Computational mechanisms of distributed value representations and mixed learning strategies

Learning appropriate representations of the reward environment is challenging in the real world where there are many options, each with multiple attributes or features. Despite existence of alternative solutions for this challenge, neural mechanisms underlying emergence and adoption of value representations and learning strategies remain unknown. To address this, we measure learning and choice during a multi-dimensional probabilistic learning task in humans and trained recurrent neural networks (RNNs) to capture our experimental observations. We find that human participants estimate stimulus-outcome associations by learning and combining estimates of reward probabilities associated with the informative feature followed by those of informative conjunctions. Through analyzing representations, connectivity, and lesioning of the RNNs, we demonstrate this mixed learning strategy relies on a distributed neural code and opponency between excitatory and inhibitory neurons through value-dependent disinhibition. Together, our results suggest computational and neural mechanisms underlying emergence of complex learning strategies in naturalistic settings.

The current burden of Japanese encephalitis and the estimated impacts of vaccination: Combining estimates of the spatial distribution and transmission intensity of a zoonotic pathogen

Japanese encephalitis virus (JEV) is a major cause of neurological disability in Asia and causes thousands of severe encephalitis cases and deaths each year. Although Japanese encephalitis (JE) is a WHO reportable disease, cases and deaths are significantly underreported and the true burden of the disease is not well understood in most endemic countries. Here, we first conducted a spatial analysis of the risk factors associated with JE to identify the areas suitable for sustained JEV transmission and the size of the population living in at-risk areas. We then estimated the force of infection (FOI) for JE-endemic countries from age-specific incidence data. Estimates of the susceptible population size and the current FOI were then used to estimate the JE burden from 2010 to 2019, as well as the impact of vaccination. Overall, 1,543.1 million (range: 1,292.6-2,019.9 million) people were estimated to live in areas suitable for endemic JEV transmission, which represents only 37.7% (range: 31.6-53.5%) of the over four billion people living in countries with endemic JEV transmission. Based on the baseline number of people at risk of infection, there were an estimated 56,847 (95% CI: 18,003-184,525) JE cases and 20,642 (95% CI: 2,252-77,204) deaths in 2019. Estimated incidence declined from 81,258 (95% CI: 25,437-273,640) cases and 29,520 (95% CI: 3,334-112,498) deaths in 2010, largely due to increases in vaccination coverage which have prevented an estimated 314,793 (95% CI: 94,566-1,049,645) cases and 114,946 (95% CI: 11,421-431,224) deaths over the past decade. India had the largest estimated JE burden in 2019, followed by Bangladesh and China. From 2010-2019, we estimate that vaccination had the largest absolute impact in China, with 204,734 (95% CI: 74,419-664,871) cases and 74,893 (95% CI: 8,989-286,239) deaths prevented, while Taiwan (91.2%) and Malaysia (80.1%) had the largest percent reductions in JE burden due to vaccination. Our estimates of the size of at-risk populations and current JE incidence highlight countries where increasing vaccination coverage could have the largest impact on reducing their JE burden.

The current burden of Japanese encephalitis and the estimated impacts of vaccination: Combining estimates of the spatial distribution and transmission intensity of a zoonotic pathogen

Japanese encephalitis virus (JEV) is a major cause of neurological disability in Asia and causes thousands of severe encephalitis cases and deaths each year. Although Japanese encephalitis (JE) is a WHO reportable disease, cases and deaths are significantly underreported and the true burden of the disease is not well understood in most endemic countries. Here, we first conducted a spatial analysis of the risk factors associated with JE to identify the areas suitable for sustained JEV transmission and the size of the population living in at-risk areas. We then estimated the force of infection (FOI) for JE-endemic countries from age-specific incidence data. Estimates of the susceptible population size and the current FOI were then used to estimate the JE burden from 2010 to 2019, as well as the impact of vaccination. Overall, 1.15 billion (range: 982.1-1543.1 million) people were estimated to live in areas suitable for endemic JEV transmission, which represents 28.0% (range: 24.0-37.7%) of the over four billion people living in countries with endemic JEV transmission. Based on the baseline number of people at risk of infection, there were an estimated 45,017 (95% CI: 13,579-146,375) JE cases and 16,319 (95% CI: 1,804-60,041) deaths in 2019. Estimated incidence declined from 61,879 (95% CI: 18,377-200,406) cases and 22,448 (95% CI: 2,470-83,588) deaths in 2010, largely due to increases in vaccination coverage which have prevented an estimated 214,493 (95% CI: 75,905-729,009) cases and 78,544 (95% CI: 8,243-325,755) deaths over the past decade. India had the largest estimated JE burden in 2019, followed by Bangladesh and China. From 2010-2019, we estimate that vaccination had the largest absolute impact in China, with 142,471 (95% CI: 56,208-484,294) cases and 52,338 (95% CI: 6,421-185,285) deaths prevented, while Taiwan (91.1%) and Malaysia (80.5%) had the largest percent reductions in JE burden due to vaccination. Our estimates of the size of at-risk populations and current JE incidence highlight countries where increasing vaccination coverage could have the largest impact on reducing their JE burden.Author SummaryJapanese encephalitis is a vector-transmitted, zoonotic disease that is endemic throughout a large portion of Asia. Vaccination has significantly reduced the JE burden in several formerly high-burden countries, but vaccination coverage remains limited in several other countries with high JE burdens. A better understanding of both the spatial distribution and the magnitude of the burden in endemic countries is critical for future disease prevention efforts. To estimate the number of people living in areas within Asia suitable for JEV transmission we conducted a spatial analysis of the risk factors associated with JE. We estimate that over one billion people live in areas suitable for local JEV transmission. We then combined these population-at-risk estimates with estimates of the force of infection (FOI) to model the national-level burden of JE (annual cases and deaths) over the past decade. Increases in vaccination coverage have reduced JE incidence from over 60,000 cases in 2010 to 45,000 cases in 2019. We estimate that vaccination has prevented over 214,000 cases and 78,000 deaths in the past decade. Our results also call attention to the countries, and high-risk areas within countries, where increases in vaccination coverage are most needed.

Neural mechanisms of distributed value representations and learning strategies

Learning appropriate representations of the reward environment is extremely challenging in the real world where there are many options to learn about and these options have many attributes or features. Despite existence of alternative solutions for this challenge, neural mechanisms underlying emergence and adoption of value representations and learning strategies remain unknown. To address this, we measured learning and choice during a novel multi-dimensional probabilistic learning task in humans and trained recurrent neural networks (RNNs) to capture our experimental observations. We found that participants estimate stimulus-outcome associations by learning and combining estimates of reward probabilities associated with the informative feature followed by those of informative conjunctions. Through analyzing representations, connectivity, and lesioning of the RNNs, we demonstrate this mixed learning strategy relies on a distributed neural code and distinct contributions of inhibitory and excitatory neurons. Together, our results reveal neural mechanisms underlying emergence of complex learning strategies in naturalistic settings.

Understanding Australia’s national feral cat control effort

Context. Feral cats (Felis catus) pose a significant threat to Australia’s native species and feral cat control is, therefore, an important component of threatened species management and policy. Australia’s Threatened Species Strategy articulates defined targets for feral cat control. Yet, currently, little is known about who is engaged in feral cat control in Australia, what motivates them, and at what rate they are removing feral cats from the environment. Aims. We aim to document who is engaging in feral cat control in Australia, how many cats they remove and to estimate the number of feral cats killed in a single year. Furthermore, we seek to better understand attitudes towards feral cat control in Australia. Methods. We used a mixed methods approach combining quantitative and qualitative techniques. Feral cat control data were obtained from existing data repositories and via surveys targeting relevant organisations and individuals. A bounded national estimate of the number of feral cats killed was produced by combining estimates obtained from data repositories and surveys with modelled predictions for key audience segments. Attitudes towards feral cat control were assessed by exploring qualitative responses to relevant survey questions. Key results. We received information on feral cat control from three central repositories, 134 organisations and 2618 individuals, together removing more than 35000 feral cats per year. When including projections to national populations of key groups, the estimated number of feral cats removed from the environment in the 2017–2018 financial year was 316030 (95% CI: 297742–334318). Conclusions. Individuals and organisations make a significant, and largely unrecorded, contribution to feral cat control. Among individuals, there is a strong awareness of the impact of feral cats on Australia’s biodiversity. Opposition to feral cat control focussed largely on ethical concerns and doubts about its efficacy. Implications. There is significant interest in, and commitment to, feral cat control among some groups of Australian society, beyond the traditional conservation community. Yet more information is needed about control methods and their effectiveness to better understand how these efforts are linked to threatened species outcomes.

