial/district hospital providers (n = 27), health centre staff (n = 8) and village health volunteers (n = 10). Data was analysed informed by Hanson et al's health system constraint framework. RESULTS: In each of the demonstration sites, the Maternal, Neonatal and Child Health program was generally well-understood and the different activities were being implemented. Perceived implementation constraints related mainly to a mix of supply and demand factors. Supply-side constraints related to inadequate human resources, poor remuneration, weak technical guidance, minimal supervision and limited equipment. Demand-side constraints related mainly to cost, limited access to transport, cultural practices and language. Other constraints related to broader strategic management and cross-sectoral contextual constraints. Contextual constraints included low levels of limited education, women's position in society and poor transport and communications networks. These factors influenced the implementation process and if not addressed, may reduce the effectiveness of the policy and scale-up. CONCLUSION: The Lao PDR has a well-defined Maternal, Neonatal and Child Health program. Analysis of the constraints experienced by service providers in implementing the program however, is essential for scaling-up the initiative. To achieve effective implementation and scale-up a number of concurrent interventions are needed to address identified constraints. More research is needed to identify the optimal combination of interventions to improve these constraints. The broader contextual characteristics require longer-term, cross-sectoral action.

[Effects of bisphenol A on the placental function and reproduction in rats].

OBJECTIVES: The aim of this study was to investigate the effects of bisphenol A (BPA), an estrogen-like environmental endocrine disrupter, on the placental function and reproduction in rats. The mRNA levels of the placental prolactin-growth hormone(PRL-GH) gene family, placental trophoblast cell frequency and reproductive data were analyzed. METHODS: The pregnancies of F344 Fisher rats (160 g +/-20 g) were detected by the presence of the copulatory plug or sperm in the vaginal smear, which marked Day 0 of pregnancy. Pregnant rats were divided into three groups. The control group was intraperitoneally injected with a sesame oil vehicle. The two remaining groups were injected with 50 or 500 mg/kg B.W/day of BPA, resuspended in sesame oil, on either days 7 to 11 or 16 to 20 of pregnancy, with the rats sacrificed on either day 11 or 20, respectively. The mRNA levels of PRL-GH and Pit-1a and b isotype genes were analyzed by Northern blot hybridization and reverse transcription-polymerase chain reaction. The hormone concentrations were analyzed by radioimmunoassay, and the frequency of the placental trophoblast cells observed by a histochemical study. Reproductive data, such as the placental weight and litter size, were surveyed on day 20. The fetal weight was surveyed for 4 weeks after birth. A statistical analysis was carried out using the SAS program (version 8.1). RESULTS: The mRNA levels of the PRL-GH gene family, such as placental lactogen I, Iv and II, prolactin like protein A, C and Cv, and decidual prolactin-related protein were significantly reduced due to BPA exposure. The mRNA levels of the Pit-1a and b isotype genes, which induce the expression of the PRL-GH gene family in the rat placenta, were also reduced due to BPA exposure. The PL-Iv and PL-II concentrations were reduced in the BPA exposed group. During the middle to last stage of pregnancy (Days 11-20), a high dose of BPA exposure reduced the frequency of spongiotrophoblast cells, which are responsible for the secretion of the PRL-GH hormones. Reproductive data, such as the placental and fetal weights and the litter size, were reduced, but that of the pregnancy period was extended in the BPA exposed compared to the control group. CONCLUSIONS: BPA disrupts the placental functions in rats, which leads to reproductive disorders.