In regular statistical models, the leave-one-out cross-validation is asymptotically equivalent to the Akaike information criterion. However, since many learning machines are singular statistical models, the asymptotic behavior of the cross-validation remains unknown. In previous studies, we established the singular learning theory and proposed a widely applicable information criterion, the expectation value of which is asymptotically equal to the average Bayes generalization loss. In the present paper, we theoretically compare the Bayes cross-validation loss and the widely applicable information criterion and prove two theorems. First, the Bayes cross-validation loss is asymptotically equivalent to the widely applicable information criterion as a random variable. Therefore, model selection and hyperparameter optimization using these two values are asymptotically equivalent. Second, the sum of the Bayes generalization error and the Bayes cross-validation error is asymptotically equal to 2λ/n, where λ is the real log canonical threshold and n is the number of training samples. Therefore the relation between the cross-validation error and the generalization error is determined by the algebraic geometrical structure of a learning machine. We also clarify that the deviance information criteria are different from the Bayes cross-validation and the widely applicable information criterion.
Sparse coding--that is, modelling data vectors as sparse linear combinations of basis elements--is widely used in machine learning, neuroscience, signal processing, and statistics. This paper focuses on the large-scale matrix factorization problem that consists of learning the basis set in order to adapt it to specific data. Variations of this problem include dictionary learning in signal processing, non-negative matrix factorization and sparse principal component analysis. In this paper, we propose to address these tasks with a new online optimization algorithm, based on stochastic approximations, which scales up gracefully to large data sets with millions of training samples, and extends naturally to various matrix factorization formulations, making it suitable for a wide range of learning problems. A proof of convergence is presented, along with experiments with natural images and genomic data demonstrating that it leads to state-of-the-art performance in terms of speed and optimization for both small and large data sets.
Rapid advances in hardware-based technologies during the past decades have opened up new possibilities for life scientists to gather multimodal data in various application domains, such as omics, bioimaging, medical imaging, and (brain/body)–machine interfaces. These have generated novel opportunities for development of dedicated data-intensive machine learning techniques. In particular, recent research in deep learning (DL), reinforcement learning (RL), and their combination (deep RL) promise to revolutionize the future of artificial intelligence. The growth in computational power accompanied by faster and increased data storage, and declining computing costs have already allowed scientists in various fields to apply these techniques on data sets that were previously intractable owing to their size and complexity. This paper provides a comprehensive survey on the application of DL, RL, and deep RL techniques in mining biological data. In addition, we compare the performances of DL techniques when applied to different data sets across various application domains. Finally, we outline open issues in this challenging research area and discuss future development perspectives.
<p>This study presents an Early Warning System (EWS) for coastal flooding that integrates wind, wave, and sea level forecasts which are validated using in situ records. The system employs the SWAN spectral wave model to simulate nearshore hydrodynamics while an empirical approach is used to assess Total Watel Level (TWL) exceedances over a user-defined morphological threshold, deriving from repeated topographic surveys. This approach utilizes widely used empirical methods for wave run-up estimation and makes use of the most effective one after calibration. The performance of the EWS is assessed through seven monitored surge events of varying magnitude and hydrodynamic conditions, demonstrating strong agreement between projected TWL exceedances over predefined morphological thresholds, particularly under high-energy wave conditions. Minor discrepancies are noted during events with marginal TWL exceedances over short durations. Results underline the system's potential as a valuable tool for coastal hazard assessment and risk management, with future improvements focusing on appropriate updates of the beach morphology and the integration of suitable numerical techniques and machine learning algorithms.</p>
The Yunnan-Kweichow Plateau (YKP), a representative ecologically fragile zone, is subject to dual pressures from intensified climate change and anthropogenic activities. The specific mechanisms of how Net Ecosystem Productivity (NEP) responds to these changes remain unclear, whose relative contributions remain poorly quantified. This study conducts spatiotemporal quantification analysis of NEP dynamics and influencing factors in the YKP from 2001 to 2020, which integrated a linear regression, shifting center of gravity, Mann-Kendall trend test, partial correlation analysis, and random forest. The results showed an enhancement in NEP within the YKP (slope = 2.42 g C·m−2·yr−1, p < 0.05). Overall, climate change and anthropogenic activities contributed 1.86 g C·m−2·yr−1 and 0.76 g C·m−2·yr−1 to NEP variations, respectively. In terms of climate impact, temperature and precipitation are the main drivers affecting vegetation change, while radiation has the least influence. The importance of precipitation on NEP has been increasing by an upward trend, particularly in non-humid regions (slope = 0.31, p < 0.05) and grassland (slope = 0.45, p < 0.05). Besides, although the impact of climate change is dominant throughout the region, in areas affected by anthropogenic activities and climate change, the influence of anthropogenic activities is dominant and has a positive impact on the vegetation growth of YKP and NEP, especially in forest areas. The research elucidates the coupling mechanisms of how anthropogenic activities and climate change drive vegetation dynamics in the YKP region, providing key insights for boosting carbon sink capacity and promoting ecological sustainability.
Price volatility in grain markets, especially for maize, has substantial socio-economic impacts, particularly in low-income regions where food security remains a critical concern. Accurate forecasting of grain price volatility is therefore crucial in safeguarding the financial interests of commodity traders, as well as shielding consumers from detrimental effects of inflationary food prices. This study proposes a hybrid Bi-directional Long Short-Term Memory (BLSTM) model, integrated with generalised autoregressive conditional heteroscedasticity (GARCH)-type methods, to forecast white maize futures volatility in South Africa. By comparing the forecasting accuracy of the hybrid BLSTM model against several benchmarks, including standard LSTM and BLSTM models, our results demonstrate notable improvements in prediction accuracy, as shown through heteroscedasticity-adjusted performance metrics. The key contribution of this research is its enhancement of volatility forecasting by combining advanced machine learning with traditional econometric approaches, bridging a gap in predictive accuracy for commodity price dynamics. Additionally, this study supports the United Nations Sustainable Development Goals (SDGs), particularly Zero Hunger and Responsible Consumption and Production, by improving food price stability and risk management in agriculture. This approach exemplifies the evolving role of data science in financial analysis, offering market participants an effective tool to manage price risk and improve food security.Impact Statement This study introduces a novel hybrid forecasting model that integrates GARCH-type econometric techniques with Bi-directional Long Short-Term Memory (BLSTM) neural networks to predict the realised volatility of white maize futures. As white maize is a staple food, accurate volatility forecasting directly contributes to improved food security and price stability. The model significantly outperforms traditional approaches and standard deep learning models across multiple forecast horizons, offering a powerful risk management tool for farmers, traders, and policymakers. By enhancing the accuracy of agricultural price forecasts, this research supports the United Nations Sustainable Development Goals (SDGs), particularly Zero Hunger (SDG 2) and Responsible Consumption and Production (SDG 12), while also demonstrating the value of advanced data science methods in addressing real-world socio-economic challenges.