Summary: Background: The US overdose crisis is driven by fentanyl, heroin, and prescription opioids. One evidence-based policy response has been to broaden naloxone distribution, but how much naloxone a community would need to reduce the incidence of fatal overdose is unclear. We aimed to estimate state-level US naloxone need in 2017 across three main naloxone access points (community-based programmes, provider prescription, and pharmacy-initiated distribution) and by dominant opioid epidemic type (fentanyl, heroin, and prescription opioid). Methods: In this modelling study, we developed, parameterised, and applied a mechanistic model of risk of opioid overdose and used it to estimate the expected reduction in opioid overdose mortality after deployment of a given number of two-dose naloxone kits. We performed a literature review and used a modified-Delphi panel to inform parameter definitions. We refined an established model of the population at risk of overdose by incorporating changes in the toxicity of the illicit drug supply and in the naloxone access point, then calibrated the model to 2017 using data obtained from proprietary data sources, state health departments, and national surveys for 12 US states that were representative of each epidemic type. We used counterfactual modelling to project the effect of increased naloxone distribution on the estimated number of opioid overdose deaths averted with naloxone and the number of naloxone kits needed to be available for at least 80% of witnessed opioid overdoses, by US state and access point. Findings: Need for naloxone differed by epidemic type, with fentanyl epidemics having the consistently highest probability of naloxone use during witnessed overdose events (range 58–76% across the three modelled states in this category) and prescription opioid-dominated epidemics having the lowest (range 0–20%). Overall, in 2017, community-based and pharmacy-initiated naloxone access points had higher probability of naloxone use in witnessed overdose and higher numbers of deaths averted per 100 000 people in state-specific results with these two access points than with provider-prescribed access only. To achieve a target of naloxone use in 80% of witnessed overdoses, need varied from no additional kits (estimated as sufficient) to 1270 kits needed per 100 000 population across the 12 modelled states annually. In 2017, only Arizona had sufficient kits to meet this target. Interpretation: Opioid epidemic type and how naloxone is accessed have large effects on the number of naloxone kits that need to be distributed, the probability of naloxone use, and the number of deaths due to overdose averted. The extent of naloxone distribution, especially through community-based programmes and pharmacy-initiated access points, warrants substantial expansion in nearly every US state. Funding: National Institute of Health, National Institute on Drug Abuse.

Many potential applications of artificial intelligence involve making real-time decisions in physical systems while interacting with humans. Automobile racing represents an extreme example of these conditions; drivers must execute complex tactical manoeuvres to pass or block opponents while operating their vehicles at their traction limits1. Racing simulations, such as the PlayStation game Gran Turismo, faithfully reproduce the non-linear control challenges of real race cars while also encapsulating the complex multi-agent interactions. Here we describe how we trained agents for Gran Turismo that can compete with the world's best e-sports drivers. We combine state-of-the-art, model-free, deep reinforcement learning algorithms with mixed-scenario training to learn an integrated control policy that combines exceptional speed with impressive tactics. In addition, we construct a reward function that enables the agent to be competitive while adhering to racing's important, but under-specified, sportsmanship rules. We demonstrate the capabilities of our agent, Gran Turismo Sophy, by winning a head-to-head competition against four of the world's best Gran Turismo drivers. By describing how we trained championship-level racers, we demonstrate the possibilities and challenges of using these techniques to control complex dynamical systems in domains where agents must respect imprecisely defined human norms.

The authors have requested that this preprint be removed from Research Square.

We propose a novel method for analyzing and visualizing the complexity of standard reinforcement learning (RL) benchmarks based on score distributions. A large number of policy networks are generated by randomly guessing their parameters, and then evaluated on the benchmark task; the study of their aggregated results provide insights into the benchmark complexity. Our method guarantees objectivity of evaluation by sidestepping learning altogether: the policy network parameters are generated using Random Weight Guessing (RWG), making our method agnostic to (i) the classic RL setup, (ii) any learning algorithm, and (iii) hyperparameter tuning. We show that this approach isolates the environment complexity, highlights specific types of challenges, and provides a proper foundation for the statistical analysis of the task's difficulty. We test our approach on a variety of classic control benchmarks from the OpenAI Gym, where we show that small untrained networks can provide a robust baseline for a variety of tasks. The networks generated often show good performance even without gradual learning, incidentally highlighting the triviality of a few popular benchmarks.

Existing structural coloration methods using thin films, commonly implemented in high-purity aluminium, produce colours which are highly dependent on the viewing angle because of the inherent angular dependence of thin film interference. Adapting the thin film coloration mechanism to anodisation of industrial-quality aluminium alloys, which scatter light more efficiently than their high-purity counterparts, reduces angle dependence in the colour produced. This reduction of angle dependence, as well as the wide use of anodised aluminium in consumer products, suggests that structural colour based on anodised aluminium could potentially be scaled up for commercial scale production.