Fundamental uncertainty levels of 21 cm power spectra from a delay analysis

Several experimental efforts are underway to measure the power spectrum of 21 cm fluctuations from the epoch of reionization (EoR) using low-frequency radio interferometers. Experiments like the Hydrogen Epoch of Reionization Array (HERA) and Murchison Widefield Array Phase II (MWA) feature highly redundant antenna layouts, building sensitivity through redundant measurements of the same angular Fourier modes, at the expense of diminished UV coverage. This strategy limits the numbers of independent samples of each power spectrum mode, thereby increasing the effect of sample variance on the final power spectrum uncertainty. To better quantify this effect, we measure the sample variance of a delay-transform based power spectrum estimator, using both analytic calculations and simulations of flat-spectrum EoR-like signals. We find that for the shortest baselines in HERA, the sample variance can reach as high as 20 per cent, and up to 30 per cent for the wider fields of view of the MWA. Combining estimates from all the baselines in a HERA- or MWA-like 37 element redundant hexagonal array can lower the variance to 1−3 per cent for some Fourier modes. These results have important implications for observing and analysis strategies, and suggest that sample variance can be non-negligible when constraining EoR model parameters from upcoming 21 cm data.

COMBINING ESTIMATES OF CONDITIONAL TREATMENT EFFECTS

Estimating a treatment’s effect on an outcome conditional on covariates is a primary goal of many empirical investigations. Accurate estimation of the treatment effect given covariates can enable the optimal treatment to be applied to each unit or guide the deployment of limited treatment resources for maximum program benefit. Applications of conditional treatment effect estimation are found in direct marketing, economic policy, and personalized medicine. When estimating conditional treatment effects, the typical practice is to select a statistical model or procedure based on sample data. However, combining estimates from the candidate procedures often provides a more accurate estimate than the selection of a single procedure. This article proposes a method of model combination that targets accurate estimation of the treatment effect conditional on covariates. We provide a risk bound for the resulting estimator under squared error loss and illustrate the method using data from a labor skills training program.

What’s Up: an assessment of Causal Inference in the Perception of Verticality

The perceptual upright is thought to be constructed by the central nervous system (CNS) as a vector sum; by combining estimates on the upright provided by the visual system and the body’s inertial sensors with prior knowledge that the upright is usually above the head. Results from a number of recent studies furthermore show that the weighting of the respective sensory signals is proportional to their reliability, consistent with a Bayesian interpretation of the idea of a vector sum (Forced Fusion, FF). However, findings from a study conducted in partial gravity suggest that the CNS may rely on a single sensory system (Cue Capture, CC), or choose to process sensory signals differently based on inferred signal causality (Causal Inference, CI). We developed a novel Alternative-Reality system to manipulate visual and physical tilt independently, and tasked participants (n=28) to indicate the perceived upright for various (in-)congruent combinations of visual-inertial stimuli. Overall, the data appear best explained by the FF model. However, an evaluation of individual data reveals considerable variability, favoring different models in about equal proportions of participants (FF, n=12; CI, n=7, CC, n=9). Given the observed variability, we conclude that the notion of a vector sum does not provide a comprehensive explanation of the perception of the upright.