26 January 2024 in IEEE Access
This study addresses the critical global issue of food security, particularly under the influence of climate change on agricultural land suitability. The primary objective of our research is to predict the risks associated with land suitability degradation and changes in irrigation patterns, directly impacting food security. This research aligns with the United Nations’ sustainable development goals to reduce hunger and malnutrition. Central Eurasia, a region facing unique economic and social challenges, serves as the focal point of our investigation, providing a pertinent example for analyzing the effects of climate change on food security. In our approach, we employ interpretable machine learning techniques to analyze the impact of climate change on agricultural land suitability under different carbon emission scenarios. The developed model demonstrates strong performance, evidenced by an accuracy of 86% and a mean average precision of 72% in a multi-class land suitability classification task. Focusing on the most vulnerable regions in Eastern Europe and Northern Asia, our research provides crucial insights for policymakers. These insights are instrumental for strategic planning, including the allocation of critical resources like water and fertilizers, to prevent humanitarian crises. The results demonstrate that machine learning can be a powerful tool in predicting and managing the impacts of climate change on food security.
10 April 2025 in Geoenergy Science and Engineering
1 December 2000 in Plasma Physics Reports
Generation of electron Bernstein waves by the ordinary-extraordinary-Bernstein (O-X-B) mode conversion process has been successfully demonstrated on W7-AS. According to Kirchoff’s law, the inverse pro
19 November 2026 in ACM Transactions on Information Systems
Transformer-based sequential recommender systems are retrained frequently, yet their training is dominated by cross-entropy (CE) over item catalogs that can reach millions of items. Its logit tensor scales with catalog size × batch size × sequence length, forcing practitioners to trade these quantities against each other or use negative sampling. However, how to allocate memory among negatives, batch size, and sequence length remains an open question. We study this trade-off with nearly 1,000 SASRec training runs on six datasets and find that no single memory-related hyperparameter should be maximized in isolation. Scaling both the number of negatives and batch size outperforms maximizing only one, while the choice between full CE and CE with negative sampling is dataset-dependent. We further introduce a memory-efficient and GPU-optimized replacement for CE with negative sampling that avoids materializing the logit tensor. It speeds up training by up to 3.5× and reduces memory by 4–24× relative to PyTorch CE with negative sampling. Reallocating memory saved by our replacement and Cut Cross-Entropy, which we adapt to sequential recommender systems, into larger batches, longer sequences, and more negatives improves NDCG@10 by up to 54.7% under the same memory budget. We release the Triton kernel and a public dataset of all training runs at https://github.com/On-Point-RND/MemoryEfficientSRS .