We demonstrate optical rectification in a reconfigurable and relatively simple nanoscopic tunneling junction formed via resistive switching. In optical rectification, electrons must keep up with the rapid oscillations of an illuminating optical field and harness the nonlinearities of a tunneling contact to produce the desired DC field. Among the intrinsic requirements for such devices are tunneling junctions with an exceedingly small capacitance and surface area. In contrast to tunneling junctions formed by different methods, the resistive switching approach explored here allows the system to be tuned, set, and reset via the application of DC electric fields. This makes it ideally suitable for exploring optical rectification phenomena under different tunneling conditions and for dynamically tuning the device's responsivity. This on-the-go tunability opens the possibility for adaptive devices, such as ultrafast photon detectors, wireless power transmitters, and energy harvesting systems.

Physical coloration without chemicals offers a pathway to develop pollution-free coloration technology, and can be applied to colorimetric sensing of gases, toxic and chemical agents. In this paper, we report on realization of a high-purity red color using mechanism of mode-selective absorption in a thin-film optical cavity. By placing an ultra-thin absorber layer at the antinode of a targeted spectral band in a Fabry-Perot cavity, its otherwise conventional dichroic reflection spectrum is shaped into a broad rectangular flat-bottom one that gives the desired vivid red. The purity of our demonstrated red color reaches 76%, which is increased by 16% compared with those reported in prior thin-film structures. Our method of mode-selective absorption is adaptable to more general-purpose spectral shaping, and could be applied in producing other target colors as well as broadband light absorption for energy harvesting and infrared detection.

Structural coloration is a quickly growing field, encompassing physical and photonic processes such as interference, diffraction, and scattering. In this study, we investigated the optical effects in the visible wavelength range, and in particular, the colour gamuts achievable with absorber–dielectric–metal sandwich structures. These chemical‐free layered structures are highly tunable, easily scaled, optical cavities that are capable of generating remarkable colours whose properties are determined completely by material and structural parameters. We employed experimental and numerical strategies to demonstrate that each absorber spans a unique colour gamut, i.e. a subset of the full chromaticity space. While gamut overlap is observed between different absorber types, the gamut areas unique to each absorber occur at different hues of high excitation purity. A comprehensive understanding of how these colour gamuts develop and how different materials may be combined to expand larger subsets of the chromaticity space is required in order to maximize the variety of colours achievable with this system and elevate it into a ‘structural coloration technology’.

Methods to produce colored surfaces using physical mechanisms are proposed, with aluminum used as a demonstration vehicle. Our approach aims to achieve tunable, vibrant colors through the interaction of three well-established physical mechanisms of color formation, namely (1) resonant scattering and interference effects in a nano-structured surface (e.g. via anodization of aluminum), (2) plasmonic resonances in metal nanoparticles, and (3) spectrally selective absorption enhancement in ultrathin films, such as amorphous germanium and carbon. We provide proof-of-concept by fabricating and characterizing representative aluminum-based samples showing vivid colors throughout the visible spectral range. (C) 2016 Elsevier Ltd. All rights reserved.

We investigate conduction phenomena in ultrathin bismuth (Bi) films that are thermally evaporated onto flat quartz. Critical points in the conductance as a function of deposition time are identified and used to scale the data from time dependence to coverage dependence. The resulting nonlinear coverage scaling equation is verified independently via analysis done on transmission electron microscope images of the evaporated films. The scaled data yields critical exponents in very good agreement with classical percolation theory, and clearly shows the transition from the tunneling regime into percolation. Surprisingly, no noticeable signatures of size-quantization effects in the nucleation sites as a function of deposition time is observed in either regime. We discuss our findings in light of Boltzmann transport modeling of 1D conduction as an approximation to the narrow percolative paths that form at the onset of percolation. Our results suggest that lack of a preferred crystallite orientation in the nucleation process may indeed cause quantum-confinement to be too smeared out to be observable in the tunneling to percolation transition.

We report on uncooled mid-infrared photovoltaic responses at 300 K arising in heterojunctions of reduced graphene oxide with p-Si. Two major photoresponse spectral peaks are observed, one in the near infrared starting at 1.1 μm corresponding to electron-hole pair generation in the Si substrate, and another at wavelengths below 2.5 μm, arising from properties of the reduced graphene oxide-Si heterojunction. Our analysis of the current-voltage characteristics at various temperatures suggests that the two materials form a type-II (broken-gap) heterojunction, with a characteristic transition between direct tunneling to field emission, to over-the-barrier excitation with increasing reverse voltage. Illumination was found to affect the onset of the transition between direct tunneling and field-emission, suggesting that the mid infrared response results from the excitation of minority carriers (electrons) from the Si and their collection in the reduced graphene oxide contact. The photoresponse near 1.1 μm showed a time constant at least five times faster than the one at 2.5 μm, which points to surface defects as well as high series resistance and capacitance as potentially limiting factors in this mode of operation. With proper device engineering considerations, these devices could be promising as a graphene-based platform for infrared sensing.