No abstract

Neural Architecture Search (NAS) has seen an explosion of research in the past few years. A variety of methods have been proposed to perform NAS, including reinforcement learning, Bayesian optimization with a Gaussian process model, evolutionary search, and gradient descent. In this work, we design a NAS algorithm that performs Bayesian optimization using a neural network model.We develop a path-based encoding scheme to featurize the neural architectures that are used to train the neural network model. This strategy is particularly effective for encoding architectures in cell-based search spaces. After training on just 200 random neural architectures, we are able to predict the validation accuracy of a new architecture to within one percent of its true accuracy on average, for popular search spaces. This may be of independent interest beyond Bayesian neural architecture search.We test our algorithm on the NASBench (Ying et al. 2019) and DARTS (Liu et al. 2018) search spaces, and we show that our algorithm outperforms other NAS methods including evolutionary search, reinforcement learning, AlphaX, ASHA, and DARTS. Our algorithm is over 100x more efficient than random search, and 3.8x more efficient than the next-best algorithm on the NASBench dataset. As there have been problems with fair and reproducible experimental evauations in the field of NAS, we adhere to the recent NAS research checklist (Lindauer and Hutter 2019) to facilitate NAS research. In particular, our implementation has been made publicly available, including all details needed to fully reproduce our results.

No abstract

To determine if residential water sampling corroborates the expectation that formation of stable PbO2 coatings on lead service lines (LSLs) provides an effective lead release control strategy, lead profile sampling was evaluated for eight home kitchen taps in three U.S. cities with observed PbO2-coated LSLs (Newport, Rhode Island; Cincinnati and Oakwood, Ohio). After various water standing times, these LSLs typically released similar or lower peak lead levels (1 to 18 μg/L) than the lead levels from the respective kitchen faucets (1 to 130 μg/L), and frequently 50-80% lower than the lead levels typically reported from Pb(II)-coated LSLs in comparable published sampling studies. Prolonged stagnation (10-101 h) at the Cincinnati sites produced varying results. One site showed minimal (0-4 μg/L) increase in lead release from the PbO2-coated LSL, and persistence of free chlorine residual. However, the other site showed up to a 3-fold increase proportional to standing time, with essentially full depletion of the chlorine residual. Overall, lead release was consistently much lower than that reported in studies of Pb(II)-coated LSL scales, suggesting that natural formation of PbO2 in LSLs is an effective lead "corrosion" control strategy.

Attention deficit hyperactivity disorder (ADHD) is a complex, multifactorial disorder characterized by physical hyperactivity and behavioural disinhibition. Short interval cortical inhibition (SICI), measured in motor cortex with transcranial magnetic stimulation, is reduced in ADHD and correlates with symptom severity. However, ADHD medication-induced changes in SICI vary widely among normal individuals and have not been well studied in children with ADHD. Therefore, we undertook this study to measure and compare effects of two ADHD medications, methylphenidate (MPH), a psychostimulant, and atomoxetine (ATX), a selective norepinephrine reuptake inhibitor, on SICI in children with ADHD. In addition, we wished to determine whether a genetic variation in the dopamine transporter (DAT1), a site of action of MPH, could influence the effects of MPH or ATX on SICI. We performed a randomized, double-blind, single-dose, crossover study comparing 0.5 mg/kg MPH with 1.0 mg/kg ATX in 16 children with ADHD, aged 8-17. Seven were homozygotes and 9 heterozygotes for the DAT1 variable number of tandem repeats 10-repeat allele. Medication and genotype effects on SICI were estimated with repeated measures, mixed model regression. We found that MPH and ATX had similar effects on SICI. However, medication effects differed significantly by DAT1 genotype [F(2,13) = 13.04, P = 0.0008]. Both MPH and ATX increased SICI in heterozygotes but not in 10-repeat homozygotes. In conclusion, MPH and ATX have similar effects on SICI in children with ADHD. A genetic variation in DAT1, previously linked to ADHD risk and MPH behavioural responses, influences the neurophysiological effects of both MPH and ATX.