Weather forecasting is a fundamental problem for anticipating and mitigating the impacts of climate change. Recently, data-driven approaches for weather forecasting based on deep learning have shown great promise, achieving accuracies that are competitive with operational systems. However, those methods often employ complex, customized architectures without sufficient ablation analysis, making it difficult to understand what truly contributes to their success. Here we introduce Stormer, a simple transformer model that achieves state-of-the art performance on weather forecasting with minimal changes to the standard transformer backbone. We identify the key components of Stormer through careful empirical analyses, including weather-specific embedding, randomized dynamics forecast, and pressure-weighted loss. At the core of Stormer is a randomized forecasting objective that trains the model to forecast the weather dynamics over varying time intervals. During inference, this allows us to produce multiple forecasts for a target lead time and combine them to obtain better forecast accuracy. On WeatherBench 2, Stormer performs competitively at short to medium-range forecasts and outperforms current methods beyond 7 days, while requiring orders-of-magnitude less training data and compute. Additionally, we demonstrate Stormer’s favorable scaling properties, showing consistent improvements in forecast accuracy with increases in model size and training tokens. Code and checkpoints are available at https://github.com/tung-nd/stormer.

We consider the problem of correctly identifying the mode of a discrete distribution $\mathcal{P}$ with sufficiently high probability by observing a sequence of i.i.d. samples drawn from $\mathcal{P}$. This problem reduces to the estimation of a single parameter when $\mathcal{P}$ has a support set of size $K = 2$. After noting that this special case is handled very well by prior-posterior-ratio (PPR) martingale confidence sequences (Waudby-Smith and Ramdas, 2020), we propose a generalisation to mode estimation, in which $\mathcal{P}$ may take $K \geq 2$ values. To begin, we show that the "one-versus-one" principle to generalise from $K = 2$ to $K \geq 2$ classes is more efficient than the "one-versus-rest" alternative. We then prove that our resulting stopping rule, denoted PPR-1v1, is asymptotically optimal (as the mistake probability is taken to 0). PPR-1v1 is simple and computationally light, and incurs significantly fewer samples than competitors even in the non-asymptotic regime. We demonstrate its gains in two practical applications of sampling: election forecasting and verification of smart contracts in blockchains.

MOTIVATION: Genetic map construction is a foundational step in the analysis of structured experimental populations. For markers that hybridize to several genetically similar locations, or where several alleles are present (such as in multiparental populations), current methods often discard the marker or incorrectly call the genotypes. These errors result in information loss, or incorrect genotypes that can corrupt map construction. RESULTS: We present a new approach for simultaneously performing genetic map construction and marker calling. Our new approach allows the calling of a larger number of markers, a larger number of unique alleles per marker and the correct use of markers which hybridize to multiple genetically similar locations. We demonstrate our new approach using simulations, a biparental wheat population and an eight-parent population of spring bread wheat. Applying our method to the eight-parent population increased the number of mapped markers by 71%. We show that the new genetic map allows the investigation of synteny in ways that were not previously possible in that dataset. AVAILABILITY AND IMPLEMENTATION: The method described in this article has been incorporated into R package mpMap2. It is available from CRAN and also from https://github.com/rohan-shah/mpMap2. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Emergency responders increasingly use social media as a means to monitor public information, gather information that could be used in response efforts, and communicate important information during emergency events. However, the adoption of social media into emergency management processes poses socio-technical challenges such as issues of credibility and trust, lack of organizational support, poor tools, and a shortage of resources and training. This study designs, implements, and evaluates an application that supports the work practice of emergency public information officers and their need to gather, monitor, sort, and report social media activity. Based on prior work that examines how social media and the forms of public participation enabled by it are changing public information practice, we iteratively design and evaluate application prototypes using a human-centered process--moving from low-fidelity paper prototypes to a high-fidelity digital prototype that is ready for field use